albert casullo, joshua c. thurow-the a priori in philosophy-oxford university press (2013)

The A Priori in Philosophy

This page intentionally left blank

The A Priori inPhilosophy

EDITED BY

Albert Casulloand Joshua C. Thurow

1

3Great Clarendon Street, Oxford, OX2 6DP,United Kingdom

Oxford University Press is a department of the University of Oxford.It furthers the University’s objective of excellence in research, scholarship,and education by publishing worldwide. Oxford is a registered trade mark ofOxford University Press in the UK and in certain other countries

# the several contributors 2013

The moral rights of the authors have been assertedImpression: 1

All rights reserved. No part of this publication may be reproduced, stored ina retrieval system, or transmitted, in any form or by any means, without theprior permission in writing of Oxford University Press, or as expressly permittedby law, by licence or under terms agreed with the appropriate reprographicsrights organization. Enquiries concerning reproduction outside the scope of theabove should be sent to the Rights Department, Oxford University Press, at theaddress above

You must not circulate this work in any other formand you must impose this same condition on any acquirer

British Library Cataloguing in Publication DataData available

ISBN 978–0–19–969533–1

Printed in Great Britain byCPI Group (UK) Ltd, Croydon, CR0 4YY

For my grandchildren (A. C.)For my family—Marchell, Hannah, Kayleen, and Lola ( J. C. T.)

Table of Contents

List of Contributors ixAcknowledgements x

IntroductionJoshua C. Thurow and Albert Casullo 1

I: Intuitions, Experimental Philosophy,and the A Priori

1. Philosophical Naturalism and Intuitional MethodologyAlvin I. Goldman 11

2. Experimental Philosophy and ApriorityJonathan Jenkins Ichikawa 45

3. The Implicit Conception and Intuition Theory of theA Priori, with Implications for Experimental PhilosophyJoshua C. Thurow 67

4. The Prospects for an Experimentalist Rationalism, or Why It’s Ok if theA Priori Is Only 99.44 Percent Empirically PureJonathan M. Weinberg 92

II: The Nature and Scope of the A Priori

5. On the Armchair Justification of Conceptually GroundedNecessary TruthsDavid Henderson and Terry Horgan 111

6. Concepts, Teleology, and Rational RevisionChristopher S. Hill 134

7. A Priori Testimony RevisitedAnna-Sara Malmgren 158

8. Intuitions and Foundations: The Relevance of Moore and WittgensteinErnest Sosa 186

III: Skepticism and the A Priori

9. Skepticism, Reason, and ReidianismJoel Pust 205

10. A Priori BootstrappingRalph Wedgwood 226

IV: Challenges to the A Priori

11. Articulating the A Priori–A Posteriori DistinctionAlbert Casullo 249

12. Naturalistic Challenges to the A PrioriC. S. I. Jenkins 274

13. How Deep is the Distinction between A Priori andA Posteriori Knowledge?Timothy Williamson 291

Index 313

viii TABLE OF CONTENTS

List of Contributors

ALBERT CASULLO—Professor of Philosophy, University of Nebraska-Lincoln

ALVIN I. GOLDMAN—Board of Governors Professor of Philosophy and Cognitive Science,Rutgers University

DAVID HENDERSON—Robert R. Chambers Distinguished Professor of Philosophy and theMoral Sciences, University of Nebraska-Lincoln

CHRISTOPHER S. HILL—Professor of Philosophy, Brown University

TERRY HORGAN—Professor of Philosophy, University of Arizona

JONATHAN JENKINS ICHIKAWA—Assistant Professor of Philosophy, University of BritishColumbia

C. S. I. Jenkins—Canada Research Chair and Associate Professor of Philosophy, University ofBritish Columbia, and Chair in Theoretical Philosophy at the Northern Institute of Philosophy,University of Aberdeen

ANNA-SARA MALMGREN—Assistant Professor of Philosophy, Stanford University

JOEL PUST—Associate Professor of Philosophy, University of Delaware

ERNEST SOSA—Board of Governors Professor of Philosophy, Rutgers University

JOSHUA C. THUROW—Assistant Professor of Philosophy, University of Texas at San Antonio

RALPH WEDGWOOD—Professor of Philosophy, University of Southern California

JONATHAN M. WEINBERG—Associate Professor of Philosophy, University of Arizona

TIMOTHY WILLIAMSON—Wykeham Professor of Logic, University of Oxford

Acknowledgements

We would like to thank the original publishers for permission to reprint all or part ofthe following essays:

“Philosophical Naturalism and Intuitional Methodology,” Proceedings and Addressesof the American Philosophical Association 84 (2010): 115–50.

“Articulating the A Priori–A Posteriori Distinction,” in Albert Casullo, Essays onA Priori Knowledge and Justification (New York: Oxford University Press, 2012).

Introduction

Joshua C. Thurow and Albert Casullo

The a priori has long been a central topic of philosophical discussion. Much has beenpublished on the topic. The last decade, however, has witnessed interesting develop-ments in philosophy that have a significant bearing on the nature and existence of the apriori, as well as on its role in philosophy. Precisely how these developments bear onthe a priori and its role in philosophy has just begun to be explored. This collection ofessays addresses these developments with the aim of enhancing philosophical under-standing of the a priori and its role in philosophy, and stimulating further work on theissues.There are four recent developments in philosophy that bear directly on the a priori.

First, there has been a surge of work exploring the nature of intuitions. Intuitions areimportant to philosophy because philosophers often appeal to intuitions as evidencein constructing or evaluating philosophical theories and arguments. Much work hasbeen done explaining what these intuitions are like and the precise role they playin philosophical methodology. In addition, experimental philosophy—a very recentdevelopment—has advocated doing empirical studies to gather more general dataregarding philosophical intuitions and has used that data to both do philosophy aswell as criticize the philosophical use of intuitions. This work raises a host of interestingquestions about the a priori. Is the philosophical use of intuitions a source of prioriknowledge? Do the results of experimental philosophy challenge the a priori status or,more generally, the evidential status of intuitions?Second, there has been renewed interest in understanding the nature and scope of a

priori knowledge. One persistent challenge facing proponents of the a priori is toprovide an account of the source of such knowledge. Proponents have exploited twostrategies in response. Some appeal to rational intuition; others appeal to features ofconcept possession. Whether either of these approaches can be suitably developed anddefended remains controversial. There has also been renewed interest in understandingmathematical and modal knowledge, which have in the past been taken to beparadigmatic examples of a priori knowledge. Some philosophers have attempted toprovide empiricist explanations for how we can have such knowledge. Proponents ofthe a priori, however, maintain that such accounts fail to capture important aspects of

such knowledge. The resulting controversy raises significant questions about the apriori. Does the a priori play an important role in such knowledge? If so, what aspectof such knowledge do empiricist accounts fail to capture? Can appeals to features ofconcept possession or rational intuition explain that aspect or are new approaches tothe a priori necessary?

Third, after a period of relative neglect, skepticism has once again emerged as a focusof work in epistemology. New skeptical problems have been raised (such as theproblem of disagreement), and new solutions to skepticism have been proposed anddeveloped (such as contextualism). In the past, the a priori was often regarded as crucialfor avoiding skepticism. Descartes attempted to evade skepticism by building a systemof beliefs on the grounds of what is clearly and distinctly perceived. Kant thought thatthe synthetic a priori was necessary for avoiding skepticism. More recently, LaurenceBonJour has argued that a priori justification is necessary for making justified inferencesfrom basic beliefs. This new work raises challenging questions. Is the a priori needed toavoid skepticism? If so, what distinctive role does it play in responding to skepticism?

The a priori, however, also faces skeptical challenges. Some, for example, maintainthat such knowledge is incompatible with a naturalistic perspective. Others challengethe coherence or significance of the concept of a priori knowledge. These newchallenges raise important questions. What are the requirements of a naturalisticperspective on knowledge and how do they bear on the a priori? Is there a significantrole for the a priori in contemporary epistemology?

The chapters in this volume aim to advance the discussion of the role of the a prioriin philosophy by addressing four sets of issues. The first is whether intuitions provideevidence for philosophical theories, whether that evidence is a priori, and whether theresults of experimental philosophy affect the evidential or a priori status of intuitions.The second is whether there are explanations of the a priori and what range ofpropositions can be justified and known a priori. The third is whether a priori justifiedbeliefs are needed in order to avoid some skeptical worries. The fourth is whethercertain recent challenges to the existence or significance of the a priori are successful.We offer a brief introduction to the issues and a summary of the chapters that follow.

1 Intuitions, Experimental Philosophy, and the A PrioriPhilosophers often appeal to intuitions as evidence in constructing or evaluatingphilosophical theories and arguments. Moreover, they typically maintain that theyare a priori justified in accepting or rejecting philosophical theories and arguments onthe basis of such appeals. They maintain that intuition is a source of non-experientialevidence and, as a consequence, that beliefs justified by intuition are justified a priori.

Experimental philosophers have carried out social-psychological experiments toinvestigate the intuitions of non-philosophers about thought experiments that oftenare quite similar, if not identical, to those that philosophers have employed in defenseof various theses. In a famous paper that inspired subsequent work, “Normativity and

2 JOSHUA C. THUROW AND ALBERT CASULLO

Epistemic Intuitions,” Jonathan Weinberg, Shaun Nichols, and Stephen Stich pre-sented evidence that students of Western descent and students of East Asian descenthave significantly different intuitions about a standard Gettier case. Westerners largelythink the protagonist does not know, whereas East Asians are somewhat inclined, as awhole, to think the protagonist does know. Subsequent work has suggested thatpeople’s intuitions are affected by socio-economic status and the order in whichthought experiments are presented. None of these factors—one’s cultural heritage,one’s socio-economic status, and order of cases—seems at all relevant to the truth ofbeliefs that intuitions are used to support. Many philosophers have argued that this dataraises serious questions about the reliability of intuitions and that the philosophicalpractice of using intuitions as evidence needs to be either jettisoned completely orsubstantially overhauled, perhaps by requiring empirical checks on intuition, or byrefusing to trust some classes of intuitions.The questions raised by experimental philosophy regarding the role of intuitions in

philosophical theorizing provide the context for the first set of issues addressed by thisvolume: whether intuitions provide evidence for philosophical propositions, whetherthat evidence is a priori, and whether the results of experimental philosophy affect theevidential and a priori status of intuitions. In the first chapter, Alvin Goldman distin-guishes between first-order and second-order questions about the evidential status ofintuitions. Two central first-order questions are whether intuitions are evidence at alland, if so, whether they are a priori evidence. Two central second-order questions arewhether there is evidence that intuitions provide evidence and whether the mostappropriate type of such evidence is a priori or empirical. Goldman maintains thatthe results of experimental philosophy provide prima facie evidence against the first-order evidential status of intuitions. His primary goal, however, is to lay out theadditional investigation that must be done before we can draw a final conclusionabout their evidential status. Here he argues that psychological investigations that gobeyond the survey methods of experimental philosophy are necessary. Althoughexpressing some misgivings about the a priori/a posteriori distinction, Goldman iswilling to allow that intuitional judgments have a partly a priori status. So, Goldmandefends a conciliatory position about the relationship between empirical investigationand the a priori status of intuitions: there is (partly) a priori first order evidence fromintuitions, but experimentation is required at the second order level to verify thatintuitions provide evidence.Jonathan Ichikawa, in the second chapter, denies that the evidence from experi-

mental philosophy undermines the a priori status of philosophical judgments. Hemaintains that even if the negative experimental philosophers are correct in theircontention that we have to subject philosophical judgment to experimental tests inorder to maintain confidence in our philosophical judgments, the kinds of experi-mental work that we would have to perform would not undermine the a priori statusof philosophical judgment. Why? Because experience can enable one to have an apriori justified belief without warranting the belief. One way this could happen is if one

INTRODUCTION 3

has reason to think that one has a rational shortcoming in coming to judge p a priori;such a reason defeats one’s a priori justification for that belief. However, if one acquiresnew experiential evidence (E) that one doesn’t have the shortcoming in question, thedefeater is defeated, and one is now once again rationally able to base belief on one’soriginal propositional a priori justification for the belief. Experience E enables an apriori justified belief without warranting it. Ichikawa thinks that an experimentalvindication of intuition would play just such an enabling role for a priori justifiedbeliefs.

In chapter three, Joshua Thurow defends a kind of conciliatory position between thetraditional view that many philosophical beliefs are a priori justified, and experimentalcritiques of this view. However, Thurow reaches this position by developing anddefending a theory of how a priori judgments can be reliable, and then investigatingwhat this theory would imply about the viability of experimental philosophy. Thurow’s“Implicit Conceptions and Intuitions” (ICI) theory explains how there can be a reliableconnection between intuitions and the implicit conceptions that determine the contentof concepts, and a reliable connection between those implicit conceptions and thetruth. This theory, which employs deference to one’s community for the content-determining features of a concept, implies that experimental philosophy can in principleepistemically prune away unreliable intuitions, and also assist in better understanding aconcept. However, according to the theory, intuitions are generally reliable, and so willoften serve to justify philosophical claims.

In chapter four, Jonathan Weinberg investigates whether a range of empiricallymodified traditional philosophical methods would count as a priori. He argues thatsometimes experimental results simply modify philosophical practice, separate the goodintuitions from the bad, or (as Ichikawa also observes) defeat empirical defeaters for apriori justification. In these cases, one’s justification using the modified philosophicalmethods remains a priori. However, there are cases where experimental results play adirect, and indispensable (in the circumstances), justifying role for a belief—such aswhen experimental surveys of intuitions reveal a certain pattern, which we then take toreflect some truth about the concept in question. Here, the collection of intuitions, andthe fact that they are gathered in a certain way to ensure that they are representative,constitutes empirical evidence. Weinberg argues that beliefs formed on the basis of suchevidence can nevertheless rightly be considered a priori because this way of formingbeliefs can get us strong modal conclusions, and one of our positive conceptions of the apriori is that it is a process that justifies strong modal conclusions. So, the a priori, in thissense, doesn’t have to be entirely purified of empirically warranting elements.

2 The Nature and Scope of the A PrioriMany philosophers maintain that a wide range of propositions are knowable a priori.Traditional examples include mathematical, modal, logical, conceptual, and moralpropositions, as well as epistemic principles. Others dispute the a priori status of these


propositions. One question often pressed by the doubters is this: how could we possiblyknow these things a priori? They want an explanation of how a priori knowledgeof these propositions works. It would be desirable to have a plausible explanationnot only to address the doubters, but also to deepen our understanding of the a priori.The second part of this volume is thus devoted to the issue of whether there areexplanations of the a priori and what range of propositions can be justified and knowna priori.In chapter five, David Henderson and Terry Horgan develop an account of armchair

justification of conceptually grounded necessary truths. These truths are conceptuallygrounded in the sense that they are necessary solely by virtue of their constituentconcepts; in short, these truths are analytic. Henderson and Horgan argue that suchtruths can be justified abductively as the best explanation of a variety of relevantintuitive judgments, facts about conflicting judgments and judgment-tendencies,background scientific knowledge, and other information, some of which is empirical.They categorize beliefs justified in this way as “low-grade a priori” because empiricalelements are a crucial part of the justification. They illustrate this abductive method byshowing how the truth of “water is necessarily H2O” can be explained and justified, inpart, by more general conceptually necessary truths. Henderson and Horgan’s conceptof the low-grade a priori seems to correspond with Weinberg’s notion in chapter fourthat some beliefs may be justified a priori, but not purely a priori.

In chapter six, Christopher Hill takes up the task of defending conceptual rolesemantics against a formidable challenge: how to distinguish between beliefs andinferences that are constitutive of a concept and beliefs and inferences that are not.Hill develops and defends what he calls ‘teleological conceptual role semantics’:concepts are individuated by the most fundamental cognitive interests that theyserve, and thus also by the beliefs and inferences that enable them to serve thoseinterests. Hill applies his theory to logical concepts, such as ‘or’, and abbreviativeconcepts, which are concepts that abbreviate more complex conceptual structures.An example of such a concept is ‘bachelor’, which abbreviates ‘unmarried adult male.’Concepts such as these have non-empirical functions, and as a result, Hill argues, therules of use that such concepts must have so that they fulfill their functions will entaildefinitions or laws of logic that will be empirically unrevisable and a priori knowable.Hill also carefully responds to Quine’s arguments against analyticity, which wouldundermine his teleological conceptual role semantics if sound.Traditionally, philosophers have taken knowledge by testimony to be a clear case of

a posteriori knowledge. Tyler Burge has challenged tradition by arguing that testimo-nial knowledge can be, and sometimes is, a priori. On his view, humans can have apriori warrants to rely on their understanding and to rely on the rationality of theirsource. These warrants can generate all-things-considered a priori warrant in the rightcircumstances. If the source of testimony that p knows that p a priori, then one can alsoknow p a priori by relying on the source’s testimony that p. If Burge is correct, thescope of the a priori expands considerably. Expanding on her earlier work, Anna-Sara

INTRODUCTION 5

Malmgren critiques Burge’s argument in chapter seven. She responds to a recentdefense of Burge and offers a new way of defending Burge’s view using what shecalls the ‘Fast Track Model’ of warrant preservation. However, she ultimately finds thisnew defense unsatisfactory.

In chapter eight, Ernest Sosa confronts the classic problem of how to explain howwe have common sense knowledge of a wide range of facts about the external world.Sosa rejects the views of G. E. Moore and Ludwig Wittgenstein on this issue, whilerecognizing that each possesses important insights. Building off of those insights, Sosaargues that in order to be justified in their common sense beliefs, humans musthave justified belief in, or commitment to, various general claims. Furthermore,one’s justification for these general claims cannot derive from perception, memory,testimony, or introspection because these general claims are used in conjunction withthose sources to deliver common sense beliefs. These general beliefs are justified by arationally competent intuition, where the competence is an ability to grasp the truth ofa proposition simply through understanding it properly. Sosa concludes by formulatingaccounts of human rationally competent intuition and cultural rationally competentintuition.

3 Skepticism and the A PrioriSome philosophers think that a priori knowledge is necessary for avoiding variouschallenging kinds of skepticism such as global, external world, and inferential skepti-cism. As mentioned earlier, Descartes, Kant, and BonJour provide prominentexamples. Part III of the volume considers the role of the a priori in responding toskepticism. Joel Pust and RalphWedgwood both argue—in quite different ways—thata priori justification and knowledge is necessary for avoiding skepticism.

In chapter nine, Joel Pust takes up the standard for a cause that is widely thoughtdead: the Cartesian project of justifying all of our beliefs on the basis of introspectionand a priori intuition. In particular, Pust rebuts the Reidian objection that the Cartesianproject arbitrarily restricts justification to two sources: introspection and a prioriintuition. Pust argues that this restriction is not arbitrary because a priori intuition ofsufficient strength is necessarily reliable, whereas sources like perception and memoryare possibly systematically unreliable. As a result, a priori intuitions of a certain strengthare not vulnerable to the same kinds of skeptical worries as perception and memory. Hethen turns the Reidian objection on its head, and argues that the position of theReidian who fails to acknowledge that the Cartesian perspective does not lead toskepticism is arbitrary. Pust argues that this suggests that the Cartesian perspective isunavoidable for a properly justified epistemology. If Pust is correct, then not onlyphilosophical investigation, but the whole of knowledge, is held aloft by the a priori.

Ralph Wedgwood confronts a classic skeptical argument in chapter ten, and arguesthat skepticism can be avoided only if we are a priori justified in believing that we arenot in a skeptical scenario. The skeptical argument is this: we must be justified in


believing that we are not in a skeptical scenario in order to have justified beliefs aboutthe world. Only experience could justify such a belief, but sadly in fact it cannot.Wedgwood argues that the only way out of this argument is by accepting that we are apriori justified in believing that we are not in a skeptical scenario. He notes, however,that his arguments assume that it is rational to take experience at face value. Therationality of this belief-forming process is taken for granted, and so cannot beexplained in the same way that he explains why we are justified in believing that weare not in a skeptical scenario. Wedgwood thinks, then, that his solution to the classicskeptical argument reveals another more fundamental problem: how to explain therationality of certain belief-forming processes.

4 Challenges to the A PrioriA number of challenges have been raised both to the existence of a priori knowledgeand to its importance. One of these challenges was mentioned earlier: the challenge ofexplaining how the a priori works, particularly how a priori beliefs can be reliablyformed. A second challenge arises from naturalism. Many contemporary philosophersclassify themselves as naturalists (although defining precisely what that means is notori-ously tricky), and some worry about whether a priori justification fits into a naturalisticworldview. Thirdly, the distinction between a priori and a posteriori knowledge hasbeen challenged. Some argue that there is no clear distinction to be drawn; othersmaintain that any such distinction is epistemologically arbitrary and unilluminating.The chapters in Part IV discuss the two latter challenges to the a priori.Albert Casullo opens the part by surveying a number of recent challenges to the

coherence or significance of the a priori–a posteriori distinction offered by PhillipKitcher, John Hawthorne, Carrie Jenkins, and Timothy Williamson. He argues thatthese challenges fall short of their goal because they fail to correctly articulate thedistinction that they challenge. Casullo identifies three factors that obscure the conceptof a priori knowledge and maintains that the failure to take these factors into accountleads to three common errors in attempts to articulate that concept. He goes on tocontend that the recent challenges turn on these errors or variants of them.Can the a priori fit into a naturalistic approach to the world? Many philosophers,

including Michael Devitt, Penelope Maddy, and David Papineau, have thought thatthe answer is clearly ‘no,’ or ‘not in any significant way.’ In chapter twelve, CarrieJenkins responds to several of their arguments. She distinguishes several varieties ofnaturalism, drawing special attention to epistemological naturalisms and metaphiloso-phical naturalisms. She argues that each comes in several different flavors, and that somesimply deny the existence of the a priori, others are plainly consistent with the a priori,and yet others may have negative implications towards the a priori. She focuses herattention on three arguments based on the latter sort: a Quinean argument, andarguments of Maddy and Papineau. She maintains that each argument fails to show aconflict between naturalism and the a priori.

INTRODUCTION 7

In the final chapter, Timothy Williamson challenges the a priori–a posterioridistinction. He develops an earlier argument, which is criticized by Casullo in chaptereleven, and contends that, although there is a distinction between a priori and aposteriori knowledge (and justification), this distinction is superficial and of littletheoretical interest. This is so, says Williamson, because often there is little substantialepistemic difference between cases of a priori knowledge and cases of a posterioriknowledge. Both cases often involve experiences that go beyond simply enablingthe subject to entertain the proposition in question, but that fall short of providingevidence for the proposition. Consequently, argues Williamson, the a priori–a poster-iori distinction doesn’t track a theoretically valuable or insightful feature of knowledge.We should develop more finely tuned vocabulary for describing the epistemicallysignificant features of the kinds of examples that he describes.

The chapters in this volume by no means present a comprehensive picture ofthe role of the a priori in philosophy. Although this volume contains lines of argumentthat converge on certain conclusions—or that complement each other—there is stilldissent between many of our authors, and of course between our authors and otherphilosophers (this is philosophy, after all). In addition, many relevant issues remain tobe addressed and discussed with greater care. We hope that the chapters in this volumewill catalyze further work on the role of the a priori in philosophy.


I

Intuitions, ExperimentalPhilosophy, and the A Priori

1

Philosophical Naturalism andIntuitional Methodology*

Alvin I. Goldman

1 Intuitions in Philosophy: What’s the Controversy?A debate is raging over philosophical methodology. It is a debate between philosophicaltraditionalists and science-oriented philosophical naturalists concerning the legitimacyof the widespread use of intuitions in philosophy. Not everyone finds the term‘intuition’ the best label for what philosophers rely upon in the relevant sector oftheir practice. Instead of “intuitions” some prefer to talk of intuitive judgments,thought experiments, or what have you. Nonetheless, “intuition” is themost commonlyused term in the territory, so I shall not abandon it, though other terms will be usedas well. Described fairly neutrally, the philosophical activity in question consists ofspecifying a hypothetical scenario and rendering an intuitive judgment about thecorrectness or incorrectness of classifying it under a stipulated heading. Does a givenpredicate ‘F’ apply to an event, an individual, a pair of objects, etc., in the scenario?Putting the question less linguistically, is a certain property or relation exemplified in thescenario? The central question in the debate is whether, when intuitions or intuitivejudgments are formed about such cases, they provide good evidence for some typeof philosophical conclusion. For instance, are they good evidence for a conclusion ofthe form “Case C is/is not an instance of F”? Philosophers’ ultimate interest is not incases per se. Cases are examined as a means to determining the content or compositionof the referent of ‘F’ (where the referent may be a property, a kind, a meaning, or aconcept). But this further determination is not in the foreground—though it definitelycannot be ignored.That case intuitions sometimes have a powerful impact on philosophical thinking is

evident from historically salient cases. Gettier’s famous paper of 1963 is a primeexample. It was concurring intuitive judgments about his counterexamples to thejustified true belief analysis of knowledge that propelled major changes in our theoryof knowledge, or the analysis of knowledge. Today the JTB analysis is a dead letter. So

the impact of intuitive judgments in philosophy cannot be underestimated. Whetherphilosophers are right to be guided by such judgments remains to be settled.

Let us not worry excessively about how much of philosophy is done in this mold, orhow important a facet of philosophy it constitutes. Clearly, philosophical work is notexhausted by reflection on examples and their classification. Philosophy is also in thebusiness of theoretical construction. Nonetheless, a fair bit of analytic philosophy ispreoccupied with generating and examining intuitions about cases. Is this practicelegitimate? What is it supposed to tell us, and is it a reliable method for generatingevidence about some sought-after philosophical conclusions? If so, what are theseconclusions, and does themethod produce genuinely good evidence (at least in favorablecases)? How is the practice to be understood, and is it methodologically defensible?

Consulting intuitions about imaginary cases is a paradigm of “armchair” philosophy.Whatever else is meant by an “armchair” method, it is certainly intended to contrastwith the method of empirical science, which uses “real” experimentation and obser-vation rather than reflection on hypothetical examples. Philosophical naturalists tend todistrust armchair methods. They embrace empirical science as the paradigm of soundmethodology. Philosophy should not be dragged into epistemological disrepute byutilizing suspect faculties or procedures like intuition. Traditionalists reply that intu-ition is a perfectly sound faculty; it is just an a priori faculty rather than an empiricalone. Rationalists are happy to defend intuition, which they see as closely allied to, ifnot identical with, the faculty of reason. Hence the sharp split, at least in initial instincts,between traditionalists and philosophical naturalists.

The debate has intensified with the advent of experimental philosophy (X-phi). Pro-ponents of the so-called “negative” branch of X-phi, led by Stephen Stich, claim that theirexperimental studies raise skeptical doubts about the epistemic soundness or robustness ofphilosophical intuitions. They challenge the evidential status that the bulk of the professionroutinely accords to intuitions (Weinberg, Nichols, and Stich, 2001; Machery et al., 2004;Alexander and Weinberg, 2007; Nichols and Knobe, 2007; Swain, Alexander, andWeinberg, 2008; Alexander, Mallon, and Weinberg, forthcoming).1 Although experi-mental philosophers do not speak for all philosophical naturalists, they are a subclass ofphilosophical naturalists who have raised objections to the epistemic credentials of intu-itions. Negative experimental philosophy will therefore play a prominent, though by nomeans exclusive, role in the present discussion. I shall not attend closely to the details oftheir findings to date, nor offer an assessment of the scientific quality of their experimentalwork (which is sometimes criticized). I am more interested in the ways that experimentalwork can in principle shed light on the epistemological status of philosophical intuitions.

2 Two Types of Evidential Questions: First-order andSecond-order Questions

The debate over intuitional methodology is customarily formulated in terms of“evidence” (or evidential quality), a standard term in the epistemological vocabulary,

12 ALVIN I . GOLDMAN

but one less commonly at center stage than “knowledge,” “justification,” or “rationality.”Moreover, there is much less agreement among epistemologists as to what qualifies asevidence or positive evidential status. I shall therefore advance a proposal that suits presentpurposes.Before doing this, however, I want to introduce a distinction that will play a crucial

organizing role here. This is the distinction between first-order and second-order questionsabout evidence, in particular, about the evidential status of intuitions or intuitivejudgments. As background, I assume that intuitions are occurrent mental states, eitherintuitive judgments (a species of doxastic states) or non-doxastic states such as intellec-tual “seemings” or “attractions,” which tend to generate intuitive judgments. In eithercase, they are supposed to be spontaneous or non-inferential states, not states producedby conscious inference.The following are some sample first-order questions about the evidential status of

intuitions: Are intuitions, or intuitive judgments, evidence at all? What is theirevidential quality or strength? For what propositions or hypotheses are they goodevidence? For whom are they good evidence (their subjects only, or other people aswell)? What kind of evidence do intuitions have or confer: empirical evidence or apriori evidence?The following are some sample second-order questions about the evidential status of

intuitions: Is there evidence (good evidence?) for the first-order evidential status ofintuitions? If we don’t already have good enough evidence for the first-order evidentialstatus of intuitions, how should we go about gathering evidence about the first-orderstatus of intuitions? What are the most appropriate or most helpful kinds of second-order evidence: empirical or a priori ?Looking ahead, I shall advance the view (qualified in various ways) that the two

orders of evidential status for intuitions are of different types. Specifically, intuitions orintuitive judgments may have first-order evidential status of a substantially a priori kind,while their second-order evidential status—the evidence for their evidential status—ismainly of the empirical kind. This bifurcated treatment is only possible, of course, inlight of the distinction between first- and second-order questions about evidence.However, I do not commit myself to a firm endorsement of the bifurcated view. It is atricky matter, as we shall see in the final part. Whether the first-order evidential statusof intuitions is of the a priori kind depends partly on how one defines “a priori,” andalso on what cognitive science tells us about the cognitive processes underlyingclassification intuitions.

3 The Nature of Evidential StatesWhat is evidence? That is, what kinds of entities or states of affairs constitute pieces ofevidence (for something or other), and in virtue of what relation to an appropriaterelatum do they so qualify? Kelly (2008) distinguishes between several senses of“evidence,” and I shall focus on one of the senses he pinpoints. This is the reliable

PHILOSOPHICAL NATURALISM AND INTUITIONAL METHODOLOGY 13

indicator sense of evidence, in which X is evidence for Y if and only if X is a (fairly)reliable sign, or indicator, of (the truth or existence of) Y. When using the reliable-indicator sense of “evidence,” I suggest that we take states of affairs, or facts, as the entitiesthat constitute evidence. For example, there being 743 rings on the trunk of a giventree is (good) evidence for the proposition that the tree is 743 years old. That a columnof mercury in a given thermometer is at the 70-degree mark is (good) evidence that theambient temperature is 70 degrees Fahrenheit. That a clock reads “4:45” may be(good) evidence that the current time at the clock’s location is 4:45. In all of these casesthe specified state of affairs constitutes evidence for the specified relatum because theindicated state of affairs is a rather reliable indicator of the truth of the propositionspecified.2

Can a mental state qualify as good evidence under the proposed sense of evidence?Certainly, at least if we slightly tweak the proposal to read that somebody’s being in acertain mental state is the item of evidence. For some types of contentful mental states, itis very plausible that being in one of those states reliably indicates the truth of itscontent. Seeming to see an apple on a nearby table may be a reliable indicator of therebeing an apple on a nearby table. Seeming to remember eating granola for breakfastmay be a reliable indicator that one did eat granola for breakfast.

Not all mental states with contents, however, are reliable indicators of the truth oftheir contents. Wishing that p, imagining that p, fearing that p, and so forth are unlikelyto be reliably accompanied by the truth of their contents. However, intuiting thatp may be one of the mental states that is a reliable indicator of the truth of its contents,at least for a suitably delimited class of intuitings. If so, such intuitings would qualify asevidential states in the reliable-indicator sense.

Some philosophers argue that this slant on the evidential status of thought experi-ments is in danger of “psychologizing” the evidence. Williamson (2007) warns againstthis approach and points to a number of hazards. The evidence extracted fromphilosophical thought experiments should not be construed as the mental statesphilosophers undergo, i.e., their intuitings. Rather, the evidence consists in the contentsof those intuitings, i.e., the (objective) facts or states of affairs intuited.3 When aphilosopher judges that a character in a Gettier case fails to know proposition p, theitem of evidence isn’t the philosopher’s intuitive judgment; it is the (truth of) theproposition so judged, namely, that the character doesn’t know p. Williamson con-tends that the psychological view (A) misdescribes philosophical practice, (B) rests on afalse principle about evidence that he calls “evidence neutrality,” and (C) opens up anepistemologically unfortunate gap between thought-experimental evidence and whatit is supposed to be evidence for.

There is no room to respond to these worries in detail, but none of them, in myopinion, provides a well-founded reason to reject a psychological approach to intu-itional evidence.4 With respect to the threatened “gap,” in particular, the epistemo-logical literature contains a diverse array of theories—of both internalist and externalistvarieties—for closing this gap. Moreover, it deserves emphasis that the choice of


“intuitings” as the basic kind of evidence underpinning philosophical practice doesnot exclude “intuiteds” as a derivative species of evidence that can also be invoked.Unquestionably, philosophers appeal to the singular facts they discern through thoughtexperiments as part of their evidence for or against various theories. But such non-psychological evidential facts derive their evidential status (assuming they have suchstatus) from psychological evidential facts, i.e., the occurrence of intuitions or intuitivejudgments about the hypothetical scenarios.

4 Indicator Reliability and Process ReliabilityOur chosen criterion of evidencehood is the reliable-indicator criterion. But reliableindicatorship is not unrelated to reliable processes. Specifically, processes that lead to astate often contribute to its evidential or non-evidential status. Consider our exampleof the clock that reads 4:45. Is this reading a reliable indicator of the true time (in thetime-zone where the clock is located)? This depends on prior events and processes thatled to the clock’s current reading. For example, when the clock was last set, was it setcorrectly relative to the time-zone it was in? If so, has it remained in the same time-zone until now or has it been moved to a different time-zone? Finally, how well doesthe clock’s time-keeping mechanism work? A defective mechanism, obviously, canmake a big difference to the clock’s current accuracy. In short, past events and processesthat generate the clock’s current reading bear on the question of whether the currentreading indicates the correct time (in the current time-zone). Hence, in weighing theevidential status of the clock’s position vis-a-vis the hypothesis that the current time is4:45, evidence about those past events and processes is highly germane. If distortingevents or processes are known to have transpired, they can constitute (second-order)evidence to the effect that the clock’s current reading is poor evidence—if evidence atall—of the correct time.For reasons such as this, items that are claimed to be evidence can be examined for

bona fide evidential status by investigating their history of generation (among otherthings). People can evaluate the first-order evidential status of a state or event byconsidering the processes that led to it. In such a scenario, causal processes provide (onespecies of) second-order evidence about the first-order evidential status of a given stateor event.Consider forensic evidence presented in a legal context. A standard specimen of

forensic evidence is courtroom testimony by a forensic scientist to the effect that apattern of fingerprints found at a crime scene constitutes a “match” with a defendant’sfingerprints. Is this kind of testimony sufficiently reliable and hence worthy of beingadmitted into court? To ask this question is already to ask a second-order questionabout the evidence. It is to pose the question of whether there is good evidence thatthe method of fingerprint matching, as customarily done by forensic scientists, gener-ates testimonial acts that reliably indicate whether fingerprints found at crime scenes arethose of the defendant.


Such second-order questions have recently been raised in a very critical way inAmerican courts of law. The so-called “ACE-V” method of fingerprint evidence haslong been the gold standard of forensic evidence. But several high-profile cases haveoccurred in the last number of years in which alleged “matches” generated by theF.B.I. crime laboratory—long touted as one of the best forensic laboratories—turnedout to be badly mistaken. This led academic specialists to dig deeper into this method.On closer inspection, what the community of fingerprint evidence specialists call“scientific” does not pass muster as very scientific (Mnookin, 2008). (Simply calling amethod “scientific” does not make it so.) Moreover, the method’s accuracy has neverbeen subjected to proper statistical tests. So we have no scientific second-orderevidence that this so-called evidence is good evidence.

The forensic science example illustrates the relevance of generating processes to thepotential evidential status of a generated outcome. The example also illustrates thepoint that being taken to be evidence does not make something genuine evidence, atleast not good evidence. If a state of affairs does not reliably indicate the sorts of facts itpurports to indicate, it should not be relied upon as evidence. The forensic scienceexample also illustrates the point that empirical tests can often be designed to probe thereliability of a process or method, in order to decide whether its outputs are reliableindicators of that for which they purport to be indicators. These tests would beexamples of second-order evidence. Finally, just as the legal system should surelydemand good second-order evidence about the reliability of courtroom forensictestimony, it is reasonable for philosophers to seek analogous tests to obtain second-order evidence about the evidential quality of philosophical methods.

5 The Negative Program of Experimental PhilosophyExactly what experimental evidence challenges the reliability of intuitions, specifically,singular classification intuitions? Singular classification intuitions are intuitive judg-ments that arise when a person is asked whether a certain example is an instance of aspecified property or relation.5 Properties and relations of philosophical interest includeknowledge, causation, reference, etc. Experimental philosophers of all stripes usesurvey methods to compare and contrast intuitions of groups of respondents. Theyimitate experimentalists in behavioral science by using controlled conditions of varioussorts, e.g., slightly differently worded questions. Two types of stances, however, aretaken toward the outcomes of these surveys (see Alexander, Mallon, and Weinberg,forthcoming). Members of the positive program of X-phi draw various kinds ofinferences from differences in responses, but they don’t fundamentally challenge theepistemological soundness of intuition-based methods. Members of the negativeprogram, by contrast, try to find evidence that raises doubts about the quality ofintuitional evidence. Exactly how this challenge is mounted remains to be clarified.I shall interpret their findings as a challenge to the reliable-indicator properties ofintuitions. But experimental philosophers themselves have not expressed the matter


precisely this way. So it will take a bit of work to tease out the basis of the challenge.Even if my interpretation does not capture their intent, I think it is the most effectiveway to give skeptical punch to their line of thought.The studies by negative experimental philosophers have generated a bevy of findings

that they use to challenge the epistemic standing of classification intuitions. Several ofthese studies showed variation in intuitions across (American) subjects with differentethnic origins. Weinberg et al. (2001) gave their subjects a Gettier-like examplefeaturing a protagonist named “Bob,” and asked them whether Bob “really knows” aspecified proposition or “only believes” it. A majority (74 percent) of subjects withWestern ethnic origins responded that Bob only believes it, while a majority of subjectswith East Asian (56 percent) and Indian (61 percent) ethnic origins responded that Bobknows. A different study by the same authors (Nichols et al., 2003) reported that thenumber of philosophy courses subjects had taken was a factor in influencing theirresponses. Subjects who had taken a number of philosophy courses were moresusceptible to skeptical arguments than those who had taken fewer. A study byMachery et al. (2004) concerned intuitions about reference. They found that subjectswith Western ethnic origins were more likely than those with East Asian ethnic originsto give causal-historical responses to queries describing Kripke’s (1980) thought experi-ment involving the name ‘Godel.’ In another experiment, subjects’ intuitions aboutTruetemp-like cases of putative knowledge were apparently influenced by the presen-tation of contrasting cases, in particular by the order of presentation of the variousexamples (Swain, Alexander, and Weinberg, 2008). Yet another finding in the litera-ture is that intuitions are sensitive to affective content, such as seeing disgusting stimuli likegreasy pieces of old pizza (Haidt, 2001; Nichols, 2004; Nichols and Knobe, 2007;Prinz, 2007). Subjects’ intuitions about the relationship between causal determinismand moral responsibility can depend on the presence or absence of affective content inthe narrative of the thought-experiment (Nichols and Knobe, 2007). Such dataindicate that intuitions about the relationship between causal determinism and moralresponsibility are “unstable.”Do findings of “instability” speak to the matter of intuitional unreliability?

According to Alexander and Weinberg, these data show that intuitions are “sensitiveto factors irrelevant to the content of the thought-experiments themselves” (2007: 61).In other words, people’s intuitions are influenced by “irrelevant” factors. Sensitivity toirrelevant factors, argue Alexander and Weinberg, “impugns the status of intuitions asevidentiary” (2007: 66). How, exactly, does sensitivity to irrelevant factors bear onevidential status? Return to our suggested reliable-indicator construal of evidentialstatus. To repeat an earlier example, a tree’s having N rings on its trunk is evidence forits being N years old because a tree’s being N years old is reliably correlated with itshaving N rings on its trunk. Now, if X is causally influenced by factors having nothingto do, correlation-wise, with the truth of Y, sensitivity to these factors will detract fromthe prospect that X is a reliable indicator of Y. For example, if things irrelevant to (i.e.,uncorrelated with) a tree’s age causally influence the growth of its rings, then the


number of rings on its trunk won’t reliably correlate with the tree’s age. In general,findings that items of type X are influenced by factors irrelevant to Y undercut thethesis that items of type X are reliable indicators of Y. Hence, such findings would posea challenge to the claim that items of type X are good evidence for facts of type Y.

The reliable-indicator construal of evidence also makes sense of the Alexander–Weinberg claim that order effects and affective-content effects challenge the evidentialstatus of intuitions. If intuitions about knowledge are influenced by irrelevant factorssuch as whether the subject has just heard a contrasting case, those intuitions cannot bereliable indicators of the truth of the matter. Whether there was a prior presentation ofanother case is not diagnostic of the truth in the target case.

What about the inter-subject variation that experimental philosophers emphasize?How does this challenge the evidential status of intuitions? If a group of subjectsperforms a given thought experiment, and 70 percent make one intuitive judgmentwhile the other 30 percent make a contrary judgment, they disagree with one anotherabout the case. Assuming there is one correct answer, one group or another must bewrong: either 30 percent or 70 percent. Thus, at least 30 percent are wrong. If thebreakdown is 50/50, then 50 percent must be wrong. So higher levels of variationimply higher (minimum) levels of unreliability. Under the reliable-indicatorship inter-pretation of evidence, variation poses a challenge to the evidential status of intuitivejudgments.

Swain, Alexander, and Weinberg (2008) provide ostensible confirmation that(putative) unreliability is the foundation of their challenge. They write:

We certainly do not take ourselves to have offered anything like a general proof of theunreliability of all intuitions . . . (2008: 153)

This concession that they lack any proof of unreliability carries the implicature thatunreliability worries are at the heart of their challenge.6

It is clear, then, how our interpretation of evidential status in terms of reliable-indicatorship makes sense of X-phi challenges to the evidential status of intuitions—although they themselves don’t advance this interpretation. Moreover, our interpretationcan proceed in terms of the framework we articulated earlier. Experimental philoso-phers should be understood to be presenting second-order evidence in support of theproposition that intuitions, or intuitive judgments, lack first-order evidential status. AndI think we should concede at least this much to their claims: the cited experimentalfindings at least provide prima facie evidence in support of the denial of first-orderevidential status. Of course, prima facie evidence is, by definition, defeasible evidence.So in admitting that there is some second-order evidence in favor of the non-evidential(or poor evidential) status of intuitions, we leave it open that this evidence canbe defeated, or overridden. How might it be overridden? One route to defeat isan empirical route. The aforementioned experimental studies provide a body ofempirical evidence that ostensibly conflicts with the claim that intuitions are reliableindicators—for example, reliable indicators of the truth of their contents. But new


empirical evidence might suggest a different conclusion. A second possible route todefeat is a non-empirical route. The X-phi arguments for intuitional unreliabilitycrucially depend on (vague and tacit) assumptions about what the intuitions, orintuitive judgments, might be evidence for. There is room for philosophical maneu-vering here. Maybe intuitions are highly reliable indicators for propositions thathaven’t yet been adequately considered. This is another possible way to sustain first-order evidential status for intuitions without disputing the existence of some second-order evidence pointing the other way.There is also room for different grades of skepticism directed at intuitive judgments.

A radical skepticism would claim that all intuitions are devoid of evidential value.A more qualified skepticism would only claim that some intuitions are devoid ofevidential value. There are various ways that different intuitions might merit differentevidential statuses. For example, intuitions experienced by some sort of experts mightattain strong evidential status whereas intuitions experienced by other people might failto attain such status. Or, the causal processes that generate some intuitions might becompatible with their outputs having high reliability whereas other causal processesmight be incompatible with high reliability. (This comports with our earlier discussionof a link between reliable-indicatorship status and processes of causal production.)Alexander and Weinberg (2007) endorse a rather strong skeptical stance, adoptingwhat they call a “restrictionist view” of intuitions’ evidential status. Similarly, Macheryet al. contend that “philosophers must radically revise their methodology” (2004: B9).More moderate views, however, would leave the door open to evidential respectabilityfor some subset of intuitions or intuitive judgments. In the remainder of this chapter,I shall try to pinpoint the kinds of additional work that should be done before clearconclusions can emerge.

6 A Framework for Studying Classification Judgmentsand Error Possibilities

Evidential respectability for intuitions, as for other possible sources of evidence,depends heavily on the choice of propositions for which the states or sources are takento be evidence. When a detective claims that a certain item—say, a glove—found ata crime scene is crucial evidence, he will have in mind some proposition or clusterof propositions for which it might be evidence. These might include “PersonX committed the crime,” or “Person X was at the scene of the crime,” etc. Similarly,when philosophers either claim or deny that intuitions are evidence, they usually meanthat they are evidence for some proposition(s) of philosophical interest. However, I donot think there is complete agreement as to which proposition or propositions theseare, exactly. This is open to debate. Part of the task of deciding whether classificationintuitions have evidential value for philosophy is the task of deciding which hypothesesor conclusions (of philosophical interest) they might be evidence for.


One might think that the answer is straightforward. Isn’t an intuition’s contentalways the proposition for which it is, or purports to be, evidence? Not necessarily;there are other possibilities, as we shall see. Even if an intuition’s propositional contentis initially the most likely candidate, it is not the only one. Even if we assume that thereliability of a token intuition is determined by the truth of its content, there is aproblem of using experimental data of the kind X-phi generates to assess the generalreliability of intuitions (specifically, classification intuitions). Under our interpretation,experimental studies of the survey kind shed light on intuitional reliability by deter-mining the amount of disagreement across subjects. But verbal disagreement may ormay not reflect intuitive disagreement about one and the same proposition, becausedifferent subjects’ intuitions may have different propositional contents (Nichols andUlatowski, 2007; Sosa, 2008). This could emerge from divergent interpretations of theexample they are asked to classify, of the target category they are asked to address, andso forth. Such complexities must be taken into account when trying to construct ageneral perspective on the kinds of evidence that might be relevant to determining thereliability of classification intuitions (and hence their evidential status).

I begin, then, with an analytical device: the “classification game” (cf. Ludwig, 2007).A classification game involves three players, the first two being the interrogator and therespondent. In the first move of a classification game the interrogator describes to therespondent an example or case, C, and a category, kind, property, predicate, orconcept, F, and asks whether C is a member or instance of F (or satisfies F). In otherwords, the respondent is invited to classify C—or some person or event embedded inC—as an F or a non-F. In the second move of the game, the respondent makes aspontaneous membership (or instantiation) determination: C belongs to F, or C doesnot belong to F. (Alternatively, he suspends judgment.) Finally, the respondentexpresses this determination via a verbal report or mark on a questionnaire. Thethird player in the game is the philosopher–scorekeeper, who is assumed to be omniscientconcerning what goes on in the heads of the interrogator and respondent, as well asother related matters. The scorekeeper’s job is to ponder the question of whether therespondent’s intuitive judgment—i.e., the mental state from which the verbal responseissues—is true or false, and whether his making this judgment qualifies as a piece ofevidence (for some hypothesis of interest). Assessing the judgment’s truth-value maywell require, of course, prior determination of its truth conditions, which is one of thephilosophical questions on the table. To decide if it is a legitimate piece of evidence,the scorekeeper must also consider what it might be evidence for. Perhaps it is evidencefor the proposition to which the respondent assents, e.g., that C is an instance of F. Butwhat proposition is that, exactly? Moreover, even if the respondent’s intuition turnsout to be poor evidence for that proposition, it might be good evidence for anotherproposition of interest to philosophy, as previously indicated.

I now introduce the technical term classifier to cover a wide range of things includingproperties, universals, kinds (natural or otherwise), linguistic meanings, and the contentsof individuals’ concepts. All these objects either have instances and counter-instances, or


are standards or criteria that fix, pick out, or determine sets of instances and counter-instances. A central problem for the scorekeeper is to decide which type of classifier touse in keeping a respondent’s (or group of respondents’) reliability score. Such scores arereadily influenced by the choice of classifier, as we shall soon see.I divide classifiers into two categories. One category is person-specific or group-specific

classifiers. A second category is free-floating classifiers. An example of the first category isthe content of a single person’s concept, a content the person associates with a predicate orterm of interest ‘F.’ Assume that every (English-speaking) person associates a particularcontent with the term ‘cup.’ Any such content is a satisfaction condition that some-thing must meet to qualify as a cup. Such a content supervenes (in part) on somecomplex mental state of the person, a state that, when activated, tends to gives rise tonew thoughts and to verbal behavior, including classification behavior. The term‘concept’ is used ambiguously to refer both to mental states of the requisite sort or totheir contents. In the present chapter only the latter sort of thing is viewed as aclassifier.7 Obviously, one and the same content might be realized by the mental statesof different individuals. Sometimes, however, we may want to refer to a particularperson’s concept (or conception) of a predicate ‘F’ (or F-ness), that is, to the content ofthat person’s concept, ignoring whether or not that content is shared by others. In sucha case we are interested in a person-specific classifier.In addition to person-specific classifiers, we can take an interest in community-specific

classifiers. For example, we might be interested in the meaning of ‘cup’ in English, thatis, across the entire community of English speakers. Or we might take an interest in itsmeaning in a more restricted community of English speakers. Obviously, some wordshave different meanings in different sub-communities. In British English, ‘boot’ refersto the rear part of a car used for storage; in America ‘boot’ has no suchmeaning. I assumethat meaning is a kind of content, a sub-propositional content. Thus, community-specific classifiers will also be contents. They presumably supervene, in some complexfashion, on language-related states and dispositions of the relevant community.In addition to person-specific and community-specific classifiers, there may also be

“free-floating” classifiers. These might include properties, universals, kinds, and soforth. These are not associated with, and presumably do not supervene on, anyparticular individual or group that represents them. Their nature or constitution doesnot essentially depend on their status as intentional objects of people’s mental repre-sentations.Classifiers in general are candidate targets of philosophical analysis or theorizing.

When we try to analyze, explicate, or give a theory of knowledge, justice, causation, orpersonal identity, we aim to figure out what the satisfaction conditions are for‘knowing that p,’ ‘being just,’ and so forth. It is generally assumed that we are ableto recognize instances of these classifiers. Our ultimate aim is to elucidate what thoseinstances share. This consists of making explicit the “constitution” or “composition” ofa selected classifier. Which type of classifier is appropriate? And for each candidate typeof classifier, how do we go about investigating its constitution or composition?

PHILOSOPHICAL NATURALISM AND INTUIT IONAL METHODOLOGY 21

According to time-honored practice, we consider or think up actual and possibleexamples and ask whether they are, intuitively, positive or negative instances of theclassifier. Which type of classifier makes best sense of this practice?

On any of these approaches to classifiers, there will presumably be genuine cases ofknowing, causing, being the same person over time, etc. “in the world.” Even thechoice of person-specific classifiers as targets of philosophical investigation would notimply that causal relations or relations of personal identity over time are merelysubjective rather than objective. Similarly, such a choice is fully compatible withthere being genuine knowers in the world. However, it does not follow that thechoice of classifiers is irrelevant to the viability of philosophical practice. To thecontrary, a choice among classifiers can make a notable difference to the epistemicviability of philosophy’s intuitional (or thought experimental) methodology, in par-ticular, to its vulnerability to skeptical challenges.8

How can a choice of classifier affect error rates? Suppose that a person-specific contentis our favored type of classifier, and that different people associate different contentswith a term of philosophical interest. When invited to decide whether “knows” appliesto a specified case, one respondent uses his concept of knowing, K1, while another usesher concept of knowing, K2. If the first makes a positive classification judgment and thesecond makes a negative one, their verbal responses suggest contradictory intuitions, soone of them must be in error. But if their judgments “answer” different questions, itdoes not follow that either is in error. Each might answer correctly in terms of his/herown concept, or conception, of knowing. Thus, choosing person-specific contents asthe classifier would lower the error rate. This would result in higher reliability scoresfor intuitions and possibly avert intuition skepticism.

But, it will be objected, the propositions to which respondents assent are surely notpropositions about the contents of their concepts. For example, if the posed query iswhether a certain institution (e.g., slavery) is just, a respondent’s answer presumablyaddresses justice itself, not his personal concept thereof. In saying that the specifiedpolicy is unjust, he means to classify it relative to a free-floating classifier, not relative toa person-specific or community-specific classifier.Let P (below) be the evidential fact that consists in respondent R having an intuition

with the content “Case C is an instance of justice.” P* is the proposition that case C isan instance of justice.

P: Respondent R intuits that case C is an instance of justice.P*: Case C is an instance of justice.

What might P be evidence for? One possibility, of course, is (the truth of) P*. Thisevidential relation between P and P* is the one people usually have in mind when theysay that intuitions are evidence. But attitudinal facts can be evidence for matters otherthan their own contents. Thus, P might equally be evidence for, say, P**:

P**: R’s concept of justice, J1, is such that C is an instance of (or is in the extensionof ) the content of J1.


An evidential link between P and P** has much to recommend it. Whatever R’sunderstanding of justice—whether it accords with the moral structure of the universeor not—it is eminently plausible that R’s intuition accords with (the content of) hisown concept of justice. In other words, R would plausibly apply his own concept inways that conform to that concept’s content, or satisfaction conditions. Thus, ifR assents to the proposition that case C is an instance of justice, isn’t this evidencethat C satisfies the conditions of his concept of justice, K1, just as P** asserts?

Most philosophers are unimpressed by this kind of evidential “power” of intuitivejudgments. They would prefer a tighter parallel between intuition and perception.With respect to perception, it is commonly held that propositions (facts) of the formQ are evidence for propositions of the form Q*:

Q: Person S seems to see that X.Q*: X.

Similarly, it is possible that proposition P is evidence for P*. But that is not theevidential connection offered above. The P/P* evidential link is under threat fromfindings in experimental philosophy. At the moment, then, I consider the alternativeevidential link between P and P**. Unlike the P/P* evidential link, the P/P** link isnot undermined by experimental evidence. Moreover, if the philosopher/score-keeper’s job is to be on the alert for what states of affairs intuitions might reliablyindicate, the P/P** link looks extremely promising.9

This sort of evidential scenario might motivate us to turn to person-specific contentsas the favored target of philosophical analysis (cf. Goldman and Pust, 1998).10 Alterna-tively, person-specific contents might be regarded as only a first target of philosophicalanalysis, on whose basis other targets might later be addressed (Goldman, 2007). I donot mean to preclude the latter scenario. However, most philosophers prefer sometarget of philosophical analysis that is not person-specific. So it is time to examineprospects for reliable evidence under the remaining types of candidate classifiers.

7 Problems for Evidential Status under Free-FloatingClassifiers

The next class of classifiers I shall consider as candidates for the principal targets ofphilosophical analysis is free-floating classifiers (FFCs). I turn straightaway to a firstworry, of ancient vintage. Under any approach to the proper type of classifier, respond-ents engaged in classification tasks will need to execute two prior tasks to be successful(i.e., accurate): (1) acquire a proper understanding of the case before them, and (2)acquire a proper grasp of the classifier in question. Since our focus is on the classifiers, setaside any potential problems with case understandings. The big question is how toacquire a grasp of FFCs. If such a grasp is not routinely attainable by ordinary people,reliable classification will be out of reach. Any defense of intuitional methodology by


appeal to FFCs, then, owes us a story about how such a grasp can be routinely attained.Moreover, it must be attained for all of the specific classifiers to which intuitionalmethodology is customarily applied: knowledge, causation, object composition, per-sonal identity, etc.

Divide the possible strategies for resolving this issue into two approaches: rationalistand empiricist approaches. A rationalist approach holds that people have a rationalfaculty of intellectual apprehension by which they grasp a priori the composition orconstitution of abstract objects like universals. One finds this in Plato’s treatment of thecontemplation of the Forms and in much subsequent writing, of course. But what isrational apprehension, and how does it enable us to reliably grasp truths about theconstitution of abstract objects? A majority of writing on this subject uses perceptualmetaphors to characterize intellectual apprehension. It is said to be a way of “seeing”that such-and-such a proposition is necessarily true. But talk of “seeing” is clearlymetaphorical, and does nothing to dispel the mystery. In contemporary science,perceptual processes like seeing and hearing are information-transmissional processes,which run from distal stimuli to mental experiences. Such processes are moderatelywell understood as sequences of causal links involving energy propagation, transduc-tion, and transformation. In the case of genuine perceptual modalities, it is understood,at least in general terms, how information can be reliably transmitted from externalobjects to internal cognitive states (Dretske, 1981). But it remains a mystery how suchtransmission could work in the case of intuition, where the alleged medium isunknown to science. This makes naturalists suspicious of whether there could be anyreliable process by which the properties of FFCs, i.e., their specific compositions, mightbe discerned.11

A second and less familiar puzzle about the rationalist solution is that it makesnonsense of customary philosophical practice. Customary philosophical practice,as we have emphasized, tries to get a handle on knowledge, causation, and so forthby working from singular cases in which knowledge, causation, etc. are allegedlyinvolved. We “sneak up” on knowledge itself by making inferences from instancesand counter-instances. Why do we proceed this way? Why approach the constitutionof the classifier itself by eliciting intuitions about classifier instances? The reason mustbe, in part, because we lack any direct access to the classifiers themselves (and theirconstitutions). But if so, what should we make of the foregoing rationalist story? If wedid have “direct” intellectual apprehension of the constitution of classifiers, why wouldwe resort to a cumbersome, round-about, back-door method of soliciting intuitionsabout cases? Why not just call on our intellectual faculty to grasp the classifiersthemselves directly, rather than plod through the tedious process of devising hypo-thetical examples, recording intuitive responses to them, and drawing highly fallibleexplanatory inferences to the composition of the classifiers?12

Turning to the empiricist approach, perhaps respondents are thought to gain accessto FFCs by empirical methods, including the confirmation of empirical theories. This iscertainly the view of Kornblith (2002). But Kornblith does not seek to defend standard


intuitional methodology. Our question is how FFCs (possibly including natural kinds)can be invoked as part of customary practice when the alleged access to them is viaempirical theorizing. First, it’s unclear what it would be like to have an empiricaltheory for many FFCs to which intuitional methodology could be applied. Whatwould an empirical theory of object-composition (in the mereological sense) consistin? Second, innumerable empirical theories advanced during the course of history(whether scientific or non-scientific) have been false. Many (perhaps most) scientific-ally confirmed and accepted theories later proved to be false. So why should philoso-phers place much trust in people’s theories of FFCs? Notice that philosophers routinelyappeal to folk intuitions when using customary philosophical practice. Why shouldthey assume that the folk have an accurate grasp of the nature of object-composition? Ifthey lack such a grasp, though, how will their classifications of hypothetical examplesget matters right? To repeat the critical premise, without accurate understandings ofboth hypothetical cases and the nature or constitution of a target classifier, people’sjudgments about whether the former instantiate the latter cannot be relied upon foraccuracy. Yet philosophers trust such judgments to be reliable. No plausible rationalefor such trust has been presented under the FFC approach.

8 Community-Specific ClassifiersGiven the bleak prospects facing the FFC approach, let us turn to the remainingoption: community-specific classifiers. In particular, consider linguistic meanings. Aspreviously suggested, linguistic meanings are presumably fixed in part by the contentsof person-specific contents that language users associate with various terms or predi-cates. These communal contents are not necessarily “averages” of person-specificcontents. Language communities often assign extra weight, or deference, to experts.Thus, meanings might be thought of as “normative” contents that arise from individualconcept contents. There is a problem here, however, because no individual user hasdirect access to this “social” entity (the communal meaning) in the way they haverelatively direct access to their own individual concepts. Won’t similar access problemsarise for this type of classifier as for FFCs?The problems are not as severe, however, as in the case of FFCs. Speakers acquire

their personal concepts of most linguistic terms by interaction with others who arerelatively authoritative speakers. Parents and caregivers are competent at identifyingmembers and non-members of various terms’ extensions, and learners receive feedbackon their own classification attempts. Although such procedures do not infalliblytransmit communal meanings, they are credible routes of access, far less obscure thanthose postulated for free-floating classifiers. More work in psycholinguistics and devel-opmental psychology is needed before the transmission of semantic competence tochildren is fully understood. But no comparable mysteries are encountered here aswhen we consider the prospects for FFCs. Therefore, I provisionally adopt this type ofclassifier as my favored type of classifier.


What, then, is the central type of hypothesis for which an intuition is evidence? Theprototypical form of such a hypothesis will be, “Case C meets the satisfaction condi-tions for F,” where F is the communal meaning of the queried predicate. This form ofhypothesis is also the content of the subject’s intuition; it is what the person intuits.I previously argued that intuitional states can be evidence for propositions other than(the truth of) their contents. But their propositional contents also remain goodcandidates to be hypotheses for which they are evidence. That is how we shall construetheir main evidential significance for present purposes. Thus, if classification intuitionssurvive challenges to their evidential status, what they will be evidence for (in the firstinstance) is their contents.

Suppose, then, that a reader of the Gettier article intuits, with respect its first example,that Smith does not know that the man who will get the job has ten coins in his pocket(= T). This intuition might be evidence for the hypothesis that Smith fails to meet thesatisfaction conditions of knowing with respect to T (according to the appropriatemeaning of ‘know’ in the contextually relevant language community). To arrive at thisintuitive judgment reliably, the intuiter presumably must make cognitive “contact”with a stored mental representation of the relevant sense of ‘know.’ He must (tacitly)activate that representation and compare its content with what he understands aboutT. In addition, his representation of the relevant sense of ‘know’ must correspond—closely enough—to a sense of the term ‘know’ in English. But none of these featsrequires any miracles. If he has been suitably exposed to English and has normallyoperative language-acquisition capacities, he should have managed to acquire a relevantrepresentation of ‘know.’ So it should not be a super-human task to be poised to havereliable intuitions about the proper classification of examples about knowing.

9 Processes of Classification Judgment: Preparatory andFinal

Given the choice of communal meanings as our provisional type of classifier, let us nowturn to a more detailed and wide-ranging exploration of the ways that empiricalscientific investigation can and should be used to help assess the reliability of classifica-tion intuitions, and hence their evidential status. We have already seen how experi-mental philosophy has contributed to this exploration. But the familiar paradigms ofexperimental philosophy provide only hints about the processes that produce classifica-tion judgments. Identifying such processes, as we saw in Section 4, can be extremelyrelevant evidence—second-order evidence—in assessing the evidential status of theirend products. The standard X-phi studies provide only hints because survey studies arenot designed to get at causal processes. Other types of psychological investigation,however, can and should fill this gap. Indeed, X-phi practitioners themselves sometimesemphasize the need for scientific studies outside the survey paradigm (Alexander,Mallon, and Weinberg, forthcoming).


Identifying the processes that influence classification judgments is a complex matterbecause multiple stages of processing are involved. Some processes are preparatory toclassification proper and others are constitutive of it. Nonetheless, all can influence thefinal output. Given the complexity of these matters, we should not, of course, expectanything like precise truth-ratios to emerge from our discussion. Our aim is just to testthe waters, to provide a preliminary sense of the sorts of processes that might exacerbateor moderate the threat to the evidential standing of intuitional methodology.Working within the communal meanings framework, consider the cognitive tasks

confronting a respondent in a classification game. She must first decode the questionposed by the interrogator. This consists in decoding the case description and the targetpredicate. If predicate ‘F’ (or the principal lexical constituent of ‘F’) has multiple senses,a respondent must deploy semantic and/or pragmatic processes to identify the appro-priate sense. There is room for respondent error here; and if error occurs at this stage, itmagnifies the likelihood of error in the eventual classificational output.Although errors can easily arise from insensitivity to the multivocality or polysemous

character of many philosophical terms, I shall not pursue this source of error. InsteadI shall consider a type of error that arises from the application of a certain polysemousterm of great philosophical interest, namely, ‘know.’ The conventional wisdom inepistemology is that ‘know’ (in its propositional construction) has just one sense, theone that epistemology has doggedly pursued since Plato. In contrast, I believe there areat least three senses of ‘know.’ The first is mentioned with some frequency, thoughmany epistemologists pooh-pooh it. This is the sense in which knowing somethingconsists in being completely confident of it. A second sense of ‘know’ is to believe somethingtruly. Having argued for this second sense in previous publications, here I just add (inthe attached footnote) a new argumentative wrinkle.13 The third sense of ‘know’ is themain quarry of epistemology, the justified-true-belief-plus sense of knowing (where ‘plus’is a placeholder for an anti-Gettierization condition). For present purposes, I focus onthe complete-confidence sense of knowing.In the wake of discussions of contextualism, recent epistemology has lavished

attention on ways that subtly different descriptions of one and the same case can promptdifferent responses from readers or hearers. Assorted explanations have been offeredof these differences, including psychological explanations. For example, Hawthorne(2004) appeals to the so-called “availability heuristic” (posited by psychologists) toexplain our unwillingness to impute knowledge when certain non-actualized possibil-ities of error are mentioned. Nagel (2010) offers an alternative explanation, which isinteresting because the process she invokes is known from psychological work to beerror-conducive. If this process shapes knowledge ascriptions and withholdings, it couldnicely illustrate a preparatory process that encourages inaccurate inputs to classification.Nagel first considers a “plain” story about John Doe in a furniture store, who is

described as seeing a red table and believing that it’s red. This plain story is contrastedwith an embellished version, which adds the information that a white table under redlighting would look just the same and Doe hasn’t checked the lighting. Hearing the


embellished story, says Nagel, is more likely to lead one to deny that Doe knows. Why?What transpires in a hearer’s head to prompt this difference? Nagel suggests thathearing the second story may lead one to misrepresent Doe’s mental condition byascribing to Doe the same information about an error-possibility that the hearer herselfhas. This pattern of ascription would conform to a “bias” that psychologists call “thecurse of knowledge.” In the present case, the bias would manifest itself in a hearer firstbeing herself focused on the possibility of deception through unusual illumination andthen “projecting” this focus onto Doe. Why would projecting this thought onto Doeprompt a denial of knowledge to Doe? On Nagel’s telling, once the hearer thinksabout Doe as cognizant of the deception possibility, she will view him as overly hasty(hence unjustified) in believing the table to be red. I offer a slightly different explan-ation. Once the hearer represents Doe as being cognizant of the deception possibility,she will also represent him as being less than fully confident that the table is actually red.Then, thinking of knowledge as requiring complete confidence, she would deny thatDoe knows. Plausibly, this denial of knowledge is a false classification judgment.Moreover, the process that leads to it is unreliable because it has an inherently error-generating feature, namely, attributing one’s own informational states to a target.

It is instructive to note that the process of projecting one’s own states onto others is alsocalled the simulation process of mindreading. Elsewhere I defend the thesis that simulationis a primary method of executing mindreading tasks, and a good chunk of my evidence isprecisely the documented error propensities that Nagel discusses (see Goldman, 2006:164–73). Interestingly, another theorist of philosophical methodology, Williamson(2007), takes simulation to be the method of making counterfactual judgments andhence the method of doing philosophical thought experiments. Williamson, however,seems unaware of the psychological literature on egocentric bias and its close associationwith simulation (at least in the case ofmindreading). This is important because it poses thethreat of error in precisely the domain he wishes to safeguard from error. He defends theepistemic quality of thought experiments on the ground that counterfactual judgmentscannot be broadly inaccurate. But the very method he advances as crucial to counter-factual judgments is an error-prone method—at least when applied to mindreading.

The error Nagel hypothetically pinpoints in a hearer’s process of classifying Doe asa non-knower does not occur in the classification process proper. It occurs in the“run-up” to classification, when the hearer constructs in imagination what Doe isthinking. This illustrates what I call a “preparatory process,” and it’s intended to showhow such processes can contribute to classification errors. In addition to the prospect oferror-generating properties in preparatory processes, can a classification process properalso breed error on its own? Might this provide additional examples of how cognitivescience can contribute second-order evidence that is relevant to the first-order eviden-tial status of classification judgments? The answer, I think, is “yes.” I offer twoillustrations of this idea.

A bit of background is needed about the psychological structures underlying conceptpossession and how they are deployed during categorization. “Categorization” is the


standard term in the psychological literature for classification proper. The psychologicalliterature is full of such theories (for overviews, see Margolis and Lawrence, 1999;Machery, 2009). I shall not review all such theories, but shall simply select a couple ofthem for illustrative purposes.The exemplar theory holds that concept possession consists of storing in memory a set

of previously encountered exemplars of the category. Possessing a concept of “dog,”for example, consists of storing in memory an assemblage of representations of dogexemplars. To categorize an object as a dog or non-dog, one (1) retrieves frommemory some subset of exemplars from the total set and (2) compares them to thetarget for similarity. If the similarity is extensive enough, the target is categorized as adog. If the target more closely resembles exemplars from a contrast category (e.g.,“cat”) than those of the dog category, it is categorized as a non-dog.Suppose that the exemplar theory is correct. Focus on the fact that a selection is made

from the exemplar-set during the retrieval process. What guides this selection? Perhapscontextual factors “prime” certain exemplars rather than others when the target iscalled to the subject’s attention. This is the view of Medin and Schaffer (1978), whose“context model” of categorization assigns importance to context. This raises animportant question. What fixes the content of an individual’s personal concept underthe exemplar theory? Is it fixed by all exemplars in the subject’s memory and theirproperties? Or only by those exemplars retrieved from memory on a given occasion? Ifthe latter were right, the concept’s extension would be highly variable; it would vary ina manner that accords poorly with our customary conception of concept-content.Although an individual’s concept can undergo some change as a result of new learningabout a category, it shouldn’t change as a function of the exemplars freshly sampled oneach occasion. If we therefore make the natural assumption that a concept’s extensionshould be fixed by all exemplars stored in memory, categorization might turn out to behighly fallible. Whenever a skewed choice of exemplars is used for comparison withthe target, incorrect categorizations could easily be made.This may well be what transpired in the experimental finding of order effects

in knowledge attributions (Swain et al., 2008). Subjects were asked to indicate howfirmly they agreed or disagreed with the statement that knowledge is achieved in aTruetemp-like case. Subjects were asked to make this rating either before or afterrating other cases involving knowledge or non-knowledge. Subjects first presentedwith a clear case of knowledge were significantly less willing to attribute knowledge inthe Truetemp case, whereas subjects first presented with a clear case of non-knowledgewere significantly more willing to attribute knowledge in the Truetemp case. Plaus-ibly, those first presented with a clear case of knowledge were, in effect, primed toretrieve a class of knowledge exemplars that set a high standard of comparison—whichinduced less willingness to attribute knowledge in the ensuing Truetemp case. Thosefirst presented with a clear case of non-knowledge were primed to retrieve a class ofknowledge exemplars that set a much lower standard of comparison—which inducedgreater willingness to attribute knowledge in the ensuing Truetemp case. Assuming


that at least one of these groups of attributions was rife with error, the mistakes seemattributable to the categorization process itself, which (by hypothesis) allows exemplarselection to play a major role in decision-making.

Another approach to concepts, the so-called theory-theory, takes exception to the ideathat either exemplars or prototypes contain all of the information, or the most importantinformation, stored in our concepts. Smith,Medin, andRips (1984),Murphy andMedin(1985), Gelman (2003), and others argue for a “dual” or “hybrid” theory of concepts.This approach holds that concepts often store two kinds of information about theircorresponding categories: an “identification procedure” and a “conceptual core.” Thefirst kind of information involves relatively superficial—typically perceptual—propertiesof previously encountered members of the category. The second contains informationabout deep, hidden, or essential features of members of the category. This is called theconceptual “core.” Information contained in the conceptual core is often considered bysubjects to bemore diagnostic and central to the identity of the category than informationcontained in the identification procedure. Nonetheless, when quick categorizationjudgments must be made, and subjects lack detailed information about “deeper” prop-erties of the instances to be categorized, they will use properties contained in the“identification procedure.” Naturally, this can lead to errors.14

Thus, inspection of empirically based theories of categorization suggests that infalli-bility of judgment is not to be expected. It is therefore perfectly appropriate to worryabout the level of reliability of categorization. This process cannot be assumed, a priori,to have a high enough reliability level (whatever “high enough” amounts to) to escapeskeptical challenge.

10 A Dispositional Conception of ClassificationalCompetence

Some philosophers might argue that a proper understanding of concept possessionwould avert the worries posed by X-phi and psychological theories of concepts andcategorization. On this view, we can determine from the armchair that people arecompetent in deploying their own concepts. Classification dispositions are constitutiveof the contents of their own concepts. So, as long as the scoring criterion features thecontents of person-specific concepts, it will be a priori that people are competent andhence reliable. If so, empirical evidence is irrelevant, contrary to what philosophicalnaturalists contend. Concept possession, on this view, consists in having dispositions tosort or classify things. It’s either a necessary or an a priori truth that if someone isdisposed to classify certain things as Fs and others as non-Fs, then all the things hewould classify as F are Fs (according to his concept of F, at any rate) and all the things hewould classify as non-Fs are non-Fs (according to his concept). Hence, there is no roomfor error—as long as we stick to the F-concept (or F-conception) of the individualdoing the classifying.


Roughly this view is defended by Ludwig (2007).15 Chalmers and Jackson (2001)spell out a similar idea in the following passages.

If a subject possesses a concept and has unimpaired rational processes, then sufficient empiricalinformation about the actual world puts a subject in a position to identify the concept’sextension. For example, if a subject possesses the concept ‘water’, then sufficient informationabout the distribution, behavior, and appearance of clusters of H2Omolecules enables the subjectto know that water is H2O, to know where water is and is not, and so on. This conditionalknowledge requires only possession of the concept and rational reflection, and so requires nofurther a posteriori knowledge. (2001: 323)

If something like this is right, then possession of a concept such as ‘knowledge’ or ‘water’bestows a conditional ability to identify the concept’s extension under a hypothetical epistemicpossibility, given sufficient information about that epistemic possibility and sufficient reasoning.That is, possession of these concepts in a sufficiently rational subject bestows an ability to evaluatecertain conditionals of the form E ! C, where E contains sufficient information about anepistemic possibility and where C is a statement using the concept and characterizing itsextension, for arbitrary epistemic possibilities. And conceptual analysis often proceeds preciselyby evaluating conditionals like these. (2001: 324)

It is not necessary to dispute the letter of these claims to draw a rather different moral. Itmay be true that in a sufficiently rational subject, sufficient information about a possibilitybestows a conditional ability to identify the concept’s extension. Nonetheless, it is doubtfulthat live subjects satisfy such idealized conditions. Human subjects may never be “suffi-ciently rational” and the information provided in thought experiments may characteristi-cally be insufficient. In the terrain we are dealing with, and given our question about defacto reliability, it is not germane to appeal to the ideal operations of some ideal creature.Instead we should consider the customary operations of human classificational resources,warts and all. Chalmers and Jackson speak abstractly of concept possession, with nomanifest concern for the actual routines or heuristics used in representing categories andmaking categorization decisions. We need to get “down and dirty” to see how humanconcept possession and categorization actually work to get a true measure of ways thatthey might go wrong. This, of course, invites us into the psychologist’s laboratory—or atleast invites us to read the journals where laboratory work is reported and interpreted.

11 A Social–Epistemological Perspective on IntuitionalEvidence

I now want to introduce a fresh element into the debate over the evidential value ofintuitions. Suppose we assemble all of the sorts of considerations discussed to this pointand decide that classification intuitions have a reliability of, say, 0.60, or 0.70. In otherwords, they are reliable indicators of the truth of their contents at roughly the 0.60–0.70 level—greater than 0.50, but not terribly high. Would this be a defeat of intuition


skepticism or a vindication? It could be argued either way. On the vindication side itcould be argued as follows: “With that modest level of reliability, one surely isn’tjustified in believing (the truth of) an intuition’s content” (e.g., that Smith in the Gettierexample doesn’t know). “This modest level of evidential support isn’t sufficient forflat-out belief.” On the other hand, it could be argued that this level of reliabilityprovides enough evidential support to entitle someone to adopt a moderately strongdegree of credence in the intuition’s content. That entitlement, or justification,constitutes a defeat of the skeptic—perhaps not a ringing defeat, but still a defeat.

Let us sidestep the potentially irresoluble question of what skepticism-avoidancerequires (justifiedness of full belief or only something weaker). Let us simply askwhether there is any way to enhance the evidential force of intuition. There is sucha way, I suggest, even without seeking new experimental findings or fresh intuitions.Until now our entire focus has been the evidential value of a single person’s intuitions.But why should this question absorb all of our attention? We should be prepared toadopt a social–epistemological perspective and enjoy its potential fruits. It is a datum ofphilosophical practice (with clear parallels in other fields) that when a method yields ahigh degree of consensus among practitioners, or a highly replicable finding, levels ofconfidence rise. In the present case, the philosophical community is swayed when anintuition is widely shared about a particular case. When this occurred with the Gettierexamples, the community’s reaction was a strong conviction that the cases are notinstances of knowledge. Are philosophers entitled to such a strong conviction? If asingle person’s intuitive judgment has only modest evidential weight, could it transpirethat many people’s intuitive judgments collectively have massive evidential weight?

Absolutely, and this is easy to explain in terms of the reliable-indicator approach toevidence conjoined with probabilistic facts. To say that one person’s intuitions arereliable indicators of (say) the truth of their contents is to say that this person’s intuitionshave a probability of being right greater than 1/2. Suppose that all people whoseintuitions are consulted are, in this sense, minimally competent. According to a well-known theorem, the Cordorcet Jury Theorem, the probability that a concurringmajority of such competent individuals will be right is a swiftly increasing functionof their total number, converging to 1 as their number tends to infinity (List andGoodin, 2001). In other words, as the size of the group increases, the reliability of amajority of the group being right—even a slight majority—increases dramatically.Even without unanimity, a majority of concurring individuals can have a collectiveprobability of being right that greatly exceeds the probability of any one of them beingright. Given the reliable-indicator approach to evidence, the weight of their evidencecould therefore considerably exceed the evidential weight of a single individual’sintuition. The theorem holds when three conditions are met: (1) there are two options,(2) each group member has a competence greater than 1/2, and (3) the members’beliefs are conditionally independent of one another. The independence condition isnot trivially satisfied. If there are common causes of different individuals’ beliefs, theindependence condition isn’t met. (Interesting results might still obtain even for


interdependent beliefs, but such results are weaker than under the independencecondition. See Dietrich and List, 2004 for discussion.) What may be inferred fromthis is that it is safe to trust concurring intuitions of a large group in a two-option choiceif their intuitions are independent, even if the members are singly only somewhatcompetent. Whether these two conditions are satisfied is an empirical question.

What does this social–epistemological analysis imply about the justification thataccrues to philosophers from intuitional evidence in the most favorable cases? Hereis one way a critic might respond. What the foregoing approach emphasizes is theevidential weight arising from a collective intuitional fact, e.g., the fact there is a certaindistribution of intuitions about a Gettier case among practitioners in the field, forexample, 900 to 20. But although this is a collective fact, and although this fact is a verygood indicator of the truth of the content of the majority’s intuition, this collective factis not a piece of evidence possessed by any individual. Each individual possesses only hisown intuitional piece of evidence; he doesn’t possess the collective pattern of intu-itions. Now, the critic will continue, a person is only justified or unjustified in virtue ofitems of evidence he possesses. So despite its initial attractions, the foregoing approachdoesn’t really help.The critic raises a good point. We have not said enough about evidence possession

to enable us to draw conclusions about justification. However, that matter requires ourattention anyway (as acknowledged in footnote 1), so let us turn to it forthwith. Giventhat p is an item of evidence (which entails that p is true), under what circumstances is pevidence that person S possesses? Two principles of evidence possession may beproposed, one for basic evidence possession and the other for non-basic evidencepossession.

(1) [Basic evidence-possession] If S is in a conscious mental state at t that is anevidential state, then S possesses that evidence at t.

(2) [Non-basic evidence-possession] If proposition p is true and S justifiably believesp at t, then S possesses p as an item of evidence at t.

Let us apply these principles to the question before us. If having an intuition is a pieceof evidence, then according to principle (1) a person’s having such an intuition implieshis possessing that item of evidence. Obviously, this principle does not cover the case ofa collective intuitional fact. No individual has (or is the subject of) such a collectiveintuitional fact. But nothing precludes an individual from justifiably believing thatthere is such a collective intuitional fact. He might have such a justified beliefby dreceiving appropriate testimony from the 920 individuals in question. If so, thenaccording to principle (2) he non-basically possesses this item of evidence, and hispossession of this evidence might make him justified in holding a very high degree ofbelief—or a categorical belief—in the content of the widely shared intuition.The foregoing principles only address the matter of evidence possession. They are

silent on the question of whether an evidence possessor is justified in believing (orholding another attitude) toward a hypothesis that the (possessed) item of evidence


supports. This is tricky terrain, and I won’t try to settle matters there. It is tempting torequire not only that the evidence possessed does support the hypothesis in question butalso that the evidence possessor is justified in believing that the evidence supports thathypothesis. But this additional condition might be too strong. Specifically, it might betoo strong for the case of basic evidence.

Finally, the preceding discussion suggests that the empirical study of intuitions andtheir processes of generation might be fruitfully re-directed. Instead of focusing on thetraditional question of whether intuitions are “good enough” to qualify as evidence (orjustification), we might instead ask how to structure or modify our philosophicalclassification games so as to maximize the evidential weight of intuitional outcomes.Both the choice of respondents and the conditions in which their intuitions are elicitedmight be controlled and/or tweaked so as to avoid circumstances or situations thatpromote bias and error. Such improvements could be a constructive upshot of theoriginally negative program of experimental philosophy. It would yield an ameliorativeapproach to intuitional methodology, an approach that would facilitate the prospects ofcognitive science playing a positive role in improving philosophical methods.

12 Intuitional Methodology and the A PrioriI promised at the outset to address two sorts of questions about intuitions or intuitionaljudgments. One is the question of whether such states constitute evidence at all forpertinent philosophical questions, and how strong such evidence is. The second is thequestion of what species of evidence intuitions provide: a priori evidence, a posteriorievidence, or some third type? In each case, of course, the importance of distinguishingbetween first-order questions and second-order questions was highlighted. I have nowcompleted what I wish to say about the first question. In brief, without dismissing orminimizing the existing challenges to the evidential status of intuitions, there may stillbe reason to think that intuitions constitute bona fide evidence for philosophicalconclusions of interest, at least in favorable cases (e.g., cases with high levels of intuitiveagreement). To this extent our discussion is compatible with a traditionalist positionthat embraces intuitional methods in philosophy. At the same time, much of thesecond-order evidence needed to legitimize or secure intuitional methodology wassaid to be empirical, including evidence from empirical science. The latter thesis isdefinitely not traditionalist; it resonates with the program of philosophical naturalism.All of this discussion, however, concerned the legitimacy and strength of intuitionalevidence, not the character or species of evidence it comprises. This final section turns tothe taxonomical question: Is intuitional evidence a priori or a posteriori? And whattype of second-order evidence is needed to address the taxonomical question?

It should first be noted that the terms a priori and a posteriori are not customarilyinvoked to classify types of evidence. Far more commonly they are used to classify typesof knowledge or justification. So in this section I shall nudge the discussion over to thequestion of whether intuition-based beliefs have a priori justification or warrant.


The tradition draws the a priori/a posteriori distinction (whether applied to justifi-cation or to knowledge) by linking such classifications to the cognitive source of thejustification or warrant. How shall we understand “source”? The straightforwardunderstanding is in terms of the causal processes responsible for the warrant. A poster-iori warrant is warrant produced by experiential, or perceptual, processes; a prioriwarrant is warrant produced by processes of intellection or ratiocination. Things gettricky, however, when we consider the possibility that a given justified belief might bethe product of both experience and ratiocination. How should its justification beclassified? The familiar answer accentuates the negative: a belief ’s warrant is a priorionly if it is not influenced by any experiential elements, otherwise it is a posteriori. Thisis an unsatisfactory way to configure the taxonomy. It implies that warranted doxasticstates with neither experiential nor ratiocinative sources are a priori. Belief basedexclusively on memory, for example, has a source that is neither experiential norratiocinative. It is misguided, therefore, to classify its type of warrant as a priori.A positive approach is therefore preferable, but how exactly should it be formulated?

One approach is through a distinctive phenomenology. Plantinga (1993) pursues thisline. He seeks to associate the a priori with a particular kind of intellectual “seeing,” anon-perceptual, non-sensuous seeing. He is convinced that whenever one “sees” thata proposition a true, in this sense, there is a unique phenomenology involved. Hedespairs, however, of giving it a positive characterization. “Seeing,” he says, is a way offorming a belief “with that peculiar sort of phenomenology with which we are wellacquainted but which I can’t describe in any way other than as the phenomenologythat goes with seeing that such a proposition is true” (1993: 106).16

Philosophical naturalists will be skeptical of the claim that there is a distinctivephenomenology for all a priori beliefs. It is especially doubtful for classification beliefs,which is the (alleged) species of the a priori currently being examined. Even within thecategory of classification judgments, it is obscure what might be the common phe-nomenological thread among them. It is even more obscure, by my lights, what mightbe the distinctive phenomenological thread uniting all candidates for a priori statusacross the standard domains. If one weren’t a rationalist philosopher with priortheoretical commitment to such a distinctive phenomenological unity, what are thechances that one would expect to find such a common thread across precisely thesedomains: mathematics, classification judgment, etc.? I regard the phenomenologicalunity thesis as a piece of highly “creative” speculation.Many rationalists delineate the special character of their domain by reference to some

modal properties of its distinctive propositional contents. Contemporary theorists ofthe a priori, however, have generally abandoned any simple link between a prioriwarrant and necessity. Fallibilism about the a priori is the dominant view these days,and rightly so (Casullo, 2003: chap. 3). What link between a prioricity and modal status,then, might be viable? Addressing the domain of classification judgment, Williamson(2007) proposes a tight link between knowledge of classification propositions and know-ledge of associated counterfactuals—a species of modal propositions. He interprets


philosophical thought-experiments as episodes inwhich one contemplates a counterfactualproposition and renders a judgment on its truth-value.Does this point in the right directionfor understanding a priori cognition—at least in the domain of classification judgment?Unfortunately, the proposed counterfactual interpretation of philosophical thought

experiments is too strong, as Ichikawa and Jarvis persuasively argue (Ichikawa, 2009;Ichikawa and Jarvis, 2009). Applied to a Gettier counterexample, it demands that acounter-instance to the JTB analysis of knowledge be suitably close in modal space tomake true the relevant counterfactual. This is unmotivated. Any possible instance ofjustified true belief without knowledge refutes the JTB account, whatever its locationin modal space.

A second problem is how the a priori can be fitted into this picture. Williamson’soff-line simulation account of counterfactuals assessment is not implausible, but it isquestionable whether it qualifies as a priori. He himself raises doubt about this.Furthermore, he rightly highlights important obscurities in the notion of the a priori.Specifically, it is difficult to know where to draw the line between experientialcognitive activities that play a purely evidential (justification-relevant) role in the run-up to a belief versus ones that play a merely enabling role. He comments:

In such cases, the question “A priori or a posteriori?” is too crude to be of much epistemologicaluse. The point is not that we cannot draw a line somewhere with traditional paradigms of the apriori on one side and the a posteriori on the other. Surely we can; the point is that doing soyields little insight. (2007: 169)

To the extent that this complaint expresses a general dissatisfaction with the a priori/aposteriori distinction, it resonates with a dissatisfaction of my own in an earlier article(Goldman, 1999a). Mymain dissatisfactions there were with the simplistic dichotomy ofepistemic sources and its lack of attunement to the psychological literature. If one wantsto draw fundamental distinctions between psychological sources of belief, there aremanyother, more promising paths to take. One might follow the modularity path, and dividesources of belief in terms of the mental modules from which they emanate. This mightindeed pinpoint numerical cognition and logical cognition as interesting sources ofbelief. But they would also wind up distinguished from one another, not lumpedtogether as the traditional taxonomy proposes. Another promising path is to adopt therecent “dual-process” model of the mind and seek to understand different epistemicstatuses in terms of the distinction between automatic versus controlled levels of process-ing (Evans and Frankish, 2009). Either program is likely to yieldmore fruitful insight intodiverse cognitive methods or mechanisms than the antiquated distinction epistemologyhas saddled itself with for far too long. Perhaps the old distinction was worth retainingwhen it was thought to track a distinction between grades of justification. But now that itis pretty much universally conceded that a priori status need not correlate with superiorstrength or degree of warrant, why retain a distinction with so little to be said for it?

Nonetheless, traditionalists will remain anxious to retain the a priori category andinsist that intuitional judgments exemplify the category. This move is critical if they are


to preserve the sanctity of the armchair method for philosophy (where armchairmethod is equated with a priori method). Moreover, classification judgments, at firstblush, are a clear example of a priori cognition. Don’t they essentially involve deter-minations of conceptual relations between two contents? Confronted with a Gettiercase, one (tacitly) accesses the content of one’s knowledge concept and the content ofthe specified scenario and reasons to the conclusion that what satisfies the latter (withcertain added presuppositions) does not satisfy the former. Isn’t this an unexceptionableexample of non-experiential and ratiocinative cognition?17

While this description may be a true and relevantly complete account of someprocesses of classification judgment, it doesn’t fully describe other such processes. Inparticular, philosophers often consult with their students and peers about examples;they try out the scenarios on others and solicit their reactions. The testimonies theyreceive on the matter—whether they support the subject’s own initial view or providedefeaters for it—are all justificationally relevant to the subject’s view, either positivelyor negatively.18 Since all such testimony arrives via perceptual means, these are aposteriori components of the total mental process that is evidentially relevant to thejudgment’s epistemic status. On the standard approach, then, the judgment should be aposteriori. Certainly it isn’t purely a priori.19

Of course, the foregoing “incursion” of empirical elements into the domain ofclassification judgment is not an incursion involving empirical science. The mere sharingand comparing of intuitions, as done by ordinary philosophers of a non-experimentalstripe, does not demonstrate the relevance of empirical science to the question of apriori status for intuitional judgments. Next, however, we turn evidence of this otherkind. There is evidence from cognitive science that challenges the orthodox assump-tion that the cognitive processing of relations between concepts is a purely non-experiential, i.e., “non-modal,” activity (in the psychologist’s sense of “modal”). Theorthodox view of concepts in both philosophy and cognitive science regards them as“abstract” or “amodal” mental representations, that is, representations devoid ofperceptual contents. But a growing wing of cognitive science led by Barsalou (1999,2003, 2009) has accumulated some fairly striking empirical evidence suggesting thatconceptual processing heavily recruits modality-specific (i.e., perception-derived) mater-ials. In philosophy this neo-empiricist theme has been developed and defended byPrinz (2002). While still a minority view, it shows that it cannot be taken for grantedthat conceptual processing is a non-experiential affair. This is a scientific question! (Norcan it be resolved by introspection. The level of processing that might includemodality-specific representations is below the level of consciousness, hence inaccessibleto introspection.)Here is a brief sample of the evidence Barsalou marshals in support of his neo-

empiricism (see Barsalou et al., 2003). In one set of experiments, participants wereasked to list the features of a concept verbally. One group was instructed to use imageryduring their performance whereas the others (the “neutral” group) were asked to usewhatever representations they chose. The neutral group showed very substantial


similarity to the imagery-instructed group, as judged by their response times and errors.Moreover, perceptual similarity has been found to affect property verification inconceptual tasks. When participants were asked to verify whether mane is a propertypossessed by pony, they were faster if mane had been previously verified for horse ascontrasted with being verified for lion. Apparently, verifying mane for horse primed aperceptual shape for mane that was later used to verify mane for pony, whereas verifyingmane for lion activated a different shape that could not be so used.

The role of modality-specific representations in conceptual processing is furtherconfirmed by evidence from cognitive neuroscience. Studies by Martin and colleagues(Martin et al., 2000; Martin and Chao, 2001) have localized various forms of conceptualprocessing in modality-specific brain areas. Conceptualizing object motion, forexample, activates areas near brain centers for perceiving motion; and conceptualizingactions that an agent could perform on an object activates the motor system (considereda modality-specific code as contrasted with an abstract code). All of this evidence givessuggestive empirical support to the notion that manipulating concepts—includingapplying them to examples—involves perception-related cognition, which clasheswith the classical picture of non-experiential, a priori cognition.

I am raising doubts about the contention that classification judgments have an apriori status. But I am not insisting on this. In particular, I am prepared to admit thatclassification judgments have a partly a priori status, stemming from the ratiocinativecharacter of the principal processes that influence classification judgment. Is this an oddconcession from a self-proclaimed philosophical naturalist? Is it an abandonment ofnaturalistic scruples? Not so. As argued in (Goldman 1999a), a priori warrant neednot be wholly anathema to naturalism. It is just a status that arises from a certain classof cognitive processes. However, in light of our earlier remarks about the “new” apriori—which admits fallibility and lays no claim to extraordinary evidential strength—naturalists have much less reason to decry or resist it. Naturalists are suspicious of the apriori partly in virtue of its historically “elitist” claims, especially its claim to infallibility.Since recent a priorists have substantially toned down their claims and their rhetoric,there is far less cause for resistance.

A final point. Even if cognitive scientific evidence were to push us in the orthodoxdirection, revealing that classification judgment is a purely amodal cognitive activity, thiswould not affect the “big picture” of philosophical methodology I highlighted from thestart. As I stressed from the beginning, philosophical methodology “as a whole” includestwo levels: (1) making classification judgments about cases and drawing inferences aboutthe contents of target classifiers, and (2) engaging in second-order investigationsintended to illuminate the reliability and character of activities at the first level. Thesesecond-order investigations are ones that the philosophical profession has an (epistemic)obligation to undertake. It must scrutinize the worthiness and appropriateness of its first-level methodology, and this should be done using all relevant tools at its disposal,including scientific tools. Members of any profession, especially ours, should seekjustification for believing that conclusions reached via its dominant methodologies


are justified conclusions, well supported by all relevant evidence. We should seek to bejustified in believing that many or most of the philosophical judgments we makeare justified. If this requires appeal to existing scientific evidence—or the generationof fresh scientific evidence by means of new experimentation—then so be it. Thesekinds of evidence must be consulted and/or gathered. Furthermore, I have argued thatthe justification of existing philosophical methods (as well as their improvement) doesrequire such scientific evidence. These theses comport well with the outlook ofphilosophical naturalism.20

Notes* Forthcoming in: Proceedings and Addresses of the American Philosophical Association.1. It is unclear exactly which set of intuitions, or judgments, experimental philosophers mean to

challenge with respect to evidential status. Do they mean to challenge the evidential status ofclassification judgments in general? That would constitute a very wide-ranging and radicalskepticism. An alternative approach is to defend amore selective skepticism (seeNado, submitted),i.e., an epistemological challenge to some subset of classification intuitions or judgments.Whichsubset? First, it would definitely be restricted to classification judgments rendered in response toimagined, or verbally described, hypothetical cases. Second, it may be restricted to judgmentsconcerning philosophical categories, not ordinary physical-object categories such as “table” or“cat.”Third, it may be restricted to judgments concerning difficult philosophical cases. These are,after all, the cases that receive the bulk of discussion in the philosophical literature and arecertainly the cases most intensively researched by experimental philosophers. They are also, ofcourse, among the most important cases that require resolution for purposes of philosophicaltheory testing.Well-grounded doubts about the evidential status of any of these subsetswould beof interest. Clearly, however, the wider the set, the more interesting the challenge. I shall not tryto settle precisely which set of cases is to be targeted. I leave this factor unresolved, as it is left inmost of the experimental philosophy literature. (For some attention to it, though, seeWeinberg,2007, especially pp. 334–5.)

2. This terminology concerns “objective” states of evidence and evidential relations. But wemust also be concerned with when an item of evidence is evidence for a given person. In otherwords, we shall want to be concerned with what it is for someone to possess evidence. Thistopic is postponed until Section 11.

3. The intuiting/intuited distinction is due to Lycan (1988), who also opts for intuiteds ratherthan intuitings as the primary evidential entities.

4. A thorough-going rebuttal can be found in Brown (forthcoming).5. Typically it will be a person or event featured in the example that will be the candidate bearer

of the property in question, not the example (or scenario) as a whole. For convenience,however, I shall often speak as if the example is the candidate for instantiating or exemplifyingthe target property or relation.

6. Weinberg (personal communication) comments that this was not his intention. Indeed, inWeinberg (2007) he offers a different account of the epistemological weakness of intuitionalmethods, namely, that their possible errors are not open to our detection and correction.However, even if the unreliability flaw is not Weinberg’s primary complaint, it may havebeen what his co-authors intended. At a minimum, it is a reasonable idea for others toadvance. Concerning Weinberg’s own view, see Grundmann (2010).


7. In earlier writings (Goldman and Pust, 1998; Goldman, 2007) I have sometimes muddied thewaters by talking about ‘mentalist’ and ‘extra-mentalist’ classifiers and treating concepts in thepsychological sense (rather than the content sense) as one type of classifier, indeed, a type thatshould be favored as the best for philosophical purposes. Here I offer a cleaned-up view of theterrain, which descends from the earlier treatment but is intended to improve upon it.

8. These considerations represent a different perspective than the one defended by Sosa (2007):“[W]e can engage in our . . . philosophical controversies, and regard them as objective,without ever raising the question of the ontological status of the entities involved, if any.Mostly we can conduct our controversies . . . just in terms of where the truth lies with regardto them, leaving aside questions of objectual ontology” (2007: 59). The position defendedhere, by contrast, holds that questions of objectual ontology cannot be ignored. Otherwise,we cannot be clear about which propositions intuitions purport to be right about, or provideevidence for, a matter that is crucial to intuitional reliability.

9. This linkage will be particularly attractive if one adopts a strongly pragmatic approach to theinterpretation of lexical items, especially the sort of approach associated with relevancetheory (Sperber and Wilson, 1998; Carston, 2002). According to this approach, wordmeanings “do not express full-fledged concepts, but rather concept schemas, or pointersto a conceptual space, on the basis of which, on every occasion of use, an actual concept (aningredient of a thought) is pragmatically inferred” (Carston, 2002: 360). If this is right, thereare few concepts stably expressed by public words or phrases. The content of an individual’sthought on a particular occasion, and the truth-conditions it aspires to fulfill, is very much afunction of pragmatic elaboration. Thus, the evidential reliability of intuitive judgments islikely to be maximized by considering the propositions a subject constructs “on the fly.”These propositions are not fully fixed or adequately captured by public word meanings. I amnot endorsing this position, but merely commenting on the evidential story one mightreasonably prefer if one were to adopt it.

10. Does our treatment of “personal concepts” tacitly amount to “individualism” about thecontents of (sub-propositional) attitudes? Hasn’t Burge (1979) blown individualism out ofthe water? How, then, can we take seriously the content of “person-specific” concepts orcontents, which sounds suspiciously like the endangered species “narrow content”? I wouldreply that however strong is the case for some social component in determining the contents ofpropositional and sub-propositional attitudes, that component must be limited. Inter-individ-ual differences must be honored. Even if the correct use of the verb ‘refute’ makes it anachievement term, more than a few English speakers nowadays use it in a different sense, i.e.,(merely) try to prove that someone is mistaken. (Thanks to Bob Matthews for the example.)There is a similar discrepancy between the correct use of ‘infamous’ and the widespread (mis-)use in which it is synonymous with “famous”. Wemust clearly preserve space for differences inperson-specific concepts and not allow an individual’s idiolect to be swamped by proper usage.

11. It might be suggested that, for some abstract entities, namely Lewis’s (1983) “natural”properties, it isn’t that we have a special epistemic power to grasp them but that they havea greater propensity, or “eligibility,” to become the objects of our thoughts. This should notstrike (science-oriented) philosophical naturalists as terribly helpful or illuminating. Howdoes it transpire that certain privileged abstract objects insinuate themselves into ourthoughts or thought contents? This is as objectionably mysterious as any supposed cognitivefaculty for “grasping” such objects.


12. One rationalist, Bealer, candidly admits that the most we can establish about our access tophilosophically interesting classifiers is the possibility of grasping them fully (Bealer, 1998).But if all we can secure is the mere possibility of full access, why should we have confidencein the ability of our intuitions to track the truth about purported “matches” or “fits”between a case and a classifier?

13. Sense (ii) is what I call a “weak” sense of ‘know’ (Goldman, 1999b; Goldman and Olsson,2009). To defend the existence of such a sense, consider the following. Lucy is reading amystery. She knows that her friend has already read the book, so she says to him: “Don’t tell mehow it ends. I don’t want to know how it ends, because that would ruin the story.”What doesshe mean by “I don’t want to know how it ends”? If the only sense of (propositional) knowingwere given by (iii), the state she says she doesn’t want to occupy would be the state of having ajustified and unGettierized belief in a truth about the ending. Is this the state Lucy wants toavoid? To avoid this state, it would suffice to have an unjustified belief in some such truth, or tohave a justified but Gettierized belief in such a truth. Surely, however, these are not states Lucywould be happy to occupy. She wants to avoid them too. What she means is that she doesn’twant to believe any truth about the ending. So she seems to use ‘know’ in a sense that consists ofbelieving a truth. This is the “weak” sense specified by (ii).

14. My discussion of the theory-theory approach has benefited fromBlanchard (2009), although hisconcerns are rather different frommine. Chalmers and Jackson may respond that in this case thesubject lacks (ideally) sufficient information tomake a categorization. True enough. But almost allthought experiments created by philosophers are underspecified in the sense that suitableadditional specifications could affect the proper classification. That is even true of Gettiercases. So we can hardly insist that the reliability of classification processes be scored only byreference to cases inwhich subjects receive “ideally sufficient information” about the target case.

15. I don’t mean to imply that Ludwig’s account is as simplistic as the one I present here.I introduce this mainly to provide a foil for the “structure-and-process” conception ofconcepts that I proceed to discuss in the next section. I myself suggested something vaguelysimilar to the dispositional account in Goldman (2007: 14–15).

16. Which kind of proposition does he mean in speaking of “such” a proposition? He cannotmean a necessary proposition because he admits that our a priori beliefs are fallible. Appar-ently, we can “see” that such a proposition is true and necessary when it isn’t really necessary.

17. Of course, in the normal play of a classification game, a respondent reads or hears (1) aninterrogator’s description of a case and (2) his query about whether the case exemplifies aspecified classifier. So the respondent’s cognitive episode begins with perceptual experience.This poses an obvious obstacle to the (complete) episode qualifying as justified a priori.However,we are obviously “subtracting” the initial segment of the cognitive episode. The part thatinterests us is the post-perceptual analysis of the relationship between the “case” and queriedpredicate. Since the case could have been invented by the respondent herself—without anymajor change in the “core” of the methodology—I am setting aside the perceptual input aspect.

18. I assume here that there is no inference from the testimony of others, at least at the time of thecurrent judgment. Such inference would eliminate the status of the subject’s belief as anintuition (since intuitions were defined as spontaneous, i.e., non-inferential, judgments). Still,the subject’s intuitive judgment could have been influenced by previous testimony.

19. Burge (1993) defends the view that beliefs formed on the basis of testimony are, in manycases, a priori warranted, arguing that the role of perception in testimony is not justificatory.


However, I agree with Christensen and Kornblith (1997) that this view is implausible. Asthey say, in interlocution it is precisely perception that warrants one in taking a certainproposition to have been presented as true, and therefore plays a warranting role as in typicalempirical beliefs (1997: 5).

20. The final version of this chapter has benefited from discussions of earlier versions presentedin several venues, as well as with individuals outside of those venues. An early draft was readand discussed in my graduate seminar on philosophical naturalism at Rutgers in the fall of2009. Many participants made valuable contributions in the context of that seminar, butI especially profited from criticisms, and suggestions by Robert Beddor, Thomas Blanchard,Jenny Nado, Lisa Miracchi, and Blake Roeber. Another version was presented as the(official) Romanell Lecture at the Pacific Division meeting of the APA, where JonathanWeinberg and Ron Mallon made constructive remarks during the discussion period.Subsequent revised versions were given at Arche Philosophical Research Centre at theUniversity of St. Andrews and an epistemology conference at PUCRS in Porto Alegre,Brazil. Very helpful feedback was received in both cases. I also profited from hearing JenniferBrown deliver her paper on this topic (cited here) at Arche. Finally, I have receivedextensive comments and pointers of great value from Thomas Grundmann and Holly Smith.

ReferencesAlexander, J., Mallon, R., and Weinberg, J. M. (forthcoming). “Accentuate the Negative.”

Review of Philosophy and Psychology.Alexander, J. and Weinberg, J. M. (2007). “Analytic Epistemology and Experimental Philoso-

phy.” Philosophy Compass 2(1): 56–80.Barsalou, L. W. (1999). “Perceptual Symbol Systems.” Behavioral and Brain Sciences 22: 577–660.Barsalou, L. W. (2009). “Simulation, Situated Conceptualization, and Prediction.” PhilosophicalTransactions of the Royal Society B 364: 1281–9.

Barsalou, L. W., Simmons, W. K., Barbey, A. K., and Wilson, C. D. (2003). “GroundingConceptual Knowledge inModality-Specific Systems.” Trends in Cognitive Sciences 7(2): 84–91.

Bealer, G. (1998). “Intuition and the Autonomy of Philosophy.” In DePaul and Ramsey (eds),Rethinking Intuition: The Psychology of Intuition and Its Role in Philosophical Inquiry. Lanham, MD:Rowman and Littlefield.

Blanchard, T. (2009). “Conceptual Analysis, Similarity-Based Accounts, and the Theory-Theory of Concepts.” Seminar paper, Rutgers University.

Brown, J. (forthcoming). “Thought Experiments, Intuitions, and Philosophical Evidence.”Dialectica.

Burge, T. (1979). “Individualism and the Mental.” Midwest Studies in Philosophy 4: 73–121.Burge, T. (1993). “Content Preservation.” Philosophical Review 102: 457–88.Carston, R. (2002). Thoughts and Utterances. Malden, MA: Blackwell.Casullo, A. (2003). A Priori Justification. New York: Oxford University Press.Chalmers, D. J. and Jackson, F. (2001). “Conceptual Analysis and Reductive Explanation.”

Philosophical Review 110(3): 315–60.Christensen, D. and Kornblith, H. (1997). “Testimony, Memory, and the Limits of theA Priori.”Philosophical Studies 86: 1–20.

Dietrich, F. and List, C. (2004). “A Model of Jury Decisions Where All Jurors Have the SameEvidence.” Synthese 142: 175–202.


Dretske, F. I. (1981). Knowledge and the Flow of Information. Cambridge, MA: MIT Press.Evans, J. St. B. T. and Frankish, K. (eds) (2009). In Two Minds: Dual Processes and Beyond. Oxford:Oxford University Press.

Gelman, S. (2003). The Essential Child. New York: Oxford University Press.Gettier, E. (1963). “Is Justified True Belief Knowledge?” Analysis 23: 121–3.Goldman, A. I. (1986). Epistemology and Cognition. Cambridge, MA: Harvard University Press.Goldman, A. I. (1999a). “A PrioriWarrant and Naturalistic Epistemology.” In J. Tomberlin (ed.),Philosophical Perspectives 13: 1–28. Malden, MA: Blackwell. Reprinted in Goldman,A. I. (2002). Pathways to Knowledge: Public and Private. New York: Oxford University Press.

Goldman, A. I. (1999b). Knowledge in a Social World. Oxford: Oxford University Press.Goldman, A. I. (2006). SimulatingMinds: The Philosophy, Psychology, andNeuroscience of Mindreading.New York: Oxford University Press.

Goldman, A. I. (2007). “Philosophical Intuitions: Their Target, Their Source, and TheirEpistemic Status.” Grazer Philosophische Studien 74: 1–26.

Goldman, A. I. (forthcoming). “Epistemic Relativism and Reasonable Disagreement.” InR. Feldman and T. Warfield (eds), Disagreement. New York: Oxford University Press.

Goldman, A. I. andOlsson, E. (2009). “Reliabilism and the Value of Knowledge.” In A.Haddock,A. Millar, and D. Pritchard (eds), Epistemic Value. Oxford: Oxford University Press.

Goldman, A. I. and Pust, J. (1998). “Philosophical Theory and Intuitional Evidence.” InM. DePaul and W. Ramsey (eds), Rethinking Intuition: The Psychology of Intuition and Its Rolein Philosophical Inquiry. Lanham, MD: Rowman and Littlefield.

Grundmann, T. (2010). “Some Hope for Intuitions: A Reply to Weinberg.” PhilosophicalPsychology 23(4): 481–509.

Haidt, J. (2001). “The Emotional Dog and Its Rational Tail: A Social Intuitionist Approach toMoral Judgment.” Psychological Review 108: 814–34.

Hawthorne, J. (2004). Knowledge and Lotteries. Oxford: Oxford University Press.Ichikawa, J. (2009). “Knowing the Intuition and Knowing the Counterfactual.” Philosophical

Studies 145(3): 435–43.Ichikawa, J. and Jarvis, B. (2009). “Thought-Experiment Intuitions and Truth in Fiction.”

Philosophical Studies 142(2): 221–46.Kelly, T. (2008). “Evidence,” Stanford Encyclopedia of Philosophy, E. N. Zalta (ed.), <http://plato.stanford.edu/archives/fall2008/entries/evidence/>.

Kornblith, H. (2002). Knowledge and Its Place in Nature. Oxford: Oxford University Press.Kripke, S. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press.Lewis, D. K. (1983). “New Work for a Theory of Universals.” Australasian Journal of Philosophy61: 343–77.

List, C. and Goodin, R. (2001). “Epistemic Democracy: Generalizing the Condorcet JuryTheorem.” Journal of Political Philosophy 9(3): 277–306.

Ludwig, K. (2007). “The Epistemology of Thought Experiments: First Person versus ThirdPerson Approaches.” In H. Wettstein (ed.), Midwest Studies in Philosophy 31.

Lycan, W. (1988). “Moral Facts and Moral Knowledge,” Judgement and Justification. New York:Cambridge University Press.

Machery, E. (2009). Doing without Concepts. Oxford: Oxford University Press.Machery, E., Mallon, R., Nichols, S., and Stich, S. (2004). “Semantics Cross-cultural Style.”

Cognition 92(3): B1–B12.


http://plato.stanford.edu/archives/fall2008/entries/evidence/

http://plato.stanford.edu/archives/fall2008/entries/evidence/

Margolis, E. and Lawrence, S. (eds) (1999). Concepts. Cambridge, MA: MIT Press.Martin, A., Ungerleider, L. G., and Haxby, J. V. (2000). “Category-specificity and the Brain:The Sensory-Motor Model of Semantic Representations of Objects.” In M. Gazzaniga (ed.),The New Cognitive Neurosciences, 2nd edition, pp. 1023–36. Cambridge, MA: MIT Press.

Martin, A. and Chao, L. (2001). “Semantic Memory and the Brain: Structure and Process.”Current Opinion in Neurobiology 11: 194–201.

Medin, D. and Schaffer, M. (1978). “A Context Theory of Classification Learning.” PsychologicalReview 85: 207–38.

Mnookin, J. (2008). “Of Black Boxes, Instruments, and Experts: Testing the Validity of ForensicScience.” Episteme, A Journal of Social Epistemology 5(3): 343–58.

Murphy, G. and Medin, D. (1985). “The Role of Theories in Conceptual Coherence.”Psychological Review 92: 289–316.

Nado, J. (submitted). “Intuition, Evidence, and the Threat of Skepticism.” Department ofPhilosophy, Rutgers University.

Nagel, J. (2010). “Knowledge Ascriptions and the Psychological Consequences of Thinkingabout Error.” Philosophical Quarterly 60: 286–306.

Nichols, S. (2004). “After Objectivity: An Empirical Study of Moral Judgment.” PhilosophicalPsychology 17(11): 3–26.

Nichols, S., Stich, S., and Weinberg, J. M. (2003). “Meta-Skepticism: Meditations in Ethno-Epistemology.” In S. Luper (ed.), The Skeptics: Contemporary Essays. Aldershot: Ashgate.

Nichols, S. and Knobe, J. (2007). “Moral Responsibility and Determinism: The CognitiveScience of Folk Intuitions.” Nous 41(4): 663–85.

Nichols, S. and Ulatowski, J. (2007). “Intuitions and Individual Differences: The Knobe-EffectRevisited.” Mind and Language 22(4): 346–63.

Plantinga, A. (1993). Warrant and Proper Function. New York: Oxford University Press.Prinz, J. (2002). Furnishing the Mind. Cambridge, MA: MIT Press.Prinz, J. (2007). “Can Moral Obligations Be Empirically Discovered?” In H. Wettstein (ed.),

Midwest Studies in Philosophy, 31.Smith, E., Medin, D., and Rips, L. (1984). “A Psychological Approach to Concepts: Comments

on Rey’s ‘Concepts and Stereotypes’.” Cognition 17: 265–74.Sosa, E. (2007). A Virtue Epistemology, Apt Belief and Reflective Knowledge. Oxford: Oxford

University Press.Sosa, E. (2008). “Experimental Philosophy and Philosophical Intuition.” In J. Knobe andS. Nichols (eds), Experimental Philosophy. Oxford: Oxford University Press.

Sperber, D. and Wilson, D. (1998). “The Mapping between the Mental and the PublicLexicon.” In P. Carruthers and J. Boucher (eds), Language and Thought. Cambridge: Cam-bridge University Press, pp. 184–200.

Swain, S., Alexander, J., andWeinberg, J. M. (2008). “The Instability of Philosophical Intuitions:Running Hot and Cold on Truetemp.” Philosophy and Phenomenological Research 76(1): 138–55.

Weinberg, J. M. (2007). “How to Challenge Intuitions Empirically without Risking Skepti-cism.” Midwest Studies in Philosophy 31(1): 318–43.

Weinberg, J. M., Nichols, S., and Stich, S. P. (2001). “Normativity and Epistemic Intuitions.”Philosophical Topics 29: 429–60.

Williamson, T. (2007). The Philosophy of Philosophy. Malden, MA: Blackwell.


2

Experimental Philosophy andApriority

Jonathan Jenkins Ichikawa

According to a certain kind of experimentalist critic of traditional philosophical method-ology, it is amistake to attempt to resolve philosophical questions via armchairmethods.Onthe view in question, philosophy ought to proceed by something much more like themethodology of empirical science; this is meant to contrast with the armchair engagementprototypical of contemporary and historical philosophical argument. This conclusion, saysour critic, is motivated by empirical data. Perhaps the critic has performed some relevantempirical investigation himself—he comes armed with survey data, or neuroimagingstudies, or someother sort of surprising evidence,which he takes to undermine the armchairmethodology. There is important empirical investigation, says his critique, that must beperformed before philosophical knowledge is possible. The critic in question endorses:

X-Phi: Without engaging in much more scientific investigation (perhaps in theform of surveys, neuroimaging, cognitive psychological theorizing, etc.),we cannot possess many alleged pieces of philosophical knowledge.

The critique certainly represents an important challenge to traditional philosophicalmethodology. If X-Phi is true, then many philosophers ought to be paying much moreattention to various bits of empirical research than they are. Many more of us, if thecritique is sound, should even be performing experiments ourselves, and analyzing theresults using the statistical tools prevalent in the sciences. The critique, if sound,undermines a certain armchair conception of philosophical methodology.An extremely natural thought, then, is that the critique in question undermines the

apriority of philosophy: if one cannot proceed from the armchair, but must ratherengage in empirical, scientific experiments in order to achieve philosophical know-ledge, then whatever philosophical knowledge there is or could be must be a posteriori.That is to say, X-Phi stands in obvious prima facie tension with the traditional claim:

Apriority: It is possible to have a priori knowledge of many paradigmatic philo-sophical matters (including many of those targeted by X-Phi).

Although X-Phi and Apriority appear straightforwardly inconsistent, I shall argue thatthis appearance is illusory. The proper interpretation of the experimentalist critique isnot one that bears on the apriority of philosophy. The project of this chapter is tosuggest that even if the critique is sound—even if X-Phi is true—it is plausible thatmuch philosophical investigation is a priori. Since the critique would, if sound,undermine a certain armchair conception of philosophy, it follows that apriority andarmchair methodology are less closely connected than one might initially be inclinedto think.

I won’t argue for the truth of Apriority here. I suspect it is true, but, establishing it isanother, much larger, project.1 I also won’t argue that X-Phi is true. In fact, my ownview is that X-Phi is false; but engaging that question is also a different project than thepresent one.2 My present goal is to show that X-Phi, even if true, provides no reason toreject Apriority. So I will take X-Phi as a working assumption.

In }1, I will present, identify, and precisify the experimentalist critique with whichI mean to be engaging; }2 will explore the relation between experience and apriority.In }3, I show that, thus clarified, the apriority of much philosophy is consistent with theexperimentalist complaint that philosophers need to be engaging with more experi-mental work.

1 Empirical Worries about PhilosophyIn some of its more negative paradigmatic instances, experimental philosophy collectssome empirical data—perhaps some survey data from folk judgments about philosoph-ical cases, or perhaps some neurological data about what happens in philosophers’brains when they make certain kinds of judgments—and uses it to cast doubt ontraditional armchair philosophical methodology. If these experimental philosophers areright, then philosophical methodology, as traditionally practiced, is importantly mis-guided, and stands in need of substantial revision, with a much greater emphasis onempirical investigation. In this section, I will attempt to articulate what seems to me tobe the best interpretation of the experimentalist critique, in order to explore whatwould follow, supposing that its central skeptical conclusion is correct.

1.1 Preliminaries: two caveats

Before proceeding to offer my interpretation of the critique arising from experimentalphilosophy, two caveats are in order. First, the experimental philosophy movement is awide and diverse one, and not all of its contributions are skeptical. When I say that I amcharacterizing ‘the experimentalist critique,’ I mean to be characterizing the project ofa particular significant portion of the experimental philosophy movement; I intend nosuggestion that all experimental philosophy is engaged with discrediting traditionalarmchair methods.Knobe and Nichols (2008) aptly emphasize that much experimentalphilosophy is positive; it engages with empirical philosophical questions in a waycontinuous with more traditional methodology, emphasizing only that there are

46 JONATHAN JENKINS ICHIKAWA

significant scientific resources worthy of exploitation in the investigation of philosoph-ically interesting subject matters. Much of Knobe’s own contribution to experimentalphilosophy fits this positive mode.3

But although not all instances of experimental philosophy are negative, many ofthem—includingmany of the most prominent—are. So it is, for example, thatWeinberget al. (2001) suggest that that “a sizeable group of epistemological projects—agroup which includes much of what has been done in epistemology in the analytictradition”—is “seriously undermined” by survey data they have found;4 similarly,Machery et al. (2004) tell us that they have uncovered evidence that shows traditionalphilosophical assumptions to be “spectacularly misguided,” and that as a consequence,“philosophers must radically revise their methodology.”5 Alexander et al. (2010) rec-ognize the divide between experimentalist projects I have been describing, and enjoinus to “accentuate the negative”—the skeptical pressure generated by the negativeexperimentalist program problematizes armchair philosophy and positive experimental-ist philosophy alike. When I speak in this chapter of “the experimentalist critique,” it isthis negative project of experimental philosophy that is my focus.Here are a few paradigmatic examples of negative experimental philosophy, as I am

understanding it. Weinberg et al. (2001) ran a series of surveys that apparently indicatethat some standard epistemic intuitions, such as the internalist intuition about Lehrer’s“Mr. Truetemp” case and the skeptical intuition about Gettier’s cases, vary accordingto cultural and socioeconomic background. For example, East Asian subjects seem tobe more likely than Western subjects to consider Gettier cases to be instances ofknowledge. In a similar study, Machery et al. (2004) found cross-cultural variation inintuitions about reference in Kripke’s Godel–Schmidt case. Swain et al. (2008) foundthat epistemic intuitions can vary according to the order in which they’re presented.Such survey data comprise perhaps the most typical instances of the experimentalistcritique, but negative arguments proceeding from other sorts of data should also beconsidered. For example, Greene et al. (2001) present neuroscientific data demonstrat-ing that particular brain mechanisms underlie certain kinds of moral reasoning; this isthought to undermine certain moral intuitions. The project of this section is to identifythe form of the skeptical pressure this kind of data is meant to establish.The second early clarification to be made is that the project of characterizing the

experimentalist critique is an interpretative one. Instances of negative experimentalistphilosophy that suggest the experimentalist critique are rarely explicit about just whatform the argument takes. So in characterizing ‘the’ experimentalist critique, I amattempting to distill the most plausible worry in the vicinity. Among the factorsfavoring a particular interpretation, then, will be considerations deriving from charity.Given the textual imprecision characterizing much of the skeptical argumentation inthe literature, all things equal, it is uncharitable to distill an argument form that isobviously unsound. It may be that some experimentalist critics intend arguments otherthan the one that will be the focus of this chapter—I don’t mean to be claiming that myversion of the experimentalist critique is one that all experimentalist critics would

EXPERIMENTAL PHILOSOPHY AND APRIORITY 47

endorse. But I do mean to claim that it is the best version of the argument; howeverplausible alternatives may be as experimental-philosophy exegesis, they are not plaus-ibly sound critiques of armchair methodology. I’ll begin by canvassing a couple ofinterpretive options that I will not pursue.

1.2 Facts about intuition as data

Experimentalist philosophers often emphasize that their discoveries are, from the point ofview of armchair methodology, surprising. For example, Machery et al. write that “mostphilosophers exploring the nature of reference assume that the Kripkean intuitions[about the Godel–Schmidt case] are universal.”6 They proceed to argue that thisassumption is incorrect, and take themselves, in so doing, to have undermined at leastsome of armchair methodology. On one interpretation of the argument, they understandtraditional methodology to be relying on facts about the universality of intuitions asimportant data in armchair methodology. This interpretation has the advantage ofimplying straightforwardly that the data in question are in tension with armchairmethodology: armchair methodology proceeds from premises about the universality ofintuitions—premises which the experimentalist critics take themselves to have demon-strated false or unlikely. Probably some instances of the experimentalist critique wereintended in this spirit.7 However, it is not very plausible to understand the critique asrelying on such a particular approach to philosophical methodology in generality.One reason this is so is that doing so would be too uncharitable; it would be toattribute reliance on an implausible characterization of armchair philosophy. This pointhas been adequately made elsewhere, so I won’t dwell upon it now.8 Max Deutschputs the point pithily, in discussion of another case about reference, discussed in Mallonet al. (2009):

Mallon et al. appear to believe that Evans, if only he had reflected on his own methodology,would have had to retract the claim that ‘Madagascar’ refers to the island, and would have had topatiently await the results of an opinion poll concerning competent speakers’ intuitions about thereferent of ‘Madagascar’. But that is preposterous. A philosopher of language such as Evans, just aseasily as anyone else, could have simply checked his world atlas and seen that ‘Madagascar’ refersto the island. Facts such as the fact that ‘Madagascar’ refers to the island are data for theories ofreference. Facts such as the fact that competent speakers intuit that the island is the referent of‘Madagascar’ are data for a psychological theory, one that does not have any clear bearing on atheory of reference.

Given my broader goal in this chapter, it is worth emphasizing a second reason it isunhelpful to interpret the experimentalist critique as assuming that armchair method-ology relies on facts about the distribution of intuitions as evidence. I am investigatingwhat bearing the experimentalist critique has on the putative apriority of philosophicalinquiry; if, as suggested, philosophical inquiry relies as evidence, or premises, or data,on facts about the distribution of intuitions, I take it the apriority of the relevant inquiryis off the table from the start, for the simple reason that facts about who has what


intuitions are not a priori. If armchair philosophy did depend on such empirical claims,we wouldn’t need to turn to experimental philosophy to refute the apriority of suchmethodology; we’d need only point out that any access we might have to such claimswould have to come through introspection (in the case of our own intuitions) ortestimony or observation (in the case of others’). So Apriority would be a non-starter.Insofar as my main project involves investigating what bearing the critique has onApriority, this version of the critique would pretty clearly have none.9

So there are two reasons for me not to understand the critique in the way suggestedin this subsection: it interprets it as relying on an implausible assumption aboutphilosophical methodology, and it interprets it in a way that rules out Apriority,even independent from considerations arising from X-Phi.10

1.3 Disagreement

Another way we might understand the experimentalist critique is as a skepticalchallenge arising from disagreement. When we discover that East Asian subjectsthink that Gettier cases are instances of knowledge, this challenges our own belief tothe contrary; and absent some reason to prefer our judgment to theirs, we ought not tocontinue believing that our judgments were correct. Disagreement ought to issue intoagnosticism. Sosa (2007) interprets the worry in this kind of way:

How might survey results create a problem for us? Suppose a subgroup clashes with another onsome supposed truth, and suppose they all ostensibly affirm as they do based on the sheerunderstanding of the content affirmed. We then have a prima facie problem. Suppose half ofthem affirm while half deny it, with everyone basing their respective attitudes on the sheerunderstanding of the representational content . Obviously, half of them are getting it right,and half wrong. Of those who get it right, now, how plausible can it be that their beliefsconstitute or derive from rational intuition, from an attraction to assent that manifests a realcompetence?

Not that it is logically incoherent to maintain exactly that. But how plausible can it be, absentsome theory of error that will explain why so many are going wrong when we are getting itright? Unless we can cite something different in the conditions or in the constitution of themisled, doubt will surely cloud the claim to competence by those who ex hypothesis are gettingit right.

. . .So there will definitely be a prima facie problem for the appeal to intuitions in philosophy if

surveys show that there is extensive enough disagreement on the subject matter supposedly opento intuitive access. (102)

This interpretation represents a step in the right direction from the one given previ-ously—indeed, the treatment I will ultimately give in the following subsection, anddevelop in }4, will have more than a little in common with this line. Still, there are twoproblems with understanding the experimentalist critique in this way in generality.First, as a number of authors have emphasized,11 since the relevant studies concernunreflective folk intuition, it is not clear that they ought to be afforded weight on a par

with those of considered judgments of professional philosophers. We seem to havelittle reason to regard the subjects of the surveys in question as ‘epistemic peers,’ in thesense of the burgeoning literature on the epistemic significance of disagreement.12

Second, and more to the point, understanding the experimentalist critique as oneessentially involving disagreement is insufficiently general. Although some of thestudies in question emphasize groups of people who dissent from standard intuitions,not all of them do; what are we to make of arguments arising from order effects orneuroimaging? These do not seem to involve disagreement in any straightforwardsense. To focus too narrowly on disagreement, then, seems to be an exegetical error.

1.4 Undermining defeaters

However, generalizing from the disagreement approach can, I think, avoid bothproblems given, and provide a faithful and interesting characterization of the critique.What has disagreement in common with the other sorts of experimental phenomenaused to generate skeptical pressure? In generality, such data can provide us with reasonto second-guess our initial judgments. That is to say, they can provide underminingdefeaters. In general, if I encounter someone who disagrees with me as to whether p, thisprovides me with reason to reconsider my grounds for p. Similarly, if my judgment thatp derives from a source S, and I discover some evidence that S is unreliable with respectto judgments like this one, this likewise provides me with some reason to reconsidermy grounds for p. Notice that this effect obtains even if my initial judgment was a goodone. Sometimes, upon reconsideration, I may rightly decide that my initial judgmentwas both correct and well supported after all; other times, agnosticism will be thecorrect response. But given my cognitive limitations, the discovery of such worryingpatterns provides me with reason to be circumspect.

The experimentalist concerns are easily interpretable along this model. If youdiscover that most members of a culture different from your own are inclinedintuitively to judge that Gettier’s cases are instances of knowledge, this can provideyou with some reason to consider your own reasons for thinking the contrary. If youfind that your belief derives from, e.g., general theoretical considerations about theepistemic value of non-accidentality, then you may well endorse it, having passed thisnew challenge; if, by contrast, you can identify no reason to endorse the belief, findingonly that it seems like the thing to say, then you may more properly suspend judgmentupon recognition that other people, with grounds apparently as strong as yours,disagree. Similarly, if you discover that epistemic intuitions are susceptible to ordereffects, then you know a common way in which people can make mistakes aboutepistemic judgments; it is natural and appropriate, upon such a discovery, to investigatewhether you yourself might have been making a mistake in this way. You can dothis by considering whether your judgments are stable over a variety of orders ofpresentation: does your judgment about Truetemp persist regardless of which othercases you consider immediately before it?

In its broad outlines, this form of concern is not distinctive of experimental philoso-phy. Philosophers have a long-standing tradition of questioning the grounds for one


another’s beliefs; upon discovery that a particular judgment derives from a certain sortof process, where that process is one that is not generally trusted, we find reason todiscount that judgment.13 The distinctively experimentalist critique adds two elementsto this kind of pressure. First, it sharpens our abilities to identify the sources of particularkinds of judgments. Although it is plausible that we can know a fair amount about whywe think what we do from the armchair, it should hardly be surprising that the tools ofpsychology can provide us with important new insights in this domain. As indeed it hasdone; we now know, for example, on empirical grounds, that moral judgmentsinfluence intuitions about the intentionality of actions.14 Second, experimental investi-gation can tell us more about the reliability of particular kinds of processes. This is,I take it, what occurs in the order effects case; we learn, via Swain et al.’s survey results,that unreflective intuitive judgments are susceptible to a particular kind of error; uponlearning this, we face, on pain of epistemic irresponsibility, an obligation to check tosee whether we are making this kind of error.Broadly speaking, there are two ways that this sort of defeater could manifest to

prevent philosophical knowledge. Perhaps the more obvious one derives from theinconsistency of the epistemic irresponsibility just mentioned with knowledge; giventhe empirical evidence that has been uncovered, any purely armchair method ofphilosophical investigation is irresponsible, and therefore cannot issue into knowledge.On this destruction interpretation, the experimentalist critics have found evidence thatrenders methods—ones that may previously have been reasonable—inappropriate. Onanother interpretation, which we may call a discovery interpretation, the evidence inquestion demonstrates that the methods were never reliable enough to issue intoknowledge in the first place. The easiest way to appreciate the difference betweenthe destruction and discovery interpretations is to imagine naıve armchair philosopherswho do not know about the undermining evidence. The destruction critique does notbear on such characters; they do not behave irresponsibly by failing to reckon with datathat is unavailable to them. But the discovery critique will challenge their knowledgejust as much as that of philosophers more familiar with the data: it purports to showthat, however reasonable the armchair methods may seem from a first-person point ofview, they cannot issue into knowledge. (The destruction experimentalist takes himselfto have started the fire in the armchair; the discovery experimentalist holds that it wasalways burning.15) The two interpretations needn’t be competitors; a critic mightconsistently hold that the data in question establishes both kinds of defeat—I suspectthis is the actual attitude of many defenders of X-Phi. I will say more about these kindsof undermining defeat in }4.It is easy to see, then, how the experimentalist critique demands philosophical

investigation to move beyond the armchair. There are empirical challenges to ourarmchair processes. Even if these challenges can be met, they could only be metempirically, either by demonstrating that our armchair processes are not the onesimpugned by the relevant data, or by defending those processes by showing that thedata, or its interpretation, are incorrect. According to the critique, to provide such a


response is a necessary condition for philosophical knowledge—on the destructioninterpretation, because it is necessary for responsible belief, and on the discoveryinterpretation, because until it is met, armchair methods are too unreliable to yieldknowledge. As Alexander and Weinberg (2007) write, “empirical research into thenature of intuitions generated in response to thought-experiments, rather than sup-porting the use of intuitions as evidence, challenges the suitability of intuitions tofunction in any evidentiary role” (66). Absent a defense to this challenge, philosophersproceed illegitimately.

There is much that one may say to dispute this challenge to armchair methodology,but this is not my present concern. I mean, for the purpose of argument, to bemaximally concessive to the experimentalist critique, understood as underminingphilosophical judgments in the way suggested. The central contention of this chapteris that, even conceding a rather radical form of the experimentalist critique, we have noreason to deny that philosophical investigation is often a priori. That is to say, even ifX-Phi is true—even if it is a necessary condition for some bit of philosophicalknowledge that one perform experiments to ensure that one isn’t going wrong, itmay still be that that very bit of philosophical knowledge is a priori.

This claim may strike the reader as straightforwardly contradictory. To see that it isnot, we turn now to the relation between experience and apriority.

2 Experience and the A PrioriA priori knowledge is, to a first approximation, knowledge that does not depend onexperience. So approximated, the tension between Apriority and the conclusion of theexperimentalist pressure is obvious, and leaves little grounds for optimism for reso-lution. Apriority requires that knowledge be independent from experience, but theexperimentalist pressure alleges that philosophical knowledge requires the experiencesof engaging in scientific investigation. However, the relationship between apriorityand experience is not so straightforward as this line suggests; it is widely recognized thatthe approximation needs qualification.

2.1 Experience and concept possession

A priori knowledge can depend in some sense on experience; it is consistent with p’s apriori knowability that a subject must have had certain experiences in order to knowp. That there are at least some such roles for experience is uncontroversial; consider thestandard allowance for experience that plays a role in enabling the possession ofconcepts; such are standardly thought to be exempt from the requirement that a prioriknowledge of p be independent of experience. For example, many philosophersbelieve in phenomenal concepts—concepts that can only be possessed by beingswho have undergone certain sorts of experiences. If there are such concepts, thentautologies involving them can only be thought—and so can only be known—bybeings who have had those experiences. Suppose that red is a phenomenal concept,


only possessable by those who have had the experience of seeing red things. Then inorder to think the proposition that <all red things are red>, one must have hadexperience of red things; therefore, knowledge of that proposition depends essentiallyon visual experience. (Parallel arguments could apply the point to other putative apriori knowledge beyond logical truths: <all red things are colored>, <red things areextended>, <no surface is both red and green>, etc.) But this kind of dependence isconsistent with apriority. We say that such experience is merely enabling; this contrastswith a warranting role for experience.16

Even if there are no phenomenal concepts17—no concepts such that particularexperiences are necessary for their possession—it is plausible that as a matter of fact,many of our concepts have come about via various experiences that we have had.18

Even if it is possible to possess red without seeing red things, as a matter of fact, most ofus came to this concept via that kind of experience; according to orthodoxy, this rolefor experience is still consistent with the a priori knowability of some propositionsinvolving red.This orthodoxy gives rise to two important questions. First, to what uses can the

experience in question—that necessary for concept possession—be put in a subject’scoming to believe and know p, consistent with that eventual knowledge’s apriority? Itis implausible that just any use is legitimate; otherwise, for example, it might end up apriori for me that I have had the experience of seeing something red. Intuitively, thisproposition is a posteriori, relying for its warrant on my past experience of seeing redthings; but, on the line in question, this experience is necessary for—or at least casuallyinvolved in—my possession of the concept red. So on what basis can this knowledgebe considered a priori? Second and relatedly, what if anything is special about theexperience that is necessary for concept possession? What justifies the exception to thegeneral requirement that a priori knowledge is independent of experience? Without atheoretical motivation, one suspects the rule of adhocery; is there such motivation, or isit only included to give the intuitive verdicts about cases?In the remainder of this section, I’ll lay out what strikes me as a plausible theoretical

characterization of what is essential to a priori knowledge and justification. Thischaracterization will answer both questions, explaining why, and under what circum-stances, important roles for experience necessary for concept possession are consistentwith apriority. A central upshot, for my purposes, will be that the relevant roles forexperience extend beyond experience necessary for concept possession.19

2.2 Apriority and defeasibility20

Some transitions in thought are rational, and others are not. It is rational, for example,under ordinary circumstances, to infer from <Obama went to Afghanistan> to<Obama flew on Air Force One>. The rationality of this inference is explained inpart by background knowledge concerning, e.g., Obama’s status as President of theUnited States, and the standard method of overseas transportation for the President. Itis possible for a subject to have a different course of experience, such that the inference

in question would not be a rational one; if one had experiences, for example, as of aperfectly convincing news report indicating that Air Force One had been dismantledlast year—corroborated, let us suppose, by the experience of alternate sources givingthe same news, the discussion of this surprising news by one’s peers, etc.—then onewould be rational in declining to make the inference in question.I think that this kind of dependence on experience for rational inference underwrites

the common-sense fact that the inference in question is a posteriori. However, we musttake care to articulate the reasoning here properly. For, as Philip Kitcher has empha-sized, a superficially parallel argument could threaten the apriority of any inference at all.Consider: it is rational, under ordinary circumstances, to infer from <grass is green> to<grass is green or someone is hungry>. But arguably, there is a plausible course ofexperiences such that, if a subject had those experiences, it would not be rational for himso to infer. For example, someonemight receive apparent expert testimonial advice thatdisjunctive introduction is not generally valid. True disjunctions, one might be misledto believe, require some kind of evidential or metaphysical connection between thedisjuncts. Such advice would, I suppose, bemisleading—disjunctive elimination is valid,and the intuitions requiring connectedness ought to be dismissed as Gricean pragmaticones—but it might be sufficiently authoritative such that it would be irrational to ignoreit. Given this possibility, it seems that the inference in question would be counted as aposteriori by lights similar to those marshaled earlier. And of course, the argumentwould generalize widely; once we see how experience can provide misleading evi-dence, it is plausible that some kinds of experience could undermine the rationality ofany inference whatsoever.21 It is possible to embrace this conclusion, and conclude thatthe extension of the a priori is empty—this is in effect the position of Kitcher (1983). ButI agree with Casullo (2003) that a better response is to weaken and clarify the relation-ship between defeasibility and apriority.

On my view, the ubiquity of possible empirical defeat shows that empirical defeas-ibility is not the correct model for the way in which a priori knowledge must be‘independent from experience.’That some possible experiences could underminewarrantdoes not show that contrary experiences are a part of the warrant. Compare: much of myknowledge is ‘independent of testimony’ in the sense that the rationality and know-ledge-status of my belief does not derive from testimony. I now know that I am wearinga necktie, but this knowledge has nothing to dowith anybody’s having toldme that I amdoing so.Mywarrant for this belief is perceptual, not testimonial. But it is consistent withthese obvious facts that testimonial experience could undermine my justification. If,perhaps, someone sufficiently authoritative told me that I had ingested a drug that waslikely to confuse me about what I’m wearing, or that I am the subject of an elaborateprank involving a fake necktie, this might undermine my perceptual knowledge. But ofcourse this does not show that I have only testimonial knowledge of my current outfit.

One way to attempt to draw the distinction more carefully might be to distinguishbetween, in Marcus Giaquinto’s terminology, ‘negative dependence’ and ‘positivedependence’ on experience (or testimony, or whatever).

We could mark the distinction by saying that if a belief is rationally revisable in the light of futureexperience, its retention is negatively dependent on experience; and if a belief cannot have beenjustifiably acquired unless some experience was used as grounds in the process, its acquisition ispositively dependent on experience.22

So developed, the suggestion would be that a priori knowledge is consistent withnegative dependence on experience, but inconsistent with positive dependence. Thereis much that is intuitive about this distinction, and I am not sure that it is incorrect;nevertheless, I prefer a different characterization, for two reasons. First, the notion of‘experience used as grounds,’ critical in the articulation of positive dependence, is notas straightforward to apply as might be hoped. Timothy Williamson, for example,draws attention to cases in which judgments are clearly informed—and rationallydependent on—experience, even though citing that experience as evidence wouldbe a mistake.23 This kind of feature is particularly plausibly present in various counter-factual judgments; to take one of Williamson’s examples, <if two marks had been nineinches apart, they would have been further apart than the front and back legs of anant>.24 So unless positive dependence is characterized rather carefully, linking apriorityclosely to Giaquinto’s positive dependence risks potential counterexamples. Second,Giaquinto’s distinction does not relate in any clear way to the standard story givenabout experience that is necessary for concept possession.I will attempt to characterize apriority in a way that provides a general answer to

why negative dependence reliance on concept-enabling experience is consistent withapriority, but positive dependence is not.

2.3 Propositional and doxastic justification

To properly characterize the a priori, I suggest, we should first clarify that apriority ismost fundamentally a property of propositional justification. Propositional justification isto be distinguished from doxastic justification: the former is a matter of what a subject’stotal body of evidence supports; the latter concerns whether a given belief is properlyheld. As I will use the terms, to say that, for some subject S, a proposition p is(propositionally) justified is to say that S has warrant for p; to say that S’s belief that pis (doxastically) justified is to say that S is warranted in believing as she does. To havepropositional justification is to have (good enough) reason to accept the proposition; tohave doxastic justification is to have a belief that is properly formed. A propositionmight be propositionally justified for a subject without that subject’s belief in thatproposition being doxastically justified; standard examples of such include cases inwhich a subject has good grounds in support of a proposition, but believes it on thebasis of poor grounds; for example, one might ignore one’s conclusive evidence thather husband is being unfaithful, but come unreasonably to believe that he is beingunfaithful on the basis of an astrological reading.As I think it is best understood, apriority is fundamentally a property of propositional

justification; we can define a sense of apriority for other states in the obvious ways.

A belief is doxastically justified a priori if it is appropriately held on a priorigrounds; someone knows p a priori if her a priori justified belief amounts toknowledge, etc.

Returning to the propositional sense, for p to be justified a priori is for there to bewarrant for p that does not depend for its presence on any particular past or presentexperience. Just as the presence of propositional justification in general is neutral onthe question of whether a belief in that proposition is doxastically justified, so too isthe presence of a priori propositional justification consistent with a belief inthat content’s being justified or unjustified—or justified a priori or a posteriori.That a priori propositional and doxastic justification can come apart in this wayexplains why it is consistent with the apriority of some proposition—and indeed, theapriority of justification for some belief in that proposition—that there are possiblecourses of experience that would defeat the doxastic justification for belief in thatproposition.

Although a priori justification is in this sense empirically defeasible, Ichikawa andJarvis (2012) offered a principle for distinguishing a limited subset of ways in which itcan be defeated. We suggested there that a priori (doxastic) justification can be defeatedin only two ways: first, by a subject’s rational shortcomings, and second, by evidence tothe effect that one is experiencing such rational shortcomings. As we wrote:

Indeed, almost any rational commitment to infer can be defeated due to a subject’s rationallimitations. Rational limitations may arise from limitations in one’s conceptual repertoire, limita-tions in computational capacity (e.g. in the time it takes to draw an inference), and tendencies tomake performance errors in drawing inferences, or, for that matter, any other sorts of proclivitiesthe subject has to make or exhibit confusions in attempting to execute in accordance with whathe has reason to think. (This list may not be exhaustive, but it indicates that the term ‘rationallimitations’ is not a catchall; rational limitations contrasts with limitations in experience that resultspecifically in a paucity of evidence, which might limit the subject’s ability to competently infer ina wholly different way. Rational limitations are limitations concerning the processing of evi-dence.) Testimony from a panel of expert logicians can defeat John’s rational commitment toinfer in accordance with modus ponens, but only because of John’s rational limitations vis-a-vislogic. If John were an acknowledged über-logician, he would have no reason to kowtow to thepanel of “expert” logicians any more than we have reason to defer to elementary school childrenon matters of basic arithmetic.

Similarly, almost any rational commitment to infer can be defeated due to evidence regarding asubject’s (current) rational limitations. Evidence to the effect that Jane has taken a pill that inhibitsrational capacities can defeat Jane’s rational commitment to infer in accordance with modus ponensas can evidence to the effect that Jane is crazy even if in fact Jane has not taken such a pill and isnot crazy. (Ichikawa and Jarvis 2012, p. 139)

These two forms of defeat, we said, are the only ways in which a priori justification canbe defeated.25

This framework suggests a natural explanation for why experience involved in conceptacquisition is consistent with the apriority of propositions containing those concepts: the


relevant role of such experience is merely to reduce the relevant cognitive limitations.One way to fail to be in a position to take advantage of a priori warrant is to be unable,due to limitations in one’s conceptual repertoire, to entertain the thoughts needed.Experience that plays a role in enabling such thoughts is experience that helps a subjectto capitalize on the propositional justification he already has. So construed, this experi-ence could be necessary for doxastic justification, but not for propositional justification.

3 Enabling Beyond Concept PossessionIn }2, I offered a theoretical motivation for the commonplace that experiences that arenecessary for concept possession are consistent with the apriority of knowledge ofpropositions constituted by those concepts. Such roles for experience are theoreticallyinteresting, but are of limited use for my present project of reconciling X-Phi andApriority; according to the former, epistemologists cannot know, e.g., Gettier cases tobe counterexamples to the JTB theory of knowledge unless they first engage in somefolk survey methodology—but it is utterly implausible that such surveys are necessaryfor possession of knows, or any other of the concepts constituent in the conclusion ofthe Gettier argument. Experience necessary for concept possession is pretty clearly anon-starter with respect to the present project.However, once we think of concept-enabling experiences in the way articulated in

}2—as consistent with apriority because they merely help a subject to capitalize onpropositional warrant that is already in place—it is not difficult to see that this role forexperience will generalize beyond enabling concept possession. There are other waysto be merely enabling than merely to enable concept possession.Consider the experiences typical students undergo when being trained in a priori

disciplines like math, logic, and philosophy. Sometimes, students learn contingentfacts via sensory experience—as, for example, when a student accepts an instructor’stestimony that the symbol that looks like ‘⊃’ symbolizes a truth-functional connective,with such-and-such truth conditions. So sometimes, experiences play a role in establishingpropositional justification. Prior to the relevant experiences, the students simply sufferedno rational pressure to believe the relevant facts. Such knowledge is a posteriori.But often, the way students learn in such environments is quite different; often,

students are trained to recognize rational commitments they already have. It is aconfusion to affirm the consequent; doing so reflects a rational limitation, whetherbefore the student is told so by the instructor or after. (Of course, we blame subjects forsuch failings more after they receive some training; we legitimately expect a higherdegree of rationality of some than of others.) Sometimes, experiences play importantroles in developing a priori cognitive abilities, which enables subjects better to capital-ize on their extant a priori warrant; this represents another way in which experiencecan play an important role in eventual a priori knowledge. Although the experienceplays no role in establishing the propositional a priori justification, it helps subjects tobe able to capitalize on it, thus facilitating a priori knowledge.


Plausibly, the vast majority of actual human a priori knowledge relies on experiencein this way. We are creatures who develop our cognitive abilities by relying onubiquitous experience; it is a gross understatement to say that a subject without therich sorts of educational experiences with which we have been privileged would notoften think particularly clearly. Without experience, we would not be able to havemuch knowledge at all—and this goes for the special case of philosophical knowledge,too. But this is not to say that philosophical knowledge cannot be a priori in the sensegiven earlier.

So there is no contradiction in the idea of experience—even experience unnecessaryfor concept formation—being necessary for philosophical knowledge, even thoughphilosophical knowledge is a priori. In }4, I will argue that the sorts of experiences thatare, according to the experimentalist pressure discussed in }1, necessary for philosoph-ical knowledge is indeed consistent with the apriority of philosophy. If the experi-mentalist critics are right, we must do experiments in order to achieve philosophicalknowledge. Even if we must, this provides no reason to deny that the relevantknowledge, once achieved, is a priori.

4 Experimental Philosophy and Apriority4.1 Experimental philosophy as rational improvement?

The case of rational training shows there are possible cases where subjects couldn’tknow that p without having some particular experiences first, even though p is a priori.This can happen when those experiences serve merely to decrease one’s rationallimitations, thereby making them better able to capitalize on a priori warrant. Canwe understand the experimental work we suppose to be needed for philosophicalknowledge in this way? To do so would be to suggest that engaging in experimentalwork is necessary for philosophical knowledge because it is the means we need to use inorder sufficiently to develop our a priori capacities. This would be in effect to treat therelevant sorts of experiments as providing something roughly akin to the trainingstudents receive in logic classes.

The prospects for reconciling X-Phi with Apriority in this way do not seemespecially strong. For one thing, the empirical facts to be discovered—the fact thatepistemic intuitions vary according to cultural background, or that the messiness of adesk influences moral judgments—are deeply contingent. We could have been beingswith more universal epistemic intuitions, or who processed moral judgments in a wayindependent from environmental features; the fact that we turn out not to be issomething that could not, even in principle, be discovered a priori. By contrast,when I undergo the experiences that constitute my training in logic or mathematics,although it is of course contingent that I have the particular sorts of experiences I have,the new facts which I come to appreciate—that, for example, ~(A v B) is equivalent to~A & ~B—are necessary. And even if I could not have come to appreciate this fact


without having something like the experiences that I had—even if I needed to have theexperiences of being taught logic in order to come to appreciate these truths—this is soonly because of my rational limitations. In principle, someone clever enough couldcome to this sort of knowledge without having to be taught it. At least typically, whenexperience is necessary to enable a subject to capitalize on genuinely a priori warrant, itis at least in principle possible for a sufficiently rational subject to capitalize on itwithout that experience. But this is not the case with the relevant experimental results.Even a perfectly ideal rational agent with no cognitive limitations couldn’t know,without having the appropriate experiences, that epistemic intuitions vary culturally; soif, as X-Phi says, such knowledge is necessary for philosophical knowledge, it cannot beso merely by being necessary for rational improvement.

4.2 Defeaters via evidence of rational limitation

According to Apriority, there is a priori warrant for many philosophical claims.According to X-Phi, we can have no knowledge of these claims without experimentalexperiences—on the destruction form of the experimentalist critique, this is becausewe cannot form justified beliefs on the basis of armchair methods. (We shall return tothe discovery form shortly.) So the conjunction of these views entails that the absenceof the relevant experiences defeats the transition from a priori propositional justifica-tion to doxastic justification. In }3, I suggested that there are only two ways for a prioriwarrant so to be defeated: via limitations in a subject’s rational capacities, and viaevidence that such limitations obtain. In the previous subsection, I rejected the ideathat the relevant experiences are necessary because they are necessary for the achieve-ment of sufficiently high rational capacities. We are left, then, with only one choice formodeling the destruction version of experimentalist pressure in a way consistent withApriority. If we do not engage in experimental investigation, finding sufficientlypromising results, and have the experiences that such investigation would involve,then our a priori warrant in philosophical propositions can be defeated by evidence thatwe are relevantly cognitively limited.What does this sort of defeat look like? Here again is the example given in Ichikawa

and Jarvis (2012):

Evidence to the effect that Jane has taken a pill that inhibits rational capacities can defeat Jane’srational commitment to infer in accordance with modus ponens as can evidence to the effect thatJane is crazy even if in fact Jane has not taken such a pill and is not crazy.

This sort of effect occurs in more mundane contexts, too. Suppose I am evaluating twophilosophers—in order, perhaps, to decide which one should be hired for a particularjob. We can suppose that it’s not a trivial decision—each has different strengths, andneither is far better than the other—but that in fact, Emily is better than Lauren, andI have an evidential base that determinately establishes as much. This evidential base,plausibly, includes familiarity with their respective writings, research interests, motiv-ation, teaching skills, administrative abilities, etc. It provides propositional justification


for me for the proposition that Emily is better than Lauren. All things equal, it wouldsuffice for knowledge. But now let us also suppose this complication: Emily is a dearfriend, and I want very much, for reasons external to evaluating each as a philosopher,for her to get the job. Under such circumstances, it may well be that my warrant for therelevant proposition is defeated, and I won’t be able to know which is better. This isbecause I, like most people, am susceptible to biases favoring people I like andoutcomes I desire. I have a natural tendency to emphasize the positive features anddownplay the negative ones, when it comes to people that I like. And I know thisabout myself. Under the circumstances, then, it may not be reasonable for me to trustmyself to this tricky judgment; it would be too easy for me to think Emily the bettercandidate, even if Lauren really were preferable. This psychological data concerningmy unreliability undermines my evidential warrant for Emily’s preferability. Note alsothat it is inessential that I am actually susceptible to this widespread bias; if I know thatalmost everybody is susceptible to this bias, but I happen to be exceptionally immuneto it, I still have strong evidence that I am susceptible, and this evidence still defeats mywarrant. Of course, if I engaged in some careful self-experimentation and came to knowthat I was in this way exceptional, that would be a different story; such could put me ina position to capitalize on the a priori warrant after all. So there are two roles forpsychological investigation that bear on justified belief with respect to the non-psychological question of identifying the better philosopher: it can provide evidenceof rational limitations, thus preventing the use of propositional justification, and it canprovide evidence restoring reasonable confidence in one’s rational abilities, thusdefeating the defeater and licensing the use of the non-psychological warrant.

As in the discussion of }1, there is a version of this worry that attaches to knowledgeeven among naıve subjects, without affecting justification—this will be more like the‘discovery’ interpretation of the experimentalist critique. Consider the case of Emilyand Lauren again, but suppose now that I am (blamelessly) naıve about my cognitivelimitations. I have no evidence that suggests that people tend to be biased towardshigher evaluation of their friends. (Presumably, I’ve lived a very sheltered life.) If, as wesuppose, I have had no experiences as of the relevant psychological data that couldundermine my justification, it obviously cannot be that these experiences defeat myjustification. Nevertheless, since it is a fact that these biases are operative, it is not at allunreasonable to suppose that my naıve belief that Emily is the better philosophercannot amount to knowledge, even if it retains its justification. With respect to this sortof defeater, empirical investigation would reveal that we mightn’t have had theknowledge we thought we did all along.

That Emily is a better philosopher than Lauren is, of course, not a priori. The pointof the example is rather to illustrate how experience can be relevant to establishing andundermining defeat in the form of evidence of rational failure. But the phenomenonis familiar in a priori reasoning as well. Suppose I wonder whether a certain sentence isa theorem of classical logic; after some moderate effort, I produce a proof of it, andcome thereby to believe that it is indeed a theorem. My belief may not amount to


knowledge, for it may be that, even if I have properly produced the relevant proof, it isunreasonable for me to rest confident on having done so. For, like the rest of us, I havestrong empirical evidence that I sometimes engage in fallacious reasoning; sometimes,I think I’ve produced a proof, when in reality, I haven’t. It is too risky for me simply toassume that I’ve been operating with sufficient rational capacities; it would be too easyfor me to have gone wrong. In order to come to knowledge of the proof ’s conclusion,therefore, I may need to double-check it. The sort of double-checking that can make itreasonable for me to be confident in my abilities involves, of course, experiences—thevisual experiences involved in examining the marks on my page or blackboard, forexample, that provide a record of my performance. These experiences are what it takesin order for me legitimately to be confident in my logical performance. Or, if it ispossible to imagine it, even if I have no reason to suppose I’m relevantly fallible, onemight well say that the fact of my fallibility means that I cannot have knowledgewithout double-checking.Here’s another example: suppose I perform a moderately complex mathematical

calculation in my head; I judge, say, that 26 x 31 is 806. We’ll stipulate that I came tothis answer competently in a way that, under ordinary circumstances, would lead me toknowledge that it is correct. But I go on to encounter an apparent epistemic peer whotells me that he has done the calculation in his head, and come to a different answer:628. We can suppose that my rational confidence in my own mathematical ability iscomparable with that in his, such that it would be unreasonably dogmatic simply todiscount his answer on the basis of its conflict with the correct one; since it would beunreasonable, at this moment, to continue to believe that the answer is 806, I no longerknow that fact. I restore my knowledge by having some experiences: the experiencesinvolved in, say, talking over how we came to the answer, or checking to see whetherhe’s any good at arithmetic, or using a calculator. My initial belief was a prioriknowledge; it was defeated by evidence of rational limitations, and that defeater wasovercome by experience. My experience restored my a priori knowledge.I don’t know of any simple way to explain in generality under what circumstances

this kind of experiential validation is necessary for reasonable belief. I certainly don’tthink that it’s always necessary—we don’t always need to double-check, and sometimeswe can ignore evidence that is misleading; the point is merely that sometimes suchexperiences are necessary to restore reasonable confidence. And that when they are, theexperiences involved are not properly thought of as part of the warrant—as part of theexplanation for the propositional justification in question. Instead, they are merelyenabling. They do not enable concept possession, but they enable subjects to capitalizeon their extant justification by defeating evidence that would otherwise be suggestiveof disruptive cognitive limitations. They are therefore consistent with Apriority.

4.3 X-Phi and reasonable self-confidence

The model of the previous subsection is the one to which we should apply thecase of experimental pressure against armchair philosophy; indeed, it matches the


interpretation of the experimentalist critique I argued for in }1. According to thedestruction version of the experimentalist pressure, we now have substantial empiricalevidence impugning our rational capacities, such that, to continue to rely on themunreflectively would be epistemically irresponsible. And according to the discoveryversion, even a naıve subject who lacks this evidence might, the evidence suggests, failto know, because he is likely to be making mistakes. Learning that some cultural groupsjudge Gettier cases to be cases of knowledge impugns our rational abilities, since thisshows that either we or they are wrong. Without some reason to privilege our ownjudgments over theirs, it would be unreasonably dogmatic for those of us who knowabout the data to proceed on the assumption that it is we, and not they, who are right; andthe fact of the disagreement defeats knowledge of even the naıve who aren’t aware of it.I should emphasize that I do not mean to be endorsing this line. This is how I think

the experimentalist pressure ought to be understood. Obviously, there are manypossible responses one could make at this stage in order to resist X-Phi. One couldargue that we ought to interpret the relevant subjects as engaging with differentscenarios, or using different concepts, and so do not disagree at all; or one couldsuggest that our judgments are more likely to be right, given our expertise, or ourinterested engagement. There is much one can say here, and most of it has been said.Again, my point in this chapter is that even if X-Phi is right, the experience necessary forthe relevant philosophical knowledge is consistent with Apriority.

Other experimentalist challenges are also easily thought of in this way. We learn thatorder effects influence our epistemic intuitions; this shows that epistemic intuitions areeasily influenced and so not particularly reliable; this should undermine our confidencethat our judgments are not similarly unreliable. We could restore our initial reasonableconfidence if we engaged in experimental investigation that showed that we philoso-phers are not similarly susceptible to order effects, or that our arguments did not rely onthe kinds of intuitions that are susceptible to order effects.

Of course, in saying that we could restore rational confidence and knowledge in thissort of way, I assume that our initial judgments were indeed correct and a priori.I therefore assume that the experimentalist challenge can ultimately be met. Butnothing much hangs on any particular case; if I am wrong about the Gettier intu-ition—if our ultimate investigation indicates that everyone, including philosophers, isunreliable about knowledge attributions in a way that undermines any claim to knowGettier cases to be non-knowledge, that shows that this particular case was not knowna priori, since it was not known. We evaluate judgments on a case-by-case basis.

There is much room to consider and debate how far the skeptical pressure unearthedby the experimentalists extends. A conservative interpretation would have it that onlyparticular patterns of rational failure need be investigated; one need only check to makesure there aren’t order effects operative to restore legitimate confidence. A more radicalline would suggest that the presence of order effects raises much broader worries: we’vediscovered that order effects influence intuitions; who knows what other epistemicdangers lurk? On the radical line, we’d need to do much more checking for potential


sources of error in order to be confident in our rational abilities. Messy desks influencemoral judgments; what other disruptive factors might be around? Maybe the smell ofcoffee disrupts one’s ontological intuitions? We can’t tell unless we do some experi-ments and check! The radical line demands quite a lot of experimental experience forphilosophical knowledge; even still, the knowledge in question is a priori. The truth issurely somewhere between the conservative and radical lines here, but these questionsare simply orthogonal to the apriority of philosophy.

5 ConclusionWhether some knowledge is a priori depends on the source of its warrant. WhatI have demonstrated is that this question is separable from the question whether it isachievable from the armchair; that indeed, the questions may in principle come apartrather dramatically. Even though I have not attempted to establish Apriority or tojudge definitively on X-Phi, exploring the connection—or, as I say, independence—between the two has, I hope, clarified both notions.One takeaway moral is that epistemological questions about the nature and structure

of warrant must be distinguished from methodological questions about the appropriateengagement with and processing of evidence. The methodological questions areepistemological ones too: they bear, as I have emphasized, on what is known. Butthey do so in a way importantly distinct from questions about the source of warrant,and so from questions about apriority.26

Notes1. Ichikawa and Jarvis (2009) argue that many thought-experiment judgments are a priori;

Ichikawa and Jarvis (2012) argue that much modal knowledge is known a priori. These ideasare developed much more thoroughly in Ichikawa and Jarvis (forthcoming), which alsodevelops many of the ideas in the present chapter.

2. See my (2011) for my case against X-Phi.3. See, for one influential example, Knobe (2006).4. Weinberg et al. (2001), p. 429.5. Machery et al. (2004), pp. B8 and B9.6. Machery et al. (2004), p. B8.7. Ichikawa (forthcoming) argues that one prominent strain of experimentalist pressure is best

understood in this way, but distinguishes it from the more sophisticated critique that is thefocus of this chapter.

8. See e.g. Deutsch (2009), ch. 12 of Ichikawa and Jarvis (forthcoming), or ch. 7 of Williamson(2007).

9. Given my stronger claim that X-Phi and Apriority are consistent in the sense that they might betrue, I do however carry the commitment that the relevant philosophical claims do notdepend on epistemic access to facts about the distribution of intuitions.


10. A third potential reason to worry about this interpretation of the experimentalist critiquewould charge that the very notion of an ‘intuition’ is too unclear to play the central role hereafforded; Cappelen (2012) presses something like this charge. (Thanks to a referee forencouraging discussion of this issue.) While I agree with Cappelen that in many instances,discussion of ‘intuition’ in philosophy is culpably sloppy, for present purposes, I’m happy toassume that intuitions are whatever my dialectical opponents want them to be; I assume onlythat they are or involve some kind of psychological state that can be detected by empiricalmethods. I will speak of intuitions in this chapter from time to time, in order to maintaincontact with experimentalist critics who also speak of intuitions; my own commitments areneutral on what intuitions are.

11. E.g. Hales (2006) p. 171, Ludwig (2007), Kauppinen (2007).12. E.g. Elga (2007).13. See Ichikawa (2010).14. Knobe (2006).15. Thanks to Dustin Locke for the apt locution.16. As TimothyWilliamson (2007, p. 165) writes: “Standard discussions of the a priori distinguish

between two roles that experience plays in cognition, one evidential, one enabling. Experi-ence is held to play an evidential role in my visual knowledge that this shirt is green, but amerely enabling role in my knowledge that all green things are coloured: I needed it only toacquire the concepts green and coloured, without which I could not even raise the questionwhether all green things are coloured.”Williamson goes on to critique this orthodoxy.

17. Ball (2009).18. Of course, a Fodorian nativist will here dissent. Fodor (1998). Concept nativism is obviously

inconsistent with there being experience that is both consistent with apriority and necessaryfor concept acquisition. But it is consistent with the more general story I will go on to tellabout non-warranting experience. As I will emphasize in this section and the next, not muchultimately hangs on concepts.

19. Cf. Burge (1993).20. These issues and the subsequent ones derive from Ichikawa and Jarvis (2012 and forthcom-

ing).21. Williamson (2007), ch. 4, makes much of this fact, suggesting that this rules out any

interesting notion of apriority along the lines of conceptual necessity. But it is possible tothink that there are conceptual truths without thinking that they must be appreciated by allcompetent speakers; I will give an example of such a view later.

22. Giaquinto (1998) p. 200.23. Williamson (2007), ch. 5.24. Williamson (2007), p. 167.25. Our terminology then was somewhat different; we wrote of conceptual entailment, not

apriority; our paper did not attempt to characterize apriority precisely.26. Many of the ideas in this chapter owe their inspiration to recent and forthcoming work with

Benjamin Jarvis; I’m grateful to him for inspiring my thinking in this area, and for originatingmany of the ideas that form the framework for this project. I’m grateful also for invaluablediscussions about this and related material to Derek Ball, Yuri Cath, Joachim Horvath, DustinLocke, Kaija Mortensen, Carrie Ichikawa Jenkins, and Jonathan Weinberg. Versions of thischapterwere presented at an EmmyNoether Armchair Lab inCologne in 2010, theUniversity


of British Columbia in 2011, and a Basic Knowledge seminar at the Northern Institute ofPhilosophy in Aberdeen in 2012; thanks to all these audiences for helpful discussion.

ReferencesAlexander, J. and Weinberg, J. (2007), Analytic Epistemology and Experimental Philosophy,Philosophy Compass 2(1): 56–80.

Alexander, J., Mallon, R., and Weinberg, J. (2010), Accentuate the Negative, Review of Philoso-phy and Psychology 1(2): 297–314.

Ball, D. (2009), There Are No Phenomenal Concepts, Mind 118(472): 935.Burge, T. (1993), Content Preservation, The Philosophical Review 102(4): 457–88.Cappelen, H. (2012), Philosophy Without Intuitions, Oxford: Oxford University Press.Casullo, A. (2003), A Priori Justification, Oxford: Oxford University Press.Deutsch, M. (2009), Experimental Philosophy and the Theory of Reference,Mind and Language

24(4): 445–66.Elga, A. (2007), Reflection and Disagreement. Noûs 41(3): 478–502.Fodor, J. (1998), Concepts: Where Cognitive Science Went Wrong, Oxford: Oxford University Press.Giaquinto, M. (1998), Epistemology of the Obvious: A Geometrical Case, Philosophical Studies,92(1): 181–204.

Greene, J., Sommerville, R., Nystrom, L., Darley, J., and Cohen, J. (2001), An fMRI Investi-gation of Emotional Engagement in Moral Judgment, Science 293: 2105–8.

Hales, S. D. (2006), Relativism and the Foundations of Philosophy, Cambridge, MA: MIT Press.Ichikawa, J. J. (2010), Explaining Away Intuitions, Studia Estonica Philosphica.Ichikawa, J. J. (2011), Experimentalist Pressure Against Traditional Methodology. PhilosophicalPsychology, 28 October 2011, iFirst.

Ichikawa, J. J. (forthcoming), Who Needs Intuitions? Two Experimentalist Critiques, Forth-coming in Booth, A. and Rowbottom, D. (eds), Intuitions, Oxford University Press.

Ichikawa, J. J. and Jarvis, B. (2009), Thought-Experiment Intuitions and Truth in Fiction,Philosophical Studies, 142 (2): 221–46.

Ichikawa, J. J. and Jarvis, B. (2012), Rational Imagination and Modal Knowledge, Noûs, 46(1),February 2012: 127–58.

Ichikawa, J. J. and Jarvis, B. (forthcoming),TheRules of Thought, Oxford:OxfordUniversity Press.Kauppinen, A. (2007), The Rise and Fall of Experimental Philosophy, Philosophical Explorations10(2): 95–118.

Kitcher, P. (1983), The Nature of Mathematical Knowledge, New York: Oxford University Press.Knobe, J. (2006), The Concept of Intentional Action: A Case Study in the Uses of FolkPsychology, Philosophical Studies 130(2): 203–31.

Knobe, J. and Nichols, S. (2008), An Experimental Philosophy Manifesto. In Knobe, J. andNichols, S. (eds), Experimental Philosophy, New York: Oxford University Press: 3–16.

Ludwig, K. (2007), The Epistemology of Thought Experiments: First Person versus ThirdPerson Approaches, Midwest Studies in Philosophy 31: 128–59.

Machery, E., Mallon, R., Nichols, S., and Stich, S. (2004), Semantics, Cross-Cultural Style,Cognition 92: B1–B12.

Mallon, R., Machery, E., Nichols, S., and Stich, S. (2009), Against Arguments From Reference,Philosophy and Phenomenological Research 79(2): 332–56.


Sosa, E. (2007), Experimental Philosophy and Philosophical Intuition, Philosophical Studies 132(1): 99–107.

Swain, S., Alexander, J., and Weinberg, J. (2008), The Instability of Philosophical Intuitions:Running Hot and Cold on Truetemp, Philosophy and Phenomenological Research 76(1): 138–55.

Weinberg J., Nichols, S., and Stich, S. (2001), Normativity and Epistemic Intuitions, PhilosophicalTopics 29: 429–60.

Williamson, T. (2007), The Philosophy of Philosophy, Malden, MA: Blackwell.


3

The Implicit Conception andIntuition Theory of the A Priori,with Implications for ExperimentalPhilosophy

Joshua C. Thurow

1 IntroductionOn the face of it, much of what philosophers do is a priori. We don’t know there is abarn in front of us when we’re in fake barn country . . . we could be the victims of anevil demon determined to deceive us . . . it is morally permissible to separate ourselvesfrom the ailing violinist in Thomson’s violinist scenario . . . ‘q’ really does follow from‘p’ and ‘if p then q’ . . . we would still exist after a successful brain transplant, and wewould go with our brain. Each of these claims are often justified in philosophicaldiscourse by its seeming to us on reflection to be true, or, in other words, by our havingan intuition that it is true. Furthermore, the justification that we get via these seemingsor intuitions doesn’t seem to depend on any empirical evidence, and so is a priori in thesense of “independent of reliance on empirical evidence.”The use of a priori ways of knowing or being justified (hereafter ‘the a priori,’ for

short) is distinctive, but not idiosyncratic, of philosophy. Judgments based on thoughtexperiments are used in physics, everyday folk make apparently a priori mathematicaljudgments all the time, and we all make logical inferences regularly.Despite its prominence in philosophical methodology and presence in other

common discourses, the existence and importance of the a priori has been repeatedlychallenged. Michael Devitt expresses one of these challenges when he writes, “at thispoint, it remains a mystery what it would be for something to be known a priori”(2005: 114). We have some inkling of how perception gives us knowledge of theworld, and how memory preserves that knowledge, but it is hard to see how reflectionon our concepts alone can give us such knowledge. This kind of challenge was firstdeveloped by Paul Benacerraf (1973) against the existence of mathematical knowledge

(assuming mathematical realism); Devitt (2005), Field (2005), and Thurow (forthcoming)have each argued in somewhat different ways that lacking a good theory of how a prioribeliefs can be reliable provides at least a prima facie reason to deny the existence of a priorijustification and knowledge.

A more recent challenge comes from experimental philosophy. Several studies haveshown that folk intuitions about a variety of philosophical scenarios (e.g. Gettier cases,Lehrer’s Truetemp case) vary depending upon socio-economic status and culturalbackground. Nadelhoffer and Nahmias (2007) and Alexander and Weinberg (2007)describe the ‘restrictionist’ reaction to these studies: since intuitions about cases likethese vary too widely on the basis of philosophically irrelevant factors, such intuitionscannot justify our beliefs. If restrictionism is the correct reaction, it would not followthat we should never rely on a priori methods (see Weinberg 2007), but it wouldfollow that reliance on the a priori amongst philosophers ought to be severely restrictedcompared to present practices.

In this chapter I aim to respond to both of these challenges by developing a theory—the Implicit Conception and Intuition (ICI) Theory—of how a priori beliefs can bereliably produced. This theory explains how there can be a reliable connectionbetween intuitions and implicit conceptions for relevant concept, and also a reliableconnection between one’s implicit conceptions and truths involving the relevantconcepts. I will develop this theory through critical interaction with ChristopherPeacocke’s work on the a priori, as well as Timothy Williamson’s recent critiques ofthe notion of ‘conceptual truth.’ The theory that I develop has several virtues: it is 1)consistent with the notion that a priori justification is fallible, 2) incorporable into bothinternalist and externalist theories of justification and knowledge, and 3) consistentwith internalism and externalism about content. Furthermore, I argue that according tomy theory, an extreme restrictionist view of the epistemic value of the a priori is notwarranted, but that experimental philosophy can in principle be useful for 1) develop-ing a better understanding of many concepts (and their objects) that philosophers study,and 2) identifying ways in which our intuitions can be distorted.

The broader take-home message is this: theoretical advance on the notion of the apriori is a crucial, and neglected, tool for resolving the tensions between a priorimethodology and experimental philosophy.

My main aim is to develop a plausible theory and examine some of its implications,particularly for experimental philosophy. Accordingly, I will not argue directly forevery aspect of my theory. I will motivate many components of the theory and defendthe theory against various objections. Also, I build on several ideas that others havedefended elsewhere. However, none of this will constitute a definitive case for mytheory. I simply hope to show that the theory is plausible—indeed, plausible enough tomeet the Devitt and Benacerraf-like worries about the a priori mentioned earlier (seeThurow (forthcoming) for discussion of the conditions a theory must meet to countersuch worries). Furthermore, I hope the theory can show how the a priori character ofphilosophical methodology can exist alongside work in experimental philosophy.

68 JOSHUA C. THUROW

2 Christopher Peacocke’s Theory of the A PrioriI begin with an examination of Christopher Peacocke’s account of a priori knowledge.Although I believe his account is flawed, at its heart is a plausible idea that I willincorporate into my theory of how a priori beliefs are reliably formed.In a series of papers and a recent book, Christopher Peacocke has developed an

account of how we can have a priori knowledge of a variety of propositions and of howwe can be a priori justified or entitled in making transitions in thought from one mentalstate to another where the former is to ground or warrant the latter. His theory alsoexplains how a priori beliefs are reliably formed. According to his theory, the way thatthe contents of concepts are determined explains how our a priori judgments arereliable. His account goes as follows:

An outright, non-defeasible, way of coming to know p is an a priori way if the possession-conditions for the concepts in p together with the Determination Theory jointly guarantee thatuse of that way leads to a true belief about whether p is the case. Similarly, a transition from oneset of contents to a given content is an a priori transition if the possession-conditions for thecontents involved together with the Determination Theory jointly guarantee that the transitionis truth-preserving. (Peacocke 2004: 172)

According to Peacocke’s Determination Theory, “the semantic value of a concept isthat entity of the appropriate category which makes the transitions mentioned in itspossession-condition always truth-preserving,1 and the outright principles so men-tioned always true” (2004: 171). To see exactly how this works, let’s apply it to acouple of examples. First, how does this account—Peacocke calls it the metasemanticaccount—explain howwe are a priori justified in making the transition from “A and B”to A?Well, one of the possession-conditions for CONJUNCTION is that one find theinference from “A and B” to A primitively compelling, and the Determination Theorysays that semantic values are assigned to make all transitions mentioned in the posses-sion-conditions truth-preserving. As long as there is a semantic value that will make allthese transitions truth-preserving—and in this case there is one, namely conjunction—the possession-conditions for CONJUNCTION together with the DeterminationTheory guarantee that the transition from “A and B” to A is truth-preserving, and soaccording to the second sentence in the above quote, this transition is “a priori,” that is, apriori justified and known.2

Here’s a second simple example. It is plausible to suppose that one of the possessionconditions for the concept FATHER is that one find “a father is male” primitivelycompelling. According to the determination theory, the content of FATHER must bean entity that makes all principles mentioned in the possession conditions true, so “afather is male” is thereby guaranteed to be true. Hence, finding “a father is male”primitively compelling is an a priori way of coming to know that a father is male. Again,this is because the possession conditions for FATHER, together with the determinationtheory, guarantee that “a father is male” is true.

THE IMPLICIT CONCEPTION AND INTUITION THEORY 69

Peacocke raises the following objection to his own account: there are propositionsthat we seem to know a priori, although we cannot explain how we know them apriori using the above theory. Many of the traditional counterexamples to Ayer’slinguistic theory are also counterexamples to the above theory, e.g. 1 + 1 = 2, shades of red is not green, nothing red all over is also green all over. To deal with theseexamples, Peacocke introduces the notion of an implicit conception. An implicitconception of F is a mental state whose content is some set of conditions that mustbe satisfied for something to be F. One need not be able to formulate one’s implicitconceptions; we, in fact typically cannot do so and when we try we often get it wrong.One’s implicit conception of F is also supposed to explain how one applies the conceptF. Peacocke gives a couple of helpful examples. He first considers the concept of wholenumber. He suggests that, underlying the ordinary person’s possession of the conceptWHOLE NUMBER, is an implicit conception with the following content:

(N1) 0 is a whole number;(N2) the successor of a whole number is a whole number;(N3) only what is determined to be a whole number on the basis of the preceding

two conditions is a whole number.

With this implicit conception, it follows that 1 is a whole number, since 1 is thesuccessor of a whole number, namely 0. Furthermore, I can come to know that 1 is awhole number a priori, according to Peacocke, because the truth of “1 is a wholenumber” is guaranteed by the possession conditions, now taken to include implicitconceptions,3 of WHOLE NUMBER together with the determination theory. Theway of belief that guarantees the truth of “1 is a whole number” is by being, as Peacockewrites, “suitably influenced” by the possession conditions in making the judgment.4

He also considers the concept red. Underlying RED, he claims, is an implicitconception that designates a range of shades as red, perhaps another range as borderlinered, and everything else as non-red. Similar implicit conceptions underlie other colorconcepts. As a consequence, our implicit conception will dispose us to believe that acertain shade is red, or is not red, or is borderline red, depending on which of the threecases the shade falls under, as determined by the implicit conception.Wewill also be ableto tell a priori that “nothing that is a shade of red is a shade of green” is true by comparingour implicit conceptions and noting that no shades fall under both red and green.5

There are three theoretical advantages of believing in implicit conceptions and takingthem to be possession conditions for many concepts. The first advantage, which wasalready briefly mentioned, is that implicit conceptions are supposed to influence ourjudgments in which we apply F in various scenarios, both real and imagined (simulated,as Peacocke puts it). So, implicit conceptions explain why we make these judgments.Second, they can account for our a priori knowledge of principles involving a conceptthat do not follow deductively from other principles. Third, having an implicit concep-tion is a more satisfactory account of possession conditions for a variety of concepts thanis having an acceptance condition. Acceptance conditions require one to accept certain

70 JOSHUA C. THUROW

principles or inferences involving a concept. Peacocke’s account of the possessionconditions for CONJUNCTION appeal to acceptance conditions because to possessthe concept one is required to make certain kinds of inferences with the concept.Peacocke recognizes that many concepts do not have acceptance conditions as posses-sion conditions. For example, there are no specific inferences we must find primitivelycompelling in order to possess the concepts OR or IF, THEN.6 The possessionconditions for these concepts, then, must be given in terms other than acceptanceconditions. Implicit conceptions provide another kind of possession condition thatobviates any requirement that to possess the concept one must find certain principlesor inferences compelling. One can possess a concept and be quite mistaken about howto apply the concept, as long as one possesses the implicit conception for that concept.Here’s another way of understanding Peacocke’s theory. Peacocke proposes a link

between understanding (i.e. possessing a concept) and having implicit conceptions anda link between implicit conceptions and truth.7 Put together, we then have anunderstanding/truth link. To possess a concept, one must have an implicit conceptionthat is guaranteed to be true, given how the implicit conception determines thecontent of the concept. So, one’s implicit conceptions are guaranteed to be reliable.Peacocke’s theory offers what Alvin Goldman (2007: 14) calls a constitutive approach tothe relationship between concepts and reliability: i.e. reliability follows from the natureof concepts and concept possession. It is part of the nature of concepts that one’simplicit conception for a concept is accurate, and so one’s beliefs produced by thatimplicit conception are reliably produced.

3 Problems for Peacocke’s TheoryAlthough I believe that Peacocke is on to something, his theory faces a few challenges.Here I will discuss two.8 I will use these objections to develop my own theory of how apriori beliefs can be reliable that builds off of Peacocke’s core ideas.

3.1 Content-determining conditions and implicit conceptions

Peacocke’s model relies crucially on the idea that in order to possess a concept one mustmeet certain conditions and that something about those conditions determines thecontent (or semantic value—I’ll use the two phrases interchangeably) of the concept.According to Peacocke, depending upon the concept in question, the content-determining conditions could be either acceptance conditions or implicit conceptions.There are two kinds of acceptance conditions: 1) some principle about the concept (e.g.a necessary and/or sufficient condition for the application of the concept) that one isdisposed to find primitively compelling; 2) a transition involving the concept that one isprimitively disposed to make. An implicit conception is a set of principles about theconcept that one implicitly accepts, but need not, and indeed typically does not,explicitly accept or endorse. Furthermore, the acceptance conditions and implicit

THE IMPLICIT CONCEPTION AND INTUIT ION THEORY 71

conceptions for a concept are content-determining and one must have them in order topossess the concept.

But, there are many claims that we would accept—explicitly or implicitly—that donot determine the content of any concepts. Suppose I fear spiders, and I believe thatI fear spiders. Nonetheless, the content of SPIDER is not determined to be somethingthat I fear. I believe that there are more than four books in my office, but that beliefdoes not determine the content of the concept BOOK. Plainly much of what webelieve that involves a given concept does not determine the content of that concept.Let us say that anything that we accept—implicitly or explicitly—and any inferencethat we are disposed to make that involves a concept C, is a part of our explicit andimplicit conception of C. Further, let us say that those parts of our explicit and implicitconception of C that determine the content of C are parts of the content-determiningexplicit and implicit conception (CDEIC, for short) of C. So, here’s the problem: whatmakes a part of our explicit and implicit conception of C content-determining?

I haven’t been able to find a clear answer to this question in Peacocke’s work on thea priori, however I believe that there is an attractive way of answering it. Suppose weexplicitly accept the following principle involving the concept G: All G’s are F’s.Furthermore, suppose this principle is content-determining for the concept of G.Well,then the content of G is determined to be some property of which it is true that all Gsare Fs. If we assume—and it seems plausible to assume—that a concept necessarily picksout its semantic value/content (that is, concepts are necessarily individuated by theirsemantic values, so a concept with a different semantic value would simply be adifferent concept), then it follows that it is necessary (in what Plantinga (1974: 2)calls the broadly logical sense, and Sider (Conee and Sider 2005: 190) calls the absolutesense) that a given G is an F.

Many of the things that we explicitly and implicitly accept that involve a givenconcept are not, and we do not take them to be, necessary. I gave a couple goodexamples earlier. Although I believe that I fear spiders, I don’t take it that it is necessarythat I fear spiders. I believe that there are more than four books in my office, but it isnot (and I don’t take it to be) necessary that there are at least four books in my office.So, the following seems plausible: an explicitly accepted principle P or transition T is anacceptance condition for concept C only if P or T seems necessary to one uponreflection. We have to be a little more nuanced in treating implicit conceptions becauseoften the principles in an implicit conception don’t seem necessary upon reflection—they may be too abstract to elicit such a seeming. However, for one to have an implicitconception, that implicit conception must actually influence one’s use of the concept,and so one will be able to consider many of the propositions that are entailments of theimplicit conceptions that one considers. These entailments will be necessary as well, sowe can say that an implicitly accepted principle P is an implicit conception for conceptC only if P or many of its entailments would seem necessary to one upon reflection.

This way of distinguishing implicit conceptions and acceptance conditions fromother beliefs involving a concept fit well with a claim often made about a priori

72 JOSHUA C. THUROW

knowledge. George Bealer, who holds that a priori knowledge and justification isgained via rational intuition, writes,

In our context when we speak of intuition, we mean ‘rational intuition’ or ‘a priori intuition.’This is distinguished from what physicists call ‘physical intuition.’ We have a physical intuitionthat, when a house is undermined, it will fall. This does not count as a rational intuition, for itdoes not present itself as necessary: it does not seem that a house undermined must fall. (Bealer1999: 30)

Similarly, Laurence BonJour—a prominent defender of rationalism—states, “accordingto rationalism, a priori justification occurs when the mind directly or intuitively sees orgrasps or apprehends (or perhaps merely seems to itself to see or grasp or apprehend) anecessary fact about the nature or structure of reality” (BonJour 1998: 15–16). AlvinPlantinga describes “a peculiar sort of phenomenology” that accompanies seeing a priorithat a proposition is true. This phenomenology consists, in part, in “finding yourselfutterly convinced that this proposition is not only true, but could not have been false”(Plantinga 1993: 105–6).If content-determining explicit and implicit conceptions are distinguished by a

disposition for some propositions to seem necessary, and a seeming of this sort is keyto being justified in believing propositions a priori, then content-determining explicitand implicit conceptions are, at least in principle, well suited to explain how a prioribeliefs are reliable. Such an explanation would dovetail nicely with Bealer, BonJour,and Plantinga’s view that a priori knowledge and justification is had via an intuitionthat a proposition is necessary, or a seeming with a special phenomenology.9

Following Bealer (1996, 1998, 1999), I will call a seeming that p is absolutely (orbroadly logically) necessary a rational intuition. A seeming that p is different from a beliefthat p, as it could seem that p even if you don’t believe that p (as in the Muller-Lyerillusion where the lines seem of different lengths, although once we are familiar withthe illusion we don’t believe they are of different lengths). Rational intuitions aredistinct from perceptual seemings since the latter don’t seem necessary and are basedupon a perceptual experience. Rational intuitions are also distinct from memorialseemings. Memorial seemings (i.e. a seeming that p based upon memory) often don’tseem necessary, but even those that do are different from rational intuitions. Considerthe difference between merely remembering that some complicated logical theorem istrue and necessary and showing that the theorem is true and necessary. In both cases itmight seem that the theorem is necessary, but those seemings are still different. In theformer case one is merely disposed to accept that the theorem is necessary, whereas inthe latter its necessity is more evident, or directly apparent (or so it seems).I want to add two qualifications to my definition of rational intuition. First, one

needn’t have a full philosophical understanding of the concept of absolute necessity inorder to have a rational intuition. Ordinary people have at least a rudimentary grasp ofnecessity via the notions of ‘must,’ and ‘could not.’ Second, a rational intuition needn’thave the concept of necessity as part of its content. Plantinga draws attention to a


certain sort of phenomenology that accompanies a priori justification; I take it that thisis a sort of felt necessity—a “must-y-ness”—much like how some memories areaccompanied with a felt pastness. Must-y seemings that p seem to me to count asrational intuitions that p.

My suggestion, then, is that rational intuitions distinguish principles and inferencesthat are part of (or that follow from) the CDEIC of concepts from those that do not.How exactly do rational intuitions distinguish explicit and implicit conceptions ofC that are not part of the CDEIC for C from those that are? Although this is an excellentquestion, I will not try to answer it here. Let me be clear about my goals. I’m not tryingto give necessary and sufficient conditions for explicit and implicit conceptions of C tobe CDEIC of C; my goal has been more modest. I aim to introduce the notion ofCDEIC and note some of its distinctive characteristics—including how rational intu-itions often distinguish principles that are part of the CDEIC for concepts from thosethat are not—in order to indicate howmuch of our explicit and implicit conception of agiven concept is not content-determining. So, I will conclude this section by highlight-ing three ways that rational intuitions relate to the CDEIC for concepts.

First, I take it that if p is part of the content-determining explicit conception for C,then one must be disposed to have a non-inferential rational intuition that p is necessary.This isn’t to say that one doesn’t also have other good reasons—memory, testimony, oran argument—for believing that p is necessary. One may have a rational intuition that pis necessary and have other good reasons for believing that p is necessary. If p is part ofthe content-determining implicit conception for C, one may not be disposed to have arational intuition that p is necessary, but one will find at least some of the propositionsthat are consequences of C to be intuitively necessary.

Second, one could have a rational intuition that p is necessary even if p is false, andthus not content-determining for some concept. This could happen for a variety ofreasons. One may not pay sufficient attention to p and confuse it for a similar propos-ition that one correctly finds necessary. One might confuse different senses of possibilityand necessity—epistemic, nomic, absolute. Typically, however, through further reflec-tion on the relevant concepts, employing other intuitions that seem clearly true, one cancorrect mistakes of this kind (see Bealer 2004 for further discussion).

Third, the CDEIC for a concept may not, and probably typically does not, amountto a definition of the concept, or an account of the essence of the semantic value of theconcept. The CDEIC may write in gaps that are to be filled and there are a variety ofways such gaps could be filled. Necessary a posteriori truths provide nice examples.The CDEIC forWATERmight be something like this: whatever chemical compoundthat composes (or whatever entity that best explains the surface features of) the stuffactually found in our lakes and rivers is necessarily water. We can’t know what theessence of water is a priori, but the CDEIC points us where to look to find its essence,and once we have done the relevant empirical research, CDEIC together with thatempirical data will show that water is necessarily H2O.10 So, we don’t have a rationalintuition that water is necessarily H2O, but we have rational intuitions of various

74 JOSHUA C. THUROW

principles in the CDEIC for water, which directs us to how to find its essence (namely,find the relevant chemical compound). Brian Weatherson (2003) suggests another waythis could go: the property picked out by a predicate (or a concept) is determined to bethe most simple, natural, property that “validate[s] many of our pre-theoretic intu-itions” (2003: 11). More generally, a CDEIC could dictate that the content of C isdetermined to be the simplest entity that best explains and unifies principles P1 . . . Pn(and/or the validity of a certain set of inferences). In a reply to an essay of Peacocke’s,Tyler Burge briefly suggests yet another way: “it seems to me that it would be a mistaketo think that the implicit mental structures that explain explicit judgments mustthemselves always be complete. Sometimes we depend on others. Sometimes wedepend on a combination of examples, [and] an unconceptualized sense of similarity”(2003: 386). The idea that CDEICs defer to other people or communities will bedeveloped in }6. When a CDEIC for a concept C has gaps of any of these sorts, one cangain a better understanding of C not just by having a more explicit conception of thecontent-determining implicit conception for C, but also by filling in the gaps that theCDEIC points to. Doing this may require empirical work—as with necessary a poster-iori truths—or a priori reflection—such as when we seek the simplest best explanationof a set of principles and cases. A priori theorizing about knowledge and justificationmay be an example of the latter.

3.2 Fallible a priori justification

One of Peacocke’s main goals is to provide an account of what he calls “a priorientitlement,” which means the same as “a priori justification,” as long as we grantthat one can be non-inferentially justified in believing something. What he gives us,though, is an account of how we can have a priori knowledge, not how we can be apriori justified. Perhaps Peacocke means for his theory of a priori knowledge to performboth tasks. He could accomplish this by accepting that knowledge entails justificationand by proposing that his account of when a way is an a priori way of knowing beunderstood as contributing to his theory of a priori knowledge by really being anaccount of when a way is an a priori entitling (or justifying) way. If we understandhim this way, then his account of a priori knowledge will look something like this:

S knows p iff S believes p, p is true, and S believes that p as a result of using an a priorientitling way of belief-formation,

where what it takes to be an a priori entitling way is to be suitably influenced by thepossession conditions of the concepts in p in coming to believe that p.The problem with this account of a priori justification is that it does not allow for us

to be a priori justified in believing a falsehood. This is because Peacocke asserts that thea priori way of knowing/being entitled (now that we have interpreted his theory to bedirectly about entitlement), is guaranteed to yield a true belief about whether p is thecase.11 So, it simply isn’t possible to use the a priori entitling way to arrive at a falsebelief. It isn’t even possible to have an entitled a priori belief that results from


employing an apparent proof that contains a mistake because believing p by using anapparent proof doesn’t guarantee that p is true. But, we can be a priori justified inbelieving a proposition that is false. In addition, Peacocke himself, in the first chapter ofThe Realm of Reason seems sympathetic to the possibility of false a priori justified beliefswhen he writes that, “the [a priori] entitlement may, but need not be, conclusive.”12

So, if my claim that his theory of a priori justification does not allow for fallible a priorijustified beliefs is correct, Peacocke himself would most likely recognize this to be aproblem that needs solving.

In order to address this problem, what we need is an account of what it is to form abelief in an a priori way, an explanation of how that way can lead to false beliefs, and anaccount of how that way produces justification. We will also want the a priori way tobe generally reliable—the main goal of the chapter has been to construct a theory ofhow a priori beliefs can be reliably formed.

4 The ICI (Implicit Conceptions and Intuition)Theory of the A Priori

My Implicit Conceptions and Intuitions (ICI) theory of the a priori builds off of theconnection (discussed in }3.1) between content-determining explicit and implicitconceptions for a concept C and dispositions to have rational intuitions concerningC. I will understand ‘intuition’ in the sense of rational intuition defined in }3.1—aseeming that p where p presents itself as necessary. ‘CDEIC’, again, stands for thecontent-determining explicit and implicit conceptions of a concept.

(A) S forms a belief that p in an a priori way (i.e. has an a priori belief) iff S believesthat p on the basis of an intuition that p

(B) S makes a transition t in an a priori way iff S’s making t is intuitive to S(C) The explanation for why a priori beliefs are likely to be true and a priori

transitions are likely to be truth-preserving is the following: for S to possess aconcept C, both S’s a priori beliefs involving C and the transitions S makes thatinvolve C must, most of the time, be suitably influenced by the CDEIC for C,where, when S is suitably influenced by the CDEIC in having an a priori beliefthat p or making a transition t, the CDEIC, together with the determinationtheory, guarantee that p is true, or that t is truth-conducive.

So, possessing a concept guarantees that most of one’s a priori beliefs concerning C aretrue and most of one’s transitions involving C are truth-conducive. This explains how apriori beliefs and transitions are reliable. But, the theory entails that a priori beliefs canbe false because not all a priori beliefs concerning a concept are guaranteed to be true.Some may very well be false if they are not suitably influenced by the CDEIC—perhaps they result from inadequate reflection, a bad theory, or conceptual confusion.So, as long as we add that some of these false a priori beliefs can be justified, this theory

76 JOSHUA C. THUROW

is consistent with the possibility of fallible a priori justified beliefs. In addition, (C)would explain why, traditionally, the a priori is thought to result from understanding;one understands a concept only if one possesses it and its CDEIC typically appropri-ately influence one’s a priori beliefs (and one can have a greater or lesser understandingthe more or less one’s a priori beliefs match the CDEIC).Conditions (A) and (B) conflict with Peacocke’s “umbrella characterization of the a

priori:”

A content is a priori if a thinker can be entitled to accept it without that entitlementbeing constitutively dependent upon the content or kind of her perceptual experi-ences or other conscious states.13

This characterization explicitly denies that a priori entitlement, or justification, can atall depend upon one’s conscious states, and so rules out any account of a priorijustification on which one’s being justified depends on having an intuition—a modelconscious state. I can’t see, however, why Peacocke feels bound by his “umbrellacharacterization.” There are two traditional conceptions of the a priori: a negativeconception—a priori justification is justification independent of experience, where‘experience’ is understood to refer to sensory experience—and a positive conception—a priori justification is justification by rational insight or understanding alone.14 Neitherof these traditional conceptions of the a priori is committed to Peacocke’s umbrellacharacterization; for all they say, there could be some kind of experience that plays arole in grounding justification, or being evidence for a priori beliefs, that is non-sensoryand/or that is connected to understanding in some important way. Indeed, I suggestthat intuition is just this kind of experience.The theory embodied in (A)–(C) obviously does not constitute a theory of a priori

justification or knowledge. It merely gives an account of what a priori beliefs are andhow they can be reliable. I think that one could build a variety of theories of a priorijustification and knowledge, externalist and internalist, around (A)–(C). This is avirtue of the theory because it renders the theory immune to the following kindof objection: “your theory presupposes internalism, but internalism is false (forsuch and such reasons),” or, “your theory presupposes externalism, which is false(for such and such reasons)”. My theory can be utilized by both the externalist andthe internalist for explaining how a priori beliefs can be reliable, thus enabling bothkinds of theorists to avoid concerns about how to explain the reliability of a prioribeliefs.To illustrate, an internalist could simply endorse the following epistemic principle in

addition to (A)–(C):

(Int-APJ): Necessarily, if S has an intuition that p, then S is prima facie justified inbelieving that p.15

So, on this kind of view, one’s having an intuition justifies one in believing that p. Thisview is internalist (more precisely, mentalist) because the grounds of justification are


internal to the agent.16 An accessibilist internalist could, not too implausibly, add thatthe grounds are accessible to the agent.17 Since beliefs that are justified in virtue ofsatisfying (Int-APJ) are formed in an a priori way, these beliefs are prima facie a priorijustified. Beliefs that are justified in this way are a priori justified in both the negativeand positive senses of the a priori mentioned earlier. They are justified nonexperien-tially because the grounds cited in the antecedent of (Int-APJ) are nonexperiential andthey are justified by understanding because, via (C), having intuitions is part of what itis to understand the concepts in question.

Furthermore, if both (C) and (Int-APJ) were true, there could be a priori justifiedfalse beliefs. Also, since (Int-APJ) concerns prima facie justification, one’s a priorijustification could be defeated in principle by other a priori justified beliefs or byexperientially justified beliefs.

Here’s a sample externalist theory of a priori justification around (A)–(C). Anexternalist could endorse the following epistemic principle:

(Ext-APJ) Necessarily, if S has an intuition that p and S’s intuition is produced by areliable process, and S believes that p on the basis of S’s intuition thenS is prima facie justified in believing that p.18

This principle is externalist because it entails that there is an external condition—reliability—included in the grounds for justification. The externalist who accepts (Ext-APJ) can endorse the explanation provided in (C) for why a priori beliefs are reliable,and so accept that there are a priori justified beliefs. Clearly, for reasons similar to thosediscussed previously regarding the internalist theory, this theory can accommodate thefallibility and defeasibility of a priori justification. Again, just as with the internalisttheory, I am not proposing that this is the correct theory of a priori justification, ratherI simply claim that this is an externalist way of developing such a theory, and that thistheory incorporates (A)–(C).19

Whichever way one develops an account of a priori justification and knowledgearound the ICI theory, one can also explain how some a priori beliefs are justifiednon-inferentially and others inferentially. A thinker may have a direct, non-inferen-tial intuition that p, and believe that p on the basis of the intuition, and as long as theconditions for justification and knowledge are met, the thinker’s belief will be non-inferentially a priori justified. Furthermore, if the thinker’s intuition is suitablyinfluenced by the CDEIC for her concepts, then her intuition and belief will betrue and reliably formed. A thinker might also inductively or abductively infer aproposition q from a host of non-inferentially a priori justified non-beliefs. If such aninference is carried out appropriately, then the thinker’s belief that q will beinferentially a priori justified. Philosophical theories about knowledge, causation,justice, and other such concepts are justified (if at all) in some such way, and thussome philosophical theories about these concepts may well be justified on the ICItheory.

78 JOSHUA C. THUROW

5 Williamson’s Argument and a RevisionIn recent work, Timothy Williamson (2006, 2007) argues that there isn’t anything inparticular that one must believe, or any inferences that one must be willing to make, inorder to possess a concept. He takes this to be devastating to the project of explaining apriori knowledge and justification by way of understanding. This kind of project—ofwhich Peacocke and my theories are varieties—proposes a link between understandingand belief/inference and then a link between belief/inference and truth/validity, insuch a way that one’s beliefs are guaranteed to be (mostly) true and one’s inferences areguaranteed to be valid, simply in virtue of understanding. If Williamson is right, thenhe thinks that he has severed any link between understanding and belief/inference,thus undermining this kind of explanation of a priori justification and knowledge.To establish his conclusion,Williamson presents two people who understand a concept

completely, indeed would be considered more competent than the normal person, yetwho do not accept such an evident claim as that “all vixens are vixens.” In one of hisexamples, a person does not accept this claim because they take universally quantifiedclaims to entail existential claims, and furthermore this person does not believe there areany vixens. In another example, a person does not accept this because he denies bivalence,and so there are situations in which the above claim has no truth-value. Each person hasthought a lot about his views on quantification, defends them ably, and takes his views tobe about the very same concept that everybody possesses. Furthermore, Williamsonimagines that after thinking so much about the concept, neither person is at all disposedto accept that all vixens are vixens. It’s not as though they suspect that it is true, have anintuition that it is true, but bad reasoning has convinced them otherwise. Rather, theywholeheartedly deny that all vixens are vixens. It seems, then, that neither person accepts—explicitly or implicitly—or intuits “all vixens are vixens,” but this claim might be thoughtto be one of the best potential cases of a content-determining principle for ALL, and yetboth people possess and understand ALL. So, in terms of the model I presented in }5, itseems that people can possess a concept without having a CDEIC, which is to deny anessential component of the explanation of how a priori beliefs can be reliable.I think that Williamson’s examples are compelling, so somehow the model must be

changed to account for these possibilities. How could this be done? Well, even thoughthe people inWilliamson’s examples each possess the relevant concepts, and understandthem well—I’ll even grant Williamson that they have complete linguistic understand-ing of the words associated with the relevant concepts—it doesn’t seem that theycompletely understand some of the concepts, especially the concept ALL. These peopledon’t fully understand the concept because, quite simply, they don’t employ it properly.They might employ it quite well, and be justified in thinking that their theories of ALLare true, but since their theories are false, and the truth of “all vixens are vixens” seems tobe something that falls out of the nature of the constituent concepts, they must to somedegree misunderstand ALL.


At first blush, the claim that these people misunderstand ALL to some degree mightseem unhelpful. It might seem that once we deny that they fully understand ALL, wehave gone down the path to trivializing the connection between understanding andtruth because it would seem that if a person accepted any false claim, we would bebound to say that that person misunderstands some constituent concept of that claim.But, then we have in effect stipulated that one better understands a concept to thedegree that one has more true beliefs that contain the concept. This stipulation wouldnot help one bit in explaining a priori knowledge because, on this view, understandingdoesn’t explain how our beliefs are reliable—understanding is just another way ofsaying that our beliefs are reliable.

Not only is this notion of understanding useless for explaining how our a prioribeliefs are reliable, but the notion is itself highly suspect. I can completely understandthe concept STAR while having many false beliefs about stars, such as false beliefsabout how many stars there are, how big stars can be, how stars form, etc. So, how arewe to understand the notion of understanding so that we do not walk into this deadend, yet also so that we can account for Williamson’s examples and provide someexplanation of how a priori beliefs can be reliable?

Here is how it can be done. I will first put forward my proposal for how to revise theICI theory given in }5 and then I will discuss how it avoids Williamson’s objection.First, we maintain (A) and (B), which simply describe what it is to form a belief in an apriori way and to make an a priori transition. Next, we replace (C) with the followingset of claims:

(D) One’s a priori beliefs and transitions are typically suitably influenced by20 one’sapparently content-determining explicit and implicit conceptions,

(E) To understand a concept is for one’s apparently content-determining explicitand implicit conceptions to approximately match the principles and uses thatare genuinely content-determining,

(F) One better understands a concept the better one’s apparently content-determin-ing explicit and implicit conceptions match the genuinely content-determiningexplicit and implicit conceptions.

Apparently content-determining explicit and implicit conceptions (hereafter A-CDEIC)are psychologically just like genuinely content-determining explicit and implicit con-ceptions (i.e. CDEIC). All of the ways of distinguishing CDEIC described in }3.1 alsodistinguish apparently CDEIC, we just now drop the assumption that they actuallydetermine the content.

This model can explain how our a priori beliefs are reliable. (E) establishes a reliablelink between one’s A-CDEIC and the CDEIC. Given (F), that link gets tighter thebetter one understands a concept. (D) establishes a reliable link between one’s A-CDEIC and one’s a priori beliefs and transitions. So, in virtue of understanding, one’simplicit conception is for the most part accurate, and one’s beliefs and transitions will

80 JOSHUA C. THUROW

for the most part validly follow from the implicit conception. As a result, one’s a prioribeliefs will be reliable and one’s a priori transitions will for the most part be valid.

Put a little differently, this new model establishes the following reliable links:

(L1) content-determining principles/uses—truth(L2) A-CDEIC—content-determining principles/uses(L3) a priori beliefs/transitions—A-CDEIC.

(L1) is completely reliable given how the principles/uses determine the content of aconcept. The notion of understanding described in (E) and (F) establishes (L2), and (D),in effect, posits (L3). (D) should be taken to follow from the notion of an implicitconception—it must be the kind of thing that, for the most part, actually influencesone’s beliefs in the right way. Consequently, as long as (L2) and (L3) are tight enough(and there’s no reason why they couldn’t be), there will be a reliable link betweenone’s a priori beliefs and truth and one’s a priori transitions and validity. In other words,one’s a priori beliefs will be reliable and one’s a priori transition will be for the mostpart valid.Principles (D), (E), and (F) introduce a distinction between apparently CDEIC and

genuinely CDEIC—how can the A-CDEIC not be CDEIC? Apparent CDEIC maynot be genuine if the social environment of a thinker plays a role in determining thecontent of concepts. Tyler Burge (1979, 1986) has famously argued that the socialenvironment contributes to determining the content of concepts. There is still disputeover the truth of this view as well as—amongst defenders of the view—which kinds ofour concepts are determined by aspects of our social environment (see Lau and Deutsch2010 and references therein). However, the view is widely accepted, and I propose toincorporate it into the revised ICI theory. Note, though, that there is an internalisticand externalistic way that the social environment can contribute to determiningcontent. According to the internalistic way, the individual’s A-CDEIC includesdeference to the community in determining the content of the concept. Accordingto the externalistic way, the A-CDEIC that is held by the community in generalautomatically determines the content, regardless of whether one chooses to defer.Either sort of view is consistent with the revised ICI theory, and both allow for thepossibility that one’s A-CDEIC does not match the CDEIC.The revised ICI theory can accommodate Williamson’s examples because it does not

require that one accept—explicitly or implicitly—any specific propositions or inferencesin order to possess a concept. In order to possess a concept, one’s A-CDEIC must tosome degree match the A-CDEIC of the people in one’s community, and the CDEIC issome function of the communal A-CDEIC. No specific intuition is required to possess aconcept; neither is it required that one be disposed to accept some particular inference.The people in Williamson’s examples possess the concept ALL and understand it well invirtue of the fact that their implicit conception does pretty well match the genuineconcept-determining conditions—that is, the conditions that the community accepts.The implicit conception ofWilliamson’s people and the community match fairly closely


because they would share many intuitions concerning the validity of inferences involv-ing ALL in many circumstances. As a result, Williamson’s people could come to knowthat all vixens are vixens if their implicit conceptions would come to more closely matchthe community’s. They may never come to know that all vixens are vixens, however.Their implicit conception might become so firmly entrenched, perhaps because of theirbelief that they are experts with the term, that it couldn’t be altered.

Furthermore, the notion of understanding embodied in (E) and (F) doesn’t entailthat one must have no false beliefs involving a concept in order to fully understand theconcept. This is because, first, there are many true propositions involving a conceptthat are not determined to be true by the content-determining principles and uses, soone could fully possess a concept and yet believe of one of these true propositions thatit is false. Second, even if one’s implicit conception completely matches the genuinecontent-determining principles and uses—and so one completely understands theconcept—one may still have some false beliefs involving the concept given (D),perhaps because one is influenced by the misleading testimony of someone else, orone’s judgment is influenced by a false theory about the concept, or simply because onehasn’t reflected carefully enough.

Williamson (2007: 121–33) acknowledges the role that deference to a linguisticcommunity can play in determining the reference of a term or the content of a concept.However, he argues that deference provides no resources for grounding a link betweenunderstanding a concept and having justified beliefs about the concept. His reason forthis is that ‘understanding’ is ambiguous between a) simply grasping or possessing therelevant concepts, and b) the many things that one accepts in virtue of which oneunderstands the relevant concepts to the degree that one understands it. Going withoption a won’t work because merely possessing a concept can’t explain why one wouldhave or be justified in having any given belief, since there are no beliefs onemust have inorder to possess a concept (for the reasons we’ve discussed previously). Going withoption b won’t work because “the facts that constitute your understanding of a givensentence include various cognitive capacities that are not in general necessary forunderstanding that sentence, but help to make up your particular competence withit . . . thus [these facts] are too thick to yield bases for analyticity” (2007: 131).Williamson may be right that option b won’t help to legitimate analyticity, but

nevertheless, deference can be used in an account that explains how concept possessionis reliably linked to true intuitions and a priori beliefs; my earlier account does this.Furthermore, one can explain how a priori beliefs formed via intuitions are justifiedusing Int-APJ or Ext-APJ in }4 or some other epistemological theory. So, a linkbetween understanding and justification can be grounded. In addition, the ICI theorydoesn’t imply that everything we implicitly or explicitly accept contributes to deter-mining the content of our concepts, and so constitutes our understanding, in therelevant sense. In }3.1 I gave some criteria for distinguishing which uses are apparentlycontent-determining and which are not; so the sort of understanding that the ICItheory uses falls somewhere between options a and b.

82 JOSHUA C. THUROW

6 An ObjectionAccording to the ICI theory, then, what explains why one’s A-CDEIC isn’t necessarilygenuinely content-determining is that one’s A-CDEIC may not match that of theexperts or of one’s general community. But, this invites the following challenge. Sincehaving an a priori justified belief that involves a concept C requires understanding C,doesn’t it follow that one must know or at least be justified in believing that oneunderstands C in order for one’s a priori beliefs involving C to be justified? But, since tounderstand C is for one’s A-CDEIC tomatch (to a certain degree) the genuine content-determining principles and uses, and what content-determining principles and uses aregenuine depends, often or always, on the principles and uses accepted by one’scommunity, it would follow that in order to be justified in holding some beliefinvolving C one would need to be justified in believing that one’s A-CDEIC forC matches the community’s A-CDEIC for C. But, this justification would surely be aposteriori, so a priori justification would depend upon a posteriori justification, which isincoherent. So, the new conceptual role theory of the a priori is ultimately incoherent.I don’t think this argument succeeds in showing that the new conceptual role theory

is incoherent because I deny the following premise in the above argument:

(I) if being a priori justified in believing that p requires understanding the conceptsof the elements of p, then in order to be a priori justified in believing that p onemust be justified in believing that one understands those concepts.

Here is one reason for thinking that (I) is true:

(IG1) If one must be in r (where r is some state) in order to be justified in believingthat p, then in order to be justified in believing that p one must be justifiedin believing that one is in r.

But, (IG1) is plainly false. In order to be justified in believing anything at all I mustexist, but my justification for believing that there is a computer in front of me does notdepend on my justification for believing that I exist. My justification for believing thatthere is a computer in front of me depends only on my having the concept of acomputer and on my having the right kind of experience.Here’s a slightly different reason one might offer in support of (I):

(IG2) If being in r is constitutive of being justified in believing that p, then in order tobe justified in believing that p one must be justified in believing that one is in r.

(IG2) avoids the previous examples because existing isn’t constitutive of being justifiedin believing that there is a computer in front of me. Nevertheless, (IG2) is false forsimilar reasons. Having a certain kind of experience is partly constitutive of my beingjustified in believing that there is a computer in front of me. However, my beingjustified in believing that there is a computer in front of me does not depend on mybeing justified in believing that I am having a computerish experience. Simply having


the experience is all that is required for me to be justified in believing that there is acomputer in front of me. If (IG2) were true, then perceptual justification would alwaysdepend upon introspective justification since introspection is what justifies our beliefsabout what kinds of experiences we are having. But, this seems wrong.

Here’s another reason for doubting (IG2). My justification for believing that there isa computer in front of me is partly constituted by the truth of an epistemic principlethat says something like, “if one has an experience as of a computer, then one isjustified in believing that there is a computer in front of oneself.” However, I do notneed to be justified in believing this epistemic principle in order to be justified inbelieving that there is a computer in front of me; again, all that is required is that I havethe right experience and that the principle be true.

Furthermore, if I needed to be justified in believing the epistemic principle in orderto be justified in believing (Comp) that there is a computer in front of me, then I wouldhave to be justified in believing an infinite number of propositions in order to bejustified in believing (Comp). For, then, it is constitutive of my being justified inbelieving (Comp) that the following proposition be true: that if one has an experienceas of a computer and one is justified in believing that if one has an experience as of acomputer then one is justified in believing (Comp), then one is justified in believing(Comp). But, then (IG2) entails that I must be justified in believing this proposition,but this produces a new epistemic principle that (IG2) entails that I must be justified inbelieving and so on ad infinitum. Not only does it seems obviously false that I mustknow an infinite number of propositions in order to be justified in believing (Comp),but it is far from clear that we ever could be justified in believing this infinite series ofpropositions. So, (IG2) is false.

I can’t see any better reason to believe (I) than (IG1) or (IG2), and since these aren’tgood reasons, I think there aren’t any good reasons to believe (I). In addition, thediscussion of (IG1) and (IG2) revealed that there are often a variety of conditions thatmust be satisfied in order for us to have a justified belief and that we do not need to bejustified in believing that all of these obtain in order for the belief to be justified.Understanding a concept is plausibly one of those conditions for having a priorijustified beliefs.

In fact, this claim is endorsed by some prominent contemporary advocates of the apriori (and is rejected by none, to my knowledge). Laurence BonJour writes, “a propos-ition will count as being justified a priori as long as no appeal to experience is needed forthe proposition to be justified once it is understood.”21 Tyler Burge similarly writes,

An apriori justification (entitlement) cannot rely on the specifics of sense experiences or percep-tual beliefs for its justificational force. An apriori justification may depend on sense experiences orperceptual beliefs in some way. They are typically necessary for the acquisition of understandingor belief. But such dependence is not relevant to apriority unless it is essential to justificationalforce. Distinguishing the genesis of understanding and belief from the rational or normative forcebehind beliefs is fundamental to any view that takes apriori justification seriously.22

84 JOSHUA C. THUROW

I agree with both BonJour and Burge, and would add that even if knowing that oneunderstands a concept is a posteriori, this isn’t relevant to the a priori because, inBurge’s terms, the justificational force for a proposition that is justified a priori doesnot depend on being justified in believing that one understands the constituentconcepts.23

I want to close this section by offering a somewhat different response to thisobjection. The objection, and my above response, both assume that understandingthe constituent concepts of a belief is necessary for being a priori justified in holding thebelief. Arguably, though, this assumption is false. Certainly, one must possess a conceptin order to have a priori justified beliefs involving it, but it seems that one might have apriori justified beliefs involving a concept that one possesses, but misunderstands. Let’sgo back to my arthritis example. Even though the person—call him Joe—misunder-stands ARTHRITIS, it is plausible to suppose that, until he receives testimonyotherwise, he is a priori justified in thinking that one can have arthritis in one’s forearm.Why? Because this seems utterly intuitive/obvious to him (that is, he has a strongintuition that it is true) upon reflecting on the claim. The kind of reason Joe has tobelieve what he believes is internally just like the kind of reason that I have to believethat all bachelors are unmarried—each proposition seems like it must be true uponreflection. Here’s another way of putting it. It seems to Joe as if he understands theconcept. The fact that he has such a strong intuition about arthritis is an indication thatit does seem to him that he understands the concept. As long as one seems tounderstand the concept, plausibly—as the arthritis example suggests—one can have apriori justified beliefs involving the concept.If what I have just argued is right, then understanding is in some ways analogous to

the proper functioning of one’s perceptual mechanisms. If one’s perceptual mechan-isms are functioning properly then (assuming that one is in the right environment),one’s beliefs that result from this mechanism will be reliable. However, one’s percep-tual mechanisms needn’t function properly in order for one to have justified beliefsbecause someone whose experiences were not produced in the proper way could stillhave justified beliefs about the world if he bases them appropriately on the experienceshe has. In a quite parallel fashion, if one understands a concept, one will make reliable apriori judgments involving that concept. However, one needn’t understand a conceptin order to have a priori justified beliefs involving the concept. Similar parallels wouldhold to memory as well. The fact that, if the view I have espoused here were correct,memorial, perceptual, and a priori justification would be parallel in the way describedstrikes me as extra evidence in support of this view.24

If understanding is not required for having a priori justified beliefs, then the objec-tion at hand is avoided because the antecedent of (I)—one of the assumptions of theobjection—is not true, and so the consequent (which is what produces a contradiction)cannot be drawn.25


7 Implications for Experimental PhilosophyAssuming the ICI theory I have developed and defended is correct, what followsregarding experimental philosophy? Can the findings of experimental philosophyjustify restrictionism—the view that philosophy should largely abandon a priorimethods?

The ICI theory is built to allow for false a priori belief, and according to the theorythere are three potential locations for unreliability: the connection between belief andintuition, the connection between intuition and A-CDEIC, and the connectionbetween A-CDEIC and CDEIC. In principle, experimental philosophy could helpto determine when some a priori beliefs are unreliable, and perhaps even identify thelocation of unreliability. In addition, experimental philosophy could reveal situationsin which a priori beliefs are particularly likely to be unreliably formed, or kinds of beliefcontent of which we are unreliable a priori judgers.26 We might say, then, that the ICItheory allows experimental philosophy to epistemically prune a priori beliefs and intu-itions.

There is a limit, however, to how much pruning can be done. The ICI theory entailsthat, as long as a person has some reasonable degree of understanding of a concept, thatperson’s intuitions (and a priori beliefs based on those intuitions) will be mostly reliable.Assuming that there are peoplewith a reasonable understanding of concepts (the contentsof which) philosophers discuss, ICI entails that what Alexander andWeinberg (2007) callthe eliminationist position—the view that intuitions “should be done away with in ourphilosophical practice altogether” (2007: 71)—is false. However, it could turn out thatsomewhat sizable sub-classes of intuitions are unreliable; some argue that “the peculiarand esoteric intuitions that are the philosopher’s stock-in-trade” (Alexander and Wein-berg 2007: 71) are such a class. Some form of restrictionism, then, is consistent with ICI, butextreme forms that rule out most uses of the a priori in philosophizing would seem toconflict with the theory since surely many intuitions delivered by the generally reliableprocess described by ICI will be relevant to a priori reflection and theorizing.Assuming one can identify a class of people that understand a given concept well,

ICI implies that study of patterns in their intuitions could in principle provide evidencefor various propositions involving that concept, as most of these peoples’ a priori beliefswill be reliable. Experimental philosophy, then, can in principle assist the standard apriori method of relying on intuition by, in effect, gathering intuitions of people whoputatively understand the concept reasonably well.27

Although ICI allows that experimental philosophy can in principle prune or assistthe a priori method of relying on intuitions, some elements of the theory providepractical obstacles to doing so. First, the theory explains how the a priori method ofbelieving on the basis of rational intuition is reliable. Beliefs produced by this methodcan be pruned or assisted by experimental methods only if the experimental methodsaccess peoples’ rational intuitions. Evidence that peoples’ prompted judgmentsregarding thought experiment T varies based on socio-economic factors isn’t enough

86 JOSHUA C. THUROW

to warrant pruning philosophers’ judgments about T that are based on rational intuition.An individual’s judgments could have a wide range of influences that aren’t at allconnected to her A-CDEIC for the relevant concepts. Judgments aren’t as tightly tiedto the relevant A-CDEIC as rational intuitions—as noted in }3.1, many judgmentsaren’t at all produced by one’s A-CDEIC. It is not clear that existing experimentalphilosophy data gathers rational intuitions.28

Second, the ICI theory allows for the possibility that the A-CDEIC of individualsor groups could be inaccurate. Some, perhaps even many, intuitions produced byinaccurate A-CDEIC will be false. Many people may be like the man in Burge’sarthritis example for some philosophical thought experiments. Suppose you findyourself with a strong intuition regarding a certain thought experiment. Then, theexperimental philosophers come along and tell you that folk intuitions about thisexperiment vary depending upon cultural background and socioeconomic status.What should you do? Well, since it is a live epistemic possibility that the others havean inaccurate A-CDEIC, in some cases one may have reason to stick with one’sintuition and suspect that some of the folk have an inaccurate A-CDEIC. This reactionwould seem rational if one had some reason to think that one’s understanding weremore firm than the conflicting folk; a strong intuition coupled with considerablebackground reflection on the concept seems reason enough, especially if joined withagreement from others who also have spent considerable time reflecting on therelevant concepts. If you are a philosopher with special training in thinking throughthese sorts of thought experiments, and Williamson (2011) is correct in arguing that invirtue of this training we should assume that philosophers are better at making the rightjudgments about cases than lay folk—unless good reason has been given for thinkingotherwise—then holding on to your intuition would seem even more appropriate.This isn’t to say that data about folk intuitions could never undermine a philosopher’sintuition, but that it would be difficult for such data to do so in light of the possibility ofinaccurate A-CDEIC.

8 ConclusionDeveloping a sophisticated epistemological theory of the a priori helps us to respond totwo skeptical challenges to the a priori—the challenge to explain reliability and thechallenge from experimental philosophy. I suggest that the ICI theory of how a prioribeliefs can be reliable is plausible and defensible, while having the flexibility of beingable to work with internalist and externalist theories of both justification and content.By explaining how those who understand a concept to a moderate degree have reliablea priori beliefs involving that concept, the ICI theory thus responds to the reliabilitychallenge and also rules out any extreme restrictionist or eliminationist position onrational intuition. However, the ICI theory grounds promise for the possibility ofexperimental philosophy, while also underlining some practical difficulties it faces inpruning or assisting the a priori method of relying on one’s rational intuition. If the


theory I have developed is on the right track, the best defense of the a priori from thereliability challenge both legitimates a priori methods, and opens them up to pruningand assistance from experimental philosophy.

Notes1. I think Peacocke needs to say, “always, in all possible worlds,” in order to avoid the

possibility of truth-preserving inferences that hold only in the actual world.2. See his discussion of this example in Peacocke 1987, 1992, 1998a.3. Peacocke 1998b, pp. 130–1.4. Peacocke 1998b, p. 134.5. Peacocke 2004, pp. 182–6.6. See Peacocke 1998a, pp. 61–2, and 2004, pp. 176–7.7. More precisely, he proposes a link between implicit conceptions and truth, for any

principles in the implicit conception, and a link between implicit conceptions and validityfor any inference patterns in the implicit conception.

8. I discuss a couple other objections to Peacocke’s view in Thurow 2007.9. The model I will develop of how a priori beliefs can be reliable will only explain how we

could have a priori knowledge of and justified belief in necessary propositions (although onecould have an a priori justified false belief that p, where p is merely contingent). Believers inthe contingent a priori will have to look elsewhere for an explanation of how those beliefscould be reliably formed. Personally, I am skeptical of Kripke’s examples of contingent apriori truths, for reasons given by Casullo (2003: 205–9).

10. The explanation given of necessary a posteriori truths follows accounts given by Bealer1987, Sidelle 1989, and Chalmers 1996.

11. See again Peacocke 2004, p. 172.12. Peacocke 2004, p. 25.13. Peacocke 2004, pp. 24–5.14. See BonJour 1998 for discussion of both conceptions, Casullo 2003 for a detailed discussion

of the negative conception. Jenkins 2008 contains useful discussion of the negative concep-tion as well.

15. I won’t elaborate the principle so that it covers a priori justification for making an inference.It is clear how such an elaboration would go.

16. By, “the grounds of justification are internal,” I mean the conditions that an agent must meetthat are listed in the antecedent of the epistemic principle, (Int-APJ) are internal to the agent.The epistemic principle (which is in some sense one of the grounds of justification, just notthe sense I intend here) itself isn’t internal to the agent. For an argument that the externalityof epistemic principles commits all epistemologists to externalism, see Comesana (2005).

17. For a nice discussion of varieties of accesibilist internalism, see Pappas (2005).18. As with (Int-APJ), I won’t elaborate the principle so that it covers a priori justification for

making an inference. It is clear how such an elaboration would go.19. This externalist theory is a variety of simple reliabilism, a theory that hardly any epistemolo-

gists currently accept. However, it should be clear that other varieties of externalism can bebuilt around (A)–(C) as well, including Goldman’s (1986) “normal worlds” reliabilism,

88 JOSHUA C. THUROW

Bergmann’s (2004) proper function externalism, and Comesana’s (2002) “diagonalized”reliabilism.

20. The notion of being suitably influenced described in (C) in }5 must be modified. It shouldnow be understood as follows: one’s beliefs are suitably produced by one’s implicit concep-tions when they are produced in such a way that they are true if one’s implicit conceptionsfor the concepts that constitutes one’s belief are accurate.

21. BonJour 1998, p. 10. Italics are BonJour’s.22. Burge 1993, p. 460.23. BonJour and Burge’s position is also endorsed by Roderick Chisholm (1977) in his theory of

intuitive induction, and by Jaegwon Kim (1981).24. Perhaps I should say a bit about how Joe could come to understand ARTHRITIS and come

to have reliable a priori justified beliefs about arthritis. Presumably, at some point Joe mustcome to realize that he does not understand ARTHRITIS. This could happen in a couple ofways. First, he could recognize that there is a contradiction in his a priori beliefs aboutarthritis, which could lead him to question whether he understands ARTHRITIS. Second,he could come to see that his understanding of ARTHRITIS quite radically conflicts withhow his community understands ARTHRITIS, and so come to believe that he does notunderstand the concept. He could then change his usage to more closely match that of hiscommunity. Then, we can suppose, he understands the concept, and now his a priorijudgments about that concept will be more reliable.

25. There are a couple of other objections to my theory that deserve replies, although unfortu-nately I don’t have the space here. One objection comes from the possibility of defectiveconcepts (Boghossian 2003), such as BOCHE and TONK. I address this objection inThurow 2007. Another is Jenkins’ (2008: 56–67) argument that Peacocke’s theory doesnot explain a priori knowledge because he needs an explanation for how we come to haveconcepts like those he describes, rather than some other kind of concept for which his theorywouldn’t work. Without such an explanation, it would seem lucky that we are reliable,which precludes knowledge. I don’t think that this kind of luck undermines knowledge;there is a similar kind of luck in perceptual knowledge—if we happened not to be in anormal world, but instead were in an evil-demon or brain-in-a-vat world, our experienceswouldn’t match the world and our beliefs would be unreliable. But, I don’t need to rulethese possibilities out ahead of time in order to know that I have hands. Of course, if we hadevidence that our concepts didn’t work this way, or that we had loads of defective concepts,then knowledge (and justification) would be undermined. But, I don’t think we have anysuch evidence. In any case, even if Jenkins is correct, my main aim in this chapter would notbe thwarted—namely, to respond to the challenge to explain reliability and the challengefrom experimental philosophy. If she’s right, all that would follow is that we would have todo more (something like what she does in her book) to legitimate a priori knowledge.

26. See Liao (2008) for useful further support and discussion.27. ICI thus allows for what Nadelhoffer and Nahmias (2007) call ‘Experimental Analysis.’28. See Bengson (forthcoming) for an excellent discussion of this point. As he notes, the aim of

this point, “is not to ‘immunize’ intuitions against experimental attacks, . . . [but] to identifywhat would be needed to compose a philosophically significant experimental investigationof the epistemic status of intuitions” (manuscript: 18).


ReferencesAlexander, Joshua, and Weinberg, Jonathan, “Analytic Epistemology and Experimental Phil-

osophy,” Philosophy Compass 2/1 (2007): 56–80.Bealer, George, “The Philosophical Limits of Scientific Essentialism,” Philosophical Perspectives 1

(1987): 289–365.Bealer, George, “A Priori Knowledge and the Scope of Philosophy,” Philosophical Studies 81(1996): 121–42.

Bealer, George, “Intuition and the Autonomy of Philosophy,” in DePaul and Ramsey (eds),Rethinking Intuition (Lanham: Rowman and Littlefield, 1998): 201–39.

Bealer, George, “A Theory of the A Priori,” Philosophical Perspectives 13 (1999): 29–55.Bealer, George, “The Origins of Modal Error,” Dialectica 58/1 (2004): 11–42.Benacerraf, Paul, “Mathematical Truth,” Journal of Philosophy 70 (1973): 661–79.Bengson, John, “Experimental Attacks on Intuitions and Answers,” Philosophy and Phenomeno-logical Research (forthcoming).

Bergmann, Michael, “Externalist JustificationWithout Reliability,” Philosophical Issues 14 (2004):35–60.

Boghossian, Paul,“EpistemicAnalyticity: ADefense,”Grazer Philosophische Studien 66 (2003): 15–35.BonJour, Laurence, In Defense of Pure Reason (Cambridge: Cambridge University Press, 1998).Burge, Tyler, “Individualism and the Mental,” Midwest Studies in Philosophy 4 (1979): 73–121.Burge, Tyler, “Individualism and Psychology,” Philosophical Review 95 (1986): 3–45.Burge, Tyler, “Content Preservation,” The Philosophical Review 102:4 (1993): 457–88.Burge, Tyler, “Concepts, Conceptions, and Reflective Understanding: Reply to Peacocke,” inReflections and Replies: Essays on the Philosophy of Tyler Burge, ed. Martin Hahn and BjornRamberg (Cambridge, MA: MIT Press, 2003): 383–96.

Casullo, Albert, A Priori Justification (New York: Oxford, 2003).Chalmers, David, The Conscious Mind (New York: Oxford, 1996).Chisholm, Roderick, Theory of Knowledge 2nd ed. (Englewood Cliffs, NJ: Prentice Hall, 1977).Comesana, Juan, “The Diagonal and the Demon,” Philosophical Studies 110 (2002): 249–66.Comesana, Juan, “We Are (Almost) All Externalists Now,” Philosophical Perspectives 19 (2005):

59–76.Conee, Earl, and Sider, Theodore, Riddles of Existence (New York: Oxford, 2005).Devitt, Michael, “There is no A Priori,” in Steup and Sosa (2005: 105–15).Field, Hartry, “Recent Debates about the a Priori,” Oxford Studies in Epistemology vol. 1 (2005).Goldman, Alvin, Epistemology and Cognition (Cambridge, MA: Harvard, 1986).Goldman, Alvin, “Philosophical Intuitions: Their Target, Their Source, and Their EpistemicStatus,” Grazer Philosophische Studien 74 (2007): 1–26.

Jenkins, C. S., Grounding Concepts (Oxford: Oxford, 2008).Kim, Jaegwon, “The Role of Perception in a Priori Knowledge: Some Remarks,” PhilosophicalStudies 40 (1981): 339–54.

Lau, Joe, and Deutsch, Max, “Externalism About Mental Content”, The Stanford Encyclopedia ofPhilosophy (Fall 2010 Edition), Edward N. Zalta (ed.), <http://plato.stanford.edu/archives/fall2010/entries/content-externalism/>.

Liao, S. Matthew, “A Defense of Intuitions,” Philosophical Studies 140/2 (2008): 247–62.

90 JOSHUA C. THUROW

http://plato.stanford.edu/archives/fall2010/entries/content-externalism/

http://plato.stanford.edu/archives/fall2010/entries/content-externalism/

Nadelhoffer, Thomas and Nahmias, Eddy, “The Past and Future of Experimental Philosophy,”Philosophical Explorations 10/2 (2007).

Pappas, George, “Internalist vs. Externalist Conceptions of Epistemic Justification,” StanfordEncyclopedia of Philosophy (Spring 2005 Edition), Edward N. Zalta (ed.), <http://plato.stan-ford.edu/archives/spr2005/entries/justep-intext/>.

Peacocke, Christopher, “Understanding Logical Constants: A Realist’s Account,” Proceedings ofthe British Academy 73 (1987): 153–200.

Peacocke, Christopher, A Study of Concepts (Cambridge, MA: MIT, 1992).Peacocke, Christopher, “Implicit Conceptions, Understanding, and Rationality,” PhilosophicalIssues 9 (1998a): 43–88.

Peacocke, Christopher, “Implicit Conceptions, the A Priori, and the Identity of Concepts,”Philosophical Issues 9 (1998b): 121–48.

Peacocke, Christopher, The Realm of Reason (New York: Oxford, 2004).Plantinga, Alvin, The Nature of Necessity (Oxford: Oxford, 1974).Plantinga, Alvin, Warrant and Proper Function (New York: Oxford, 1993).Sidelle, Alan, Necessity, Essence, and Individuation (Ithaca: Cornell, 1989).Steup, Matthias, and Sosa, Ernest (ed.) Contemporary Debates in Epistemology (Malden, MA:

Blackwell, 2005).Thurow, Joshua, “The a Priori: A Defense via Explanation,” Dissertation (2007).Thurow, Joshua, “The Defeater Version of Benacerraf ’s Problem for a Priori Knowledge,”

Synthese (forthcoming).Weatherson, Brian, “What Good are Counterexamples?” Philosophical Studies 115 (2003): 1–31.Weinberg, Jonathan, “How to Challenge Intuitions Empirically Without Risking Skepticism,”Midwest Studies in Philosophy 31 (2007): 318–43.

Williamson, Timothy, “‘Conceptual Truth’,” Aristotelian Society (2006) sup. 80: 1–41.Williamson, Timothy, The Philosophy of Philosophy (Oxford: Blackwell, 2007).Williamson, Timothy, “Philosophical Expertise and the Burden of Proof,” Metaphilosophy 42/3(2011): 215–29.


http://plato.stanford.edu/archives/spr2005/entries/justep-intext/

http://plato.stanford.edu/archives/spr2005/entries/justep-intext/

4

The Prospects for anExperimentalist Rationalism,orWhy It’s OK if the A Priori Is Only99.44 Percent Empirically Pure

Jonathan M. Weinberg

1 Empirically Innocent Uses of ExperimentalPhilosophy?

There are a number of arguments and positions on offer concerning the impact that therecent “experimental philosophy”movement should have on the common philosoph-ical practice of relying on intuitions as evidence for theories in such core areas asepistemology, ethics, or metaphysics. Although sometimes they can take a stridentlynegative form, proponents of experimental philosophy also often strike a more concili-atory tone, suggesting only that experimental methods supplement, not totally replace,more traditional armchair methods.1 One might worry that even these more moderateforms of experimentalism might come at too great a cost: the loss of the a priori status ofthe resulting philosophical knowledge. If experimental results play any sort of role inproviding justification for our theories, then it can seem that our knowledge of suchtheories simply must count as a posteriori.

But must indeed this be so? I will argue here that there are a number of ways thatexperimental philosophy can play a substantive role in philosophical theorizing,without thereby needing to undermine whatever a priori status such theorizing mayhave. I am not going to be considering the independent question of whether traditionalarmchair methods really should already be thought of as capable of conferring a priorijustification on its results. The topic here is a doubly conditional one: what are theimplications if armchair methods as they stand are a priori methods, and if experimentalphilosophy were to be added to the philosophical methodological mix.

Let me start with what I hope will seem two fairly clear ways that experimentalphilosophy could work in tandem with armchair methods, but without jeopardizingthose methods’ potential a priori character. In each case, although experimentalphilosophy is making a methodological contribution, nonetheless we need not seeany experimental results serving as a premise in the main philosophical arguments.

1.1 Experimental philosophy as modifier of philosophical practice

The first and easiest way that experimental philosophy could contribute to armchairphilosophical practice without disrupting its apriority is if it serves as an external refiner ofthat practice. Consider the history of mathematics in which developments in the scienceshave often been an important prod to refinements and innovations in mathematicalpractice.2 We should distinguish between our justification in our decision to use aparticular practice in a particular way at some time, and the justification that might beconferred by our so using that practice.Wemay learn, via our empirical encounters withthe world, that a priori insights may be better or more richly derived should we pursuethem in one way rather than another. What matters when considering the status of anysuch insight is whether it was itself generated in part on the basis of empirical information,not whether the choice of possible generator was itself performed on any such basis.So suppose, for example, that we were learn from experimental philosophy that our

cognition is susceptible to framing effects due to order, and that such framing effectscan be eliminated by considering a new case only after a moderate interval has passedsince the last one considered. When we engage in these more isolated acts of reflection,those empirical results would not be playing a role in our contemplations themselves.The experimental investigations, and all the working out of their methodologicalimplications, would have taken place all outside of the armchair practice. And thuswe need not, and should not, consider those results as themselves any sort of input tothose armchair practices. The epistemic dependence here is indirect, not direct, and ofa sort that has not historically been taken to threaten to undermine apriority. Con-sidering one’s thought-experiments in a temporally isolated manner would becomejust part of the background norms of the practice, like using clear notations in one’sproofs or trying not to intuit while inebriated.

1.2 Of the wheat and the chaff

The second way in which experimental philosophy might make a rationalisticallyacceptable contribution is by helping sort good intuitions from bad. A fairly standard viewin the literature on the a priori has developed, responding to arguments like that ofKitcher (1980), that a purely negative epistemic dependence on empirical matters isconsistent with a belief ’s being justified a priori. For example, the justification for mycurrent belief in the result of some modestly subtle proof that I have competently andcorrectly constructed, could be undermined by (misleading) empirical evidence that themathematical community on the whole rejects what I took to be a theorem. But in theabsence of any such empirical evidence, the justification provided by my proof stands,

THE PROSPECTS FOR AN EXPERIMENTALIST RATIONALISM 93

and stands as a priori. As Summerfield (1991) writes, proponents of the a priori can“insist that there is a significant difference between positive and negative epistemicdependence. X’s belief that p is positively dependent for its epistemic status on theprocess that produces and sustains the belief, but it is only negatively dependent onprocesses that would, in some counterfactual circumstances, undermine . . . its status asknowledge” (41–2; emphases original).3 So the mere possibility of as-yet-unperformedexperimental philosophy studies that would undermine the status of some intuitions,should not be enough to place those intuitions outside the set of the a priori.

But there is a further question that becomes relevant here. Returning to the exampleof the proof, what if I were actually to encounter such reports of this dispiritingconsensus of mathematicians contrary to my hoped-for theorem—but thenI acquired further empirical evidence to the effect that those reports were incorrect,and that my proof is indeed one vouched for by all mathematical authorities? Ingeneral, what are we to say about cases in which there is a belief that p that wouldhave a priori justification, except for the presence of putative empirical defeaters, whichare themselves empirically defeated? The verdict here is less obvious, I think. Clearlythe epistemically restorative empirical findings in this sort of case are playing some sortof actual role in securing the status of the belief, in a way in which the absence ofdefeating findings in the original case was not. But it is, importantly, a purely defensiverole. The results do not themselves provide any evidence for p itself. They only serve toallow the original a priori source of evidence to do its job unencumbered by what, inthis case, would be misleading empirical annoyances. Compare to my vouching foryour general trustworthiness in the face of allegations against your testimonial reliabil-ity, when you have asserted that p. I have not thereby offered any evidence for p. I haveonly tried to defend you from those who would lessen the rightful epistemic impact ofyour assertion. So we ought to consider that when empirical results that threatened thestanding of an intuition are themselves defused by further results, the resulting epi-stemic state is itself legitimately a priori. This seems to me not to be a positive reliance,but is merely a counter-negative one.

(Nonetheless, I cannot avoid feeling some misgivings about putting so much weighton these sorts of intuitions of positive vs. negative reliance here—perhaps views on thismatter will simply track individuals’ intuitions about causation and preemption?4—solet me also foreshadow that further considerations in section 3 will also lean in thisdirection as well.)

Transposing these considerations into the key of our topic of experimental philoso-phy, suppose that the experimentalists carry out their projects with substantial success,and in a way such that at first some intuitions that had taken to be epistemicallyunproblematic become threatened; and then later some such initially impeachedintuitions become restored to our good epistemic graces by more sustained investi-gation,5 even as others perhaps remain considered as untrustworthy. At the end of theday, in parallel with the considerations in the previous paragraph, the intuitions in theformer group can count as having had their apriority restored to them with their good

94 JONATHAN M. WEINBERG

epistemic standing. And as for the latter group, well, they weren’t good candidates tobe sources of knowledge anyway, since they were untrustworthy. So nothing wouldbe epistemically lost here by the appeals to the experimental results.We have thus far considered two roles that experimental philosophy could play in

our philosophical practices that need not impugn the a priori status of the products ofthose practices: either by helping us to revise those practices, or to help us defend somesubset of our intuitions when perhaps a larger set of them have been empiricallyimpeached. The trait that they share, that renders them potentially friendly to ration-alism, is that in each, core pieces of philosophical reasoning are doing the epistemicheavy lifting that do not themselves depend upon any experimental results, eventhough those results have had a positive methodological influence on that reasoning.Maybe we only got ourselves to a point to have the insight in the first place due tosome experimental results, or maybe we had some results that allayed some empiricallygenerated concerns about the insight. But in both cases, the insight itself is at the centerof the philosophical enterprise. One can separate off the contribution of the intuitingsthemselves, and the contribution of the empirical methods. Indeed, the empiricalmethods appear to be at the service only of amplifying the power of the intuitions. Atthe end of the day in such cases, the weight of the philosophizing is borne by instancesof intuiting that p, and thereby being able to add p to one’s stock of evidence, and thendrawing inferences from the resulting stock of intuited propositions. All the experi-mental information about the psychology of intuiting will thus remain external to thiscore step. When a philosopher is enjoying an intuitive epistemic uptake of p, they willnot need to be themselves relying in any evidential way upon the experimental work.And, as I have argued, this is an at least somewhat rationalism-compatible result.

1.3 Inescapably empirical roles for experimental philosophy?

Yet not all contributions of experimental philosophy will necessarily respect theintegrity of that core intuiting-intuited step. Consider recent work on pragmaticencroachment by Angel Pinillos. In brief, he surveyed ordinary subjects on theiranswer to the question of just how many times a student in a hypothetical casewould need to proofread a paper in order for that student to know that it containsno typos. One set of his subjects saw a version of this scenario in which the stakes arelow for the student, in that it does not matter too much whether there is or isn’t a typo,while another set of subjects saw a version in which it matters a great deal, with theretention of a valuable scholarship at stake.6 And Pinillos found an interesting statistic-ally significant difference: the high-stakes group gave a median answer of five proof-readings, whereas the low-stages group gave a median answer of two. He offers this assome evidence for the thesis of pragmatic encroachment: that whether or not S knowsthat p does depend, at least in part, on how much it matters for S that p be true.I am not interested here in these particular results themselves (though I do think it to

be a very exciting and novel line of x-phi research), but rather in the overall argumenta-tive strategy. The inference offered is from an observed pattern in their experimental

THE PROSPECTS FOR AN EXPERIMENTALIST RATIONAL ISM 95

data, to the likely correctness of anti-intellectualism about knowledge. But this inferencedoes not move through a step inwhich the individual intuitings are seriatim considered tobe correct. It is entirely consistent with this argument, for example, that people are ingeneral a bit too skeptical, and that a greatmany—more than half, even—of their refusalsto attribute knowledge are in fact incorrect. (Or, for that matter, that people are ingeneral a bit too credulous, and that their attributions arewidely over-generous.) Pinillosis only claiming that stakes play some legitimate role in determining whether a giventoken belief is really an instance of knowledge. He is not committed to the correctness ofthe overall resulting intuition in any particular cases at all. It follows that there is noscreening here, no way to take his results as merely securing the good evidential wheatand threshing out the bad psychologically flawed chaff. And thus there is no way to treattheir experimental findings as anything other than premises in their overall argument.

There is another, perhaps even more threatening, way in which the experimentalresults may be evidentially ineliminable. Methods like Pinillos’ are still somewhatcontroversial. But what is not nearly so controversial is the idea that major philosophicalarguments very often depend centrally on holistic, abductive, or ‘wide reflectiveequilibrium’ forms of inference, even in prototypically armchair areas of investigation.7

Part of the attraction of such modes of inference is precisely that they allow room tonavigate around unwieldy or unwanted intuitions, since they are powerful holisticforms of inference that make various sorts of trade-offs between catching the intuitivedata, and such theoretical desiderata as simplicity. Yet this trait that can be so attractive,also carries with it a substantial danger to the would-be rationalist. For these inferencesoperate under a norm of total evidence—one cannot select just those bits and pieces ofphenomena that one would like to explain or accommodate, but rather one must offerone’s theory in competition with other theories competing on the field of all thephenomena that might be relevant. And this is a door through which it seems decidedlya posteriori evidence might enter.

For suppose that one theory T1 would be the proper conclusion of an inference tothe best explanation over its rival T2, but only if some set of putative intuitions8 {I1,I2 . . . } are legitimately operative as evidence in that inference, whereas T2 would bethe legitimately superior explanation for an evidence set that did not include the {Ii}.Such situations are not especially uncommon in philosophy, which is why philosophersso often do try to offer debunking or “explaining away” arguments against theintuitions of their rivals. In cases like these where experimental philosophy comesinto play, one might have hoped that experimental findings speaking to the legitimacyof the {Ii} would fall under the wheat-from-chaff category considered above. That is,if those intuitions are impeached by empirical findings, then they may be legitimatelyremoved from the evidence set; alternatively, should the overall set of results considerthose intuitions legitimate, it is still the {Ii} themselves that are doing all the evidentialwork. Yet this is, unfortunately, not so. What abductive inferences can properly bedrawn from a set of evidence is not just a function of the first-order members of that set,but also, importantly, of the second-order evidence about how the set of first-order


evidence was composed. A scientist almost never makes an inference from a set ofobservations simpliciter, but rather from those observations plus some very importantinformation about how those observations were made (e.g., by means of what instru-ments); by what means it was decided that these particular observations would be made(e.g., from a random selection of possibilities, or according to some selection rubric);about how no other observations were excluded, or if some observations wereexcluded, the basis for those exclusions (e.g., due to a contamination of some ofone’s samples); and so on. Much of the power of inference to the best explanationcomes from the inclusion of this second-order evidence. Even where nothing ofinterest may follow from a few dozen data points considered by themselves, thosesame data points considered as the product of very particular and systematic processes ofselection and observation can make or break a theory. And thus, if experimentalinvestigations have ruled some intuitions as legitimate, and others as not so, thenthose very investigations must themselves be considered part of the evidence in apiece of philosophical abduction. Even if the intuitions themselves counted as a prioriwhen considered in isolation, they would only be a proper part of the philosophicalevidence, and the rest of the evidence looks to be decidedly a posteriori. And so itmight seem that the conclusions of inferences in both of these latter sorts of cases wouldbe definitively a posteriori, since empirical, experimental evidence would comprise anineliminable part of their premises.And so, as x-phi arguments like Pinillos’ begin to proliferate, would they spell a kind

of doom for a rationalistic self-understanding of at least some core areas of philosophy?My original take on this question was: yes. I had planned to write a paper in whichI would canvass the two empirically innocent options of revising our practices andseparating wheat from chaff, and from there make the case that this third kind of optionis itself empirically burdened, and so could undermine such a rationalistic conception ofphilosophy. But some further thinking about the nature of the a priori has led me todoubt that initial answer. To put these ideas in play, we need to back up for a minute,and ask ourselves what the point of the epistemic category of the a priori is supposed tobe in the first place.

2 The A Priori and the Point of PurityThe most commonly embraced arguments for the a priori today all turn on allegedinsufficiencies of purely experiential sources of justification. For example, perhaps weneed something more than experience to get us to go beyond a threatened skepticismof the present moment (BonJour 1998), or to have an unconditional grasp of themeanings of our own claims ( Jackson 1998). I will be most interested here inarguments that go back at least to Kant suggesting a similar need for the a priori toascertain necessary truths. But note that all these arguments, if they succeed at all (andI am in the minority that is inclined to think they do), only secure the existenceof supra-experiential justification. They show that there must be more to our epistemic


lives than purely experiential justification. What they do not show, however, is thatthis something more must itself be purely non-experiential. It is consistent with sucharguments that perceptual experience always play some positive role in all cognitions,just so long as in many instances that experiential component is in some waystrengthened or supplemented by something further, even if that something furtheris itself informed by experience. So the idea of the a priori as a zone that not only goesbeyond perception, but which is in some important sense untouched by perception,requires some further motivation. There is an asymmetry here, in which all the casesinvolving a mix of empirical and rational justification are lumped with any purelyperceptual cases into the category of the a posteriori. Is there any good reason for usepistemologists to posit these categorical boundaries, instead of, say, positing insteadtwo pure categories, and one mixed category?9 Or rather, since of course we canconsider and deploy such a more nuanced taxonomy, should we choose to, we shouldask: does it really make sense for us to have the kind of special interest that we clearly doin the less-nuanced way of carving up the epistemological in the way that the a priori/aposteriori distinction does currently and always has done historically?

I can think of two reasons that one might want to give such particular emphasis onthe epistemic category that is purified of all perceptual epistemic contributions. Thefirst such reason is that one is concerned about the epistemic status of perception itself,and wishes to re-found perceptual justification on what one takes to be a more securegrounding in the a priori. This is clearly part of Descartes’s objectives in theMeditations.But it risks understatement to say that this is not a project with many adherents today. Ifwe are not to simply retire “a priori” as an atavism from earlier eras of philosophy, weshould look for other considerations that motivate its relevance.10

One might think that, just as sense-perception has over time become philosophicallyless problematic than rational insight and its ilk, so too might our particular interest inthe a priori stem from particular concerns about whether such nonexperiential justifi-cation can make any sense at all. But note that that concern—as legitimate as it is—would motivate a different set of categorical emphases. If the nonexperiential itself istaken to be philosophically dubious, then it is the purely nonexperiential together withthe mixed cases that would collectively merit such concern. If it’s mysterious how wecan have any sort of nonexperiential justification at all, then mixed cases will not beespecially less mysterious. These concerns motivate the interest mentioned beforein supra-experiential justification. But they do not seem actually to be a reason to beinterested in the a priori as it has typically been configured.

But I do believe that one can locate a further legitimate candidate motivation for acategory of aperceptual purity: the fear that relying on perceptually ascertained prem-ises will thereby render one’s argument incapable of speaking to matters of necessity.The concern is not merely that experience by itself is not up to the task of providing uswith justification for a claim of necessity. It is the more epistemologically nervousconcern that appealing to experience may render impotent whatever other meanswe might have hoped to have to license such modal claims. Experience is not just


itself modally weak, but indeed it threatens to be modally enervating. If in general, aninference can only yield a conclusion no stronger than its strongest premise, thenadmitting an empirical, and hence contingent, premise into one’s reasoning will renderit incapable of justifying a modally strong conclusion.One can perhaps see some of these concerns in lines like these from Kant, from the

Second Analogy. He is here objecting to the idea that we can properly come toappreciate the necessity that inheres in the principle that changes in time follow thelaw of cause and effect. According to standard opinion, Kant writes,

. . . only through the perception and comparison of events repeatedly following in a uniformmanner upon preceding appearances are we enabled to discover a rule according to which certainevents always follow upon certain appearances, and that this is the way in which we are first led toconstruct for ourselves the concept of cause. Now the concept, if thus formed, would be merelyempirical, and the rule which it supplies, that everything which happens has a cause, would be as contingent asthe experience upon which it is based. Since the universality and necessity of the rule would not begrounded a priori, but only on induction, they would be merely fictitious and without genuinelyuniversal validity. (B241; emphasis added)

To be based on experience even at all, Kant is perhaps saying here, is sufficient topreclude any real judgment of necessity.The basis for such fear of inferential enervation is clearly manifest in how we have

designed many of our formal inference systems. In many versions of the predicatecalculus, there are rules that allow one to derive formulas of the form “(x)Fx” fromthose of the form “Fa,” but only when there are very demanding restrictions placed onwhere that “a” could have come from. These restrictions—which oftenmake the rule forUniversal Generalization much more complicated than any other rules in the system—

are meant to screen out any inadmissibly local information that may pertain only to somesubset of objects in the universe. For example, if “a” was introduced via ExistentialInstantiation from “(Ex)Fx,” then it is to be kept off-limits from any such application ofUG. Very similarly, andmost relevant to our considerations here, modal logics that allowin various forms of contingent information, such as by means of an actuality operator,have to impose similar restrictions on the Rule of Necessitation. Otherwise, one couldultimately derive from “@p” to “p” to “[ ]p”whichwould be a modal epistemic disaster.So, we have a very reasonable motivation for something like the purity aspect of

the standard negative gloss on the a priori: in general, the conclusion of an argument cannothave a stronger modal status than that of its weakest premise. In general, if an argumentappeals to particular facts about the actual world, then its conclusion cannot properlyhave the force of necessity.11 For all those “in general”s need to be taken seriously, andought not to be read as merely hedged universal claims. There are of course perfectly goodarguments that are counterexamples of a sort. For example, the following argument is valid:

1. @p ➔ [ ]q2. @pTherefore, 3. [ ]q


(Though one would of course face a question here as to where one gets justification forpremises like 1; this will be a matter of some discussion later.) This makes it somewhathard to figure out just how stringent an empirical purity requirement there shouldperhaps be on the a priori, or what specific shape any such requirement should take.

There are some clear cases in which empirically based knowledge nonetheless hassome modal strength to it, as Casullo has pointed out in the context of responding toattempts to cash out the a priori in terms of a universalized version of the Kantian idea.Casullo writes (2003):

Kant maintains that experience can teach us only what is the case. But a good deal of our ordinarypractical knowledge and the bulk of our scientific knowledge provide clear counterexamples tothe claim. My knowledge that my pen will fall if I drop it does not provide information aboutwhat is the case for the antecedent is contrary-to-fact. Scientific laws are not mere descriptions ofthe actual world. They support counterfactual conditionals and, hence, provide informationbeyond what is true of the actual world. (93)

Clearly Casullo is right about the epistemic phenomena as described there, in that we dohave such experience-based knowledge of some trans-actual propositions. But themodal potency of such inferences is consistent with the modal enervation concern stillcarrying significant weight. First, it might be that the modal enervation concerns addressonly with more robust flavors of modality than the nomological, such as metaphysical,conceptual, or logical necessity.12 Casullo’s examples of knowledge display only fairlylocal forms of modality, and indeed ones that we expect to require substantial knowledgeof the actual world in order for us to possess them, of exactly the sort that experience isrequired for. So it may be that Casullo’s examples show that the Kantian slogan as itis commonly worded is an unacceptably strong way of trying to voice the modalenervation worry, while nonetheless we could attempt more metaphysically nuancedversions, say in post-Kripkean parlance: experience only teaches us of the actual world,and of worlds that are near to the actual world, but it cannot teach us of what is true ofall worlds.

Moreover, the idea of modal enervation is consistent with our having the kind ofpractical and scientific knowledge that Casullo adverts to, if we assume that we canonly have such knowledge when we also have contingency-unblemished cognitionsthat make available to us various claims or modes of inference. For example, BonJourhas argued, in the concluding chapter of his (1998), that we need an a priori insight ofnecessary truths about induction, for our deployments of standard schemata of induct-ive and abductive inference to be justified. He offers the following as a necessary truthgleaned from a priori insight:

(I-2) So long as the possibility that observation itself affects the proportion of As that are Bs isexcluded, the best explanation, that is, the most likely to be true, for the truth of a standardinductive premise is the straight inductive explanation, namely that the observed proportion ofm/n accurately reflects (within a reasonable degree of approximation) a corresponding objective


regularity in the world (and this likelihood increases as the number of observations and thevariety of collateral circumstances of observation increases). (212)

BonJour frames his principles in terms of “objective regularities,” which are prettynaturally cashed out in terms of a local modality. That the As have an objective propensityto also be Bs should be consistent with the majority of actual As not actually being Bs—those that are both A and B being especially tasty, perhaps, whereas the non-B As aren’tworth the effort to harvest—or for that matter, with there no longer being any actualAs left at all. Thus BonJour’s (I-2) is a necessary principle that can yield some distinct trans-actual claim when combined with some premises about what is actually the case. So, onthis picture, knowledge of the counterfactual about Casullo’s pen required us also to have,as a premise inour reasoning, ametaphysical necessity thatwecouldnot havebeen justifiedin using in the first place without an experience-free epistemic license for it.Now, everything here turns on a so far undefended presupposition that experience by

itself cannot provide us with knowledge that has any modal strength to it. I think it isimportant to recognize the existence of that presupposition, and that it is one that can beseriously debated, and that debate is not likely to be resolvable on purely a priori terms,since it will likely depend onmany contingent facts about the make-up of perception.13

But if someone were to think that experience can provide knowledge of sufficientmodal strength, then it is hard to see what could motivate the purity aspect of the apriori for them. For that person, the central questions of this chapterwill just not be liveones. So for now I will take it that experience either cannot by itself provide us withknowledge with any modal strength, or at least with only restricted, local formsof modality, such as with fairly standard sorts of counterfactuals about nearby middle-sized dry goods.Where these considerations leave us, then, is with a good reason for theorists

with substantial modal ambitions—such as, typically, the mathematician or the meta-physician—to fear the modal enervation of their methods, should they be too carelessin allowing empirical, and hence likely contingent, premises into their reasoning. Andthis fear in turn motivates an empirical purity condition on the a priori. This contin-gency-phobia also makes sense of why the a priori so often has two different forms ofcharacterization, which have nonetheless never clearly been mutually entailing. Oneset of characterizations are in terms of our positive sources of justification about mattersof necessity; whereas the other, negative component is needed in order for the positiveone to do its job unhindered and unburdened by the merely contingent informationthe senses would be bringing in. Both of these positive and negative aspects of the apriori need to be in place to secure our modally strong cognition. The positive withoutthe negative risks empirical contagion and thence enervation; the negative without thepositive risks leaving matters unclear as to whether we even have a source of modallystrong justification to begin with.Nonetheless, as we have already suggested, this motivation does not obviously license

an exceptionless purity condition. If there are ways of having modally weak premises in


our inferences that somehow would not irrevocably lead to equally weak or weakerconclusions, then these would provide the raw material for such exceptions. Andpropositions like BonJour’s inductive intuitions, or Kripke’s essentialist conditionals(e.g., “if this desk is actually made of ice, then it is necessarily made of ice”) point us inthe direction of at least one class of legitimate exceptions. These propositions have aninteresting property that we might call modally invigorating—they have antecedents thatare modally weaker than their consequents. So a modally weak premise p will notenervate our inferences, should we have another modally invigorating premise with p asan antecedent, and a target necessary proposition as its consequent. Now, those twoexamples are ones where the entire proposition is (at least allegedly) itself modallystrong, and knowable on an a priori basis.Wemight therefore wonder whether that willalways be the case. It would provide a very tidy package, one that should be amenable tothe rationalist: reasoning will be a priori only when it is either empirically pure, or undera very special set of limited exceptions that can be themselves determined on an a prioribasis. This would be tidy, but it would also seem to rule out the third type of experi-mental philosophy uses discussed in I.C. Let us see whether we can allowmatters to get abit messier, while still respecting the fundamental theoretical motivations in play.

3 Experimental Philosophy and Modal InvigorationThe fear of modal enervation pertains most directly to possible epistemic practices likesubstituting observation and induction for geometric proof: trying to incorporateprotractor measurements of actually existing physical triangles would seem to defeatany inference to the full-blown mathematical necessity of the claim that the sum of theangles of a triangle in a Euclidean space sum to 180 degrees. It also seems likely, thoughperhaps not quite so obvious, that it applies correctly to more broadly theoreticarguments, like attempts to revise classical logic in the face of various phenomena inquantum physics. In these sorts of cases, there are no obvious candidates for modalinvigorators. But the way in which contingency comes in with experimental philoso-phy is rather different. Experimental philosophers (at least of the stripe under consider-ation here) are not proposing to study knowledge, causation, moral goodness, etc.directly, on any analogy with our protractor-wielding empiricist geometer.14 Nor arethey typically appealing to abductive arguments from large-scale scientific pictures ofthe world. Rather, the contingent facts that they are mustering are facts about ourcapacity to latch onto the conceptual or modal structure of the world. In studying our intuitivecapacities, they are learning contingent truths about our ability to learn necessary ones.

Here’s a commitment that anyone not a modal skeptic needs to hold in somefashion: that contingent arrangements of the actual world can reflect at least a mean-ingful portion of the structure of the modal universe that go beyond the actual worlditself. Every epistemically well-functioning human is capable of tracking a largenumber of modal truths, most of which are rather boring (that 2 + 2 is necessarily 4,or that for some particular throw of a fair die, it is possible that it could have turned out


differently). But for us to have this modal epistemic capacity, even in its most trivial andquotidian forms, means that there must be some sort of tracking relationship obtainingbetween the space of contingent physical states that our cognitive systems can enter onthe one hand, and the space of the modal universe on the other. It need not be a 100percent perfect tracking relation, of course, and I am putting aside here a large set oftechnical headaches about how to more strictly define terms like “reflect” or “track”where modal facts are concerned. It is enough here that the actual world has aspects that,should one learn about them, one can thereby be learning, even if only fallibly andperhaps quite messily, about a much greater modal structure. Most possible worlds donot have anything like this; that is, learning about the particular contingent arrangementsinmost worlds can only ever tell you, at best, about the specifics of that world andmaybesome very nearby counterfactuals. Worlds like ours are special, then, in having thisactuality–modalitymirroring property to even the extent it does.Whatmakes ourworldspecial in this way is, of course, that there are creatures like us in it, who are capable ofapprehending the modal in the all-too-contingent structures of our noggins.What this actuality–modality mirroring property presents, then, is another candidate

for modally invigorating propositions. Where some contingent structures in our worldmirror modal reality, then at least in principle we can come to learn about modal realityby learning about those contingent structures. And we can take the experimentalphilosophers to be studying the workings of this modal mirror, in our world and forcreatures like us, by learning about human intuitions. Experimental philosophy can learnabout the actual workings of our power to apprehend modal reality, and on the anti-skeptical assumption that those powers are more-or-less reliable, it can also through suchinvestigations learn about modal reality itself. And thus, because of the actuality–modality mirroring property that (we are non-skeptically assuming) obtains in ourworld, there are contingent premises that can in principle be deployed without causingmodal enervation. Whether those in-principles can be turned into in-practices is aquestion about which the jury is very much still out. It may be that the structure ofthe actuality–modality mirroring will ultimately prove unfathomable to even our bestinvestigations, even in the long term. But what I hope to have shown here is that, shouldthat pessimistic possibility not obtain—should we in fact be able to gain some empiricalunderstanding of the psychological particulars of our rational faculties—the fear ofmodal enervation would not be a reason to rule out any such knowledge as a priori.To recap, once we set aside skeptical concerns about sense-perception itself, the only

real motivation for the empirical purity condition on the a priori is our desire toachieve modally strong conclusions. It can seem that to fulfill that desire, we mustenforce a local ban on any merely contingent15 premises. And perception, evendeployed in systematic and rigorous ways, seems capable only of delivering premisesfrom exactly this unwanted genus. But once we recognize that modal invigoratorsprovide a legitimate exception to any such ban on empirical premises, and once weacknowledge further that ours is a world with the actuality–modality mirroring


property, we can see that at least some contingent premises—so long as they are of avery special kind—need not threaten this fundamental motivation for the a priori.

What does this all mean for the a priori/a posteriori status of the product of the sorts ofmodal abduction projects discussed in I.C.? On the one hand, they flagrantly fail the anystandard formulation of the negative condition on apriority. On the other hand, theyseem to operate largely within the spirit of the positive condition: these projects areleveraging exactly the sort of cognitive windows onto the modal that more traditionalrationalist projects do, but in large part by looking at those windows rather than trying topeer transparently through them. One way to proceed when in the grip of such wrestlinghands, is to seek out a reason to prefer one over the other. The earlier discussion suggestsa path towards doing so: once we consider what good motivations for the category ofthe a priori might be in the first place, allowing the results of these experimentalphilosophy projects to count as a priori seems consistent with those motivations. Infact, looking at the negative and positive conditions on apriority through the lens of themotivation of contingency-phobia, we can see that the former is in its way subservientto the latter. That is, the whole point of wanting some cognitions that are perception-free, is so that the modally strong deliverances of the relevant cognitive capacities canremain empirically unburdened, and thus preserve that modally strong status. Contin-gent premises are only a threat of contagion if there is some cognition of necessity outthere to be contaminated in the first place. In the end, this is why I backed away frommyinitial judgment that the negative condition would automatically place the modalabduction projects on the a posteriori side of the ledger. Perhaps the negative condition,in these special sorts of cases, should be considered trumped.

Those considerations point in one direction. And I also don’t see any reasons at thistime that would tilt the balance back the other way, that is, that would privilege anunrestricted version of the negative condition, as an analysis of the a priori, over aversion that puts more weight on the positive one, and thus rule the x-phi work to bedefinitively a posteriori. But I do think another sort of option needs to be considered:maybe our concept of apriority is just not up to the current methodological andepistemological situation. Hawthorne has recently mustered compelling considerationssuggesting for a set of cases that they involve a priori knowledge of decidedly contingentclaims, and he takes those considerations to put pressure on our traditional epistemo-logical categories here: “It turns out that there is no clear criterion by which the casesunder discussion get to count as deeply contingent a priori knowledge; nor is there anyobvious principle by which they get excluded” (2002, p. 249).16 For example, his“Swampscientist” is formed by a freak accident from the muck of the swamp, and has ahighly reliable capacity for intuitions about basic bodily physics, like “If someone lets goof a massy object above the ground and it is not suspended, it will fall” (p. 252).Hawthorne contends that we should credit these intuitive deliverances as knowledge,but clearly they are not based on any past experiences, since Swampscientist has had nosuch experiences. (Note that Hawthorne is explicitly working within the negativeconception of apriority: “I shall be operating with a standard conception of the a priori


according to which knowledge is a priori iff it does not rely qua knowledge uponperceptual or sensory experience” (248, italics original).)Perhaps the reasons I have offered in this chapter should be taken similarly, as a

challenge to the robustness of the very category of apriority. Certainly the picture ofthe a priori that results bears many surface dissimilarities from what our prototypes areof a priori reasoning. So those who would undermine this distinction itself couldperhaps find more ammunition in my arguments here.For all that, though, I must confess that I would reject any such “apriority

nihilism,” and would instead embrace the inclusive-albeit-revisionary option: acceptthat some experimental philosophy projects can give results that, as odd as it sounds,count as legitimately a priori. One reason to be unafraid to do so is that the concept ofapriority is our concept—it is beholden to nothing but the needs of the epistemolo-gists and some very broad and piecemeal consistency with past usages of that tribe. Ifthe physicists can change the contours of what they mean by “force” or “weight” inthe light of developments within their field, surely it is within our rights to tweak “apriori” similarly. Moreover, I think that it is useful to sever the notion, which isfundamentally epistemological, from debates about experimental philosophy, which arebetter seen as methodological. With that distinction in mind, it need not be paradoxicalfor a sector of philosophy to respect apriority in its epistemology, while embracingexperimental observation of our powers of reason in its methodology. This is myown considered view at this time: outside of a few narrow albeit highly importantzones of primarily formal inquiry, such as logic itself, philosophy cannot be pursuedas a purely armchair endeavor, as a matter of methodological practice. But it mayturn out even so that a much broader class of philosophical inquiry, includingmuch of traditional epistemology and metaphysics, could still legitimately and ap-propriately wear the badge of the a priori. And on the account offered here, we cansee why this is no mere honorific, but instead it would tell us something vitallyimportant about the epistemology in these areas: that it is capable of supportingrobustly modal conclusions.I believe that this line of thought can also de-mystify the putative problem cases

offered by the philosophers just mentioned who have questioned the cogency of the apriori/a posteriori distinction. Without going into detail here, I would close by notingthat the considerations offered here for including much experimentally informedphilosophy as still within the a priori, would end up ruling many of their problemcases over to the a posteriori side of the line. In the experimental philosophy cases, wehave explicit empirical evidence that is about our modal epistemic capacities. Bycontrast, cases like Hawthorne’s Swampscientist (as well as various cases of William-son’s and Jenkins’s) involve the use of information whose contingency could pose athreat to claims to necessity, but where such information is not used as an explicit pieceof evidence. The explicitness or implicitness of the information used is, I think, a redherring. What should matter is whether or not the information is dangerously or


innocently contingent—enervating or capable of reinvigoration—and not the particu-lar psychological form of one’s representation of that information.

Notes1. See, e.g., Nichols and Knobe (2008).2. See, e.g., Kitcher (1988). I am citing Kitcher here on the history; my reading of the

epistemological consequences here diverges from his. He takes this sort of history to beevidence against the a priori status of mathematics, but I think it more apt to be evidence—taking the apriority of mathematics as a given—that such empirically prompted methodo-logical changes are consistent with apriority.

3. See also, e.g., Manfredi (2000).4. And there is anyway some evidence that such intuitions about causation can be a bit wobbly;

see Lombrozo (2010).5. As may well be the case with Nagel (forthcoming) as a response to Weinberg et al. (2001);

though see also Stich (forthcoming) in response. I would also note that there are some deepaffinities here between this use of experimental philosophy, and what Horgan and Hen-derson have called “the low grade a priori” (2001; see also this volume).

6. I should stress here that Pinillos makes no claim that these judgments being surveyed are apriori. I am just considering his experiments in such terms for the purposes of exemplifying aparticular kind of x-phi argument, which (as I will explore shortly) may not be as easy for thewould-be rationalist to accommodate.

7. To offer just two illustrations from epistemological works that I highly admire:

“According to Subject Sensitive Invariantism . . . what you know depends on what is at stakefor you. [It] is advertised as the best explanation for stakes shifting intuitions.” Schaffer (2006;emphasis in original), p. 87.“I started with a puzzle: how can it be, when his conclusion is so silly, that the sceptic’s

argument is so irresistible? My Rule of Attention, and the version of the proviso that madethat Rule trivial, were built to explain how the sceptic manages to sway us—why hisargument seems irresistible, however temporarily. If you continue to find it eminentlyresistible in all contexts, you have no need of any such explanation. We just disagree aboutthe explanandum phenomenon.” Lewis (1996), p. 561.

8. If one does not like calling the intuitions themselves the evidence, then let this be the set ofpropositions whose status as prima facie part of one’s evidence is secured via intuition. Thesedifferent views of intuitions and evidence are not in any way interestingly distinguishable onthe sorts of methodological matters under consideration here.

9. See Murphy (2008).10. If someone did have these sorts of motivations, though, then everything that I’ve said so far

in this chapter would not apply to them: clearly the use of experimental observations in ourphilosophizing would not be appropriate to such a project of epistemic re-founding.

11. Though of course it may show that something is not necessary, by showing that its negationis actual, and hence possible; and for that matter, an argument that has a crucial premise thatis merely possible, cannot in general thereby properly show us anything about the actualworld.


12. After all, Kant himself distinguished between “unlimited universality” and “comparativeuniversality,” accepting that experience can ground the latter sort of claims, even “empiricallaws.”

13. See Casullo (2003), pp.158–60.14. Thus experimental epistemologists of this sort would be practicing a different variety of

epistemic naturalism than that exemplified in Kornblith (2003).15. Or, at most, counterfactual or nomologically necessary, as discussed previously. I will omit

this more precise formulation throughout.16. See also arguments with a similar tone in Williamson (2007); Jenkins (2008).

ReferencesBonJour, Laurence. 1998. In Defense of Pure Reason. London: Cambridge University Press.Casullo, Albert. 2003. A Priori Justification. Oxford: Oxford University Press.Hawthorne, John. 2002. Deeply Contingent A Priori Knowledge. Philosophy and Phenomeno-logical Research, 65, 247–69.

Horgan, Terry and David Henderson. 2001. The A Priori Isn’t All That It Is Cracked Up to Be,But It Is Something. Philosophical Topics, 29, 219–50.

Jackson, Frank. 1998. From Metaphysics to Ethics: A Defence of Conceptual Analysis. Oxford: OxfordUniversity Press.

Jenkins, Carrie. 2008. Modal Knowledge, Counterfactual Knowledge, and the Role of Experi-ence. The Philosophical Quarterly, 58, 693–701.

Kitcher, Paul. 1980. A Priori Knowledge. The Philosophical Review, 89, 3–23.Kitcher, Paul. 1988. Mathematical Naturalism. In Aspray and Kitcher, eds,History and Philosophyof Modern Mathematics, Minnesota Studies in the Philosophy of Science XI, Minneapolis:University of Minnesota Press. 293–325.

Kornblith, Hilary. 2003. Knowledge and its Place in Nature. Oxford: Oxford University Press.Lewis, David. 1996. Elusive Knowledge. Australasian Journal of Philosophy, 74, 549–67.Lombrozo, Tania. 2010. Causal-Explanatory Pluralism: How Intentions, Functions, and Mech-anisms Influence Causal Ascriptions. Cognitive Psychology, 61, 303–32.

Manfredi, Pat. 2000. The Compatibility of A Priori Knowledge and Empirical Defeasibility:A Defense of a Modest A Priori. Southern Journal of Philosophy, 38, 159–77.

Murphy, Peter. 2008. Rewriting the A Priori/A Posteriori Distinction. Journal of PhilosophicalResearch, 33, 279–84.

Nagel, Jennifer. Forthcoming. Intuitions and Experiments. To appear in Philosophy and Phenom-enological Research.

Nichols, Shaun and Joshua Knobe. 2008. An Experimental Philosophy Manifesto. In Nicholsand Knobe, eds, Experimental Philosophy, Oxford: Oxford University Press, 3–14.

Pinillos, Angel. 2012. Knowledge, Experiments, and Practical Interests. In Brown and Gerken,eds, Knowledge Ascriptions. Oxford: Oxford University Press, 192–219.

Schaffer, Jonathan. 2006. The Irrelevance of the Subject: Against Subject-Sensitive Invariantism.Philosophical Studies, 127, 87–107.

Stich, Stephen. Forthcoming. Do Different Groups Have Different Epistemic Intuitions? Toappear in Philosophy and Phenomenological Research.


Summerfield, Donna. 1991. Modest A Priori Knowledge. Philosophy and PhenomenologicalResearch, 51, 39–66.

Weinberg, Jonathan, Shaun Nichols, and Stephen Stich. 2001. Normativity and EpistemicIntuitions. Philosophical Topics, 29, 429–60.

Williamson, Timothy. 2007. The Philosophy of Philosophy. Oxford: Blackwell.


II

The Nature and Scope of theA Priori

5

On the Armchair Justification ofConceptually Grounded NecessaryTruths

David Henderson and Terry Horgan

In this chapter we provide an account of a form of armchair reflection that we supposeshould strike philosophers as recognizable, even familiar. It is a form of reflection thatprovides a low-grade a priori form of justification for certain claims which, if true at all,are necessarily true and furthermore are rendered necessarily true solely by virtue oftheir constituent concepts, independently of any contingent facts about our actualworld. (Sentences expressing such truths are thus analytic—i.e., they are true solelyby virtue of meaning.) We explain why the justification afforded is aptly understood asa priori, yet of a low-grade sort that has an ineliminable empirical dimension. Thejustification comes by way of cognitive processes that are not themselves special eitherto philosophy or to a priori justification. The conceptual competence that folk possessfor the non-deferential application of concepts features prominently, as does thegeneral capacity for abductive inference. On our account, the relevant reflection(common in philosophy, but often misdescribed) has three central stages. In the firststage, one generates (in the armchair) data for subsequent reflection—data that typicallytakes the form of confident first-person judgments as to the applicability or non-applicability of some concept to some specific, concrete, hypothetical scenario (e.g.,the concept knowledge vis-a-vis some instance of justified true belief, or the conceptwater vis-a-vis some Twin Earth scenario). In the second stage, one typically advances ahypothesis that is general in scope and is rather abstract, as a candidate for the status ofbeing a conceptually grounded necessary truth (e.g., the hypothesis that ‘S knows thatp’ is true just in case S believes that p, S is justified in believing p, and p is true). Thishypothesis is advanced as being prima facie abductively well supported by the judg-ments generated in the first stage—i.e., prima facie supported via “inference to the bestexplanation.” In the third stage, one engages in wide reflective “equilibration”: oneseeks out a good mesh between stage-one data and stage-two hypothesis—a processthat sometimes involves rejecting some of the stage-one judgments as mistaken and

proposing an explanation of why they seemed initially plausible, and that sometimesinstead involves rejecting the candidate hypothesis on the grounds that the items inone’s initial stock of stage-one judgments were not an adequate sample (e.g., rejectingthe justified-true-belief analysis of knowledge because it does not accommodatecertain cases one initially did not consider, namely, Gettier cases).

The plan of the chapter is as follows. In Section 1, we discuss armchair methodologyin linguistics, as a useful model for armchair philosophical reasoning. In Section 2, weelaborate on our conception of low-grade a priori reasoning in philosophy, in a waythat emphasizes some key respects in which such reasoning incorporates empiricalconsiderations. In Section 3, we illustrate low-grade a priori reasoning in action. Wediscuss a series of scenarios concerning a much-discussed concept (namely, the conceptwater), and we use these scenarios to argue for two hypotheses, each of which is apt toseem somewhat surprising in the current philosophical climate: first, metaphysicallynecessary truths that are semantically non-analytic and epistemologically a posteriori—e.g., “Water is composed of H20”—are underwritten by yet-more-fundamentalnecessary truths that are analytic; and second, it is a conceptually grounded necessarytruth that some statements expressing epistemic possibilities do not express metaphys-ical possibilities. In Section 4, we situate our conception of armchair reasoning inphilosophy in relation to two other conceptions—on the one hand, that of TimothyWilliamson, whose construal of such reasoning is less traditional than ours, and on theother hand, that of those philosophers who continue to deploy a more traditionalunderstanding of the a priori. We argue that our own conception is preferable to eachof these alternatives.

1 Armchair Methodology in LinguisticsA three-stage epistemic process is discernible in the methodology typically employedby linguists in constructing and evaluating theories of natural-language syntax. At thefirst stage, data for syntactic theory is generated in the form of judgments concerningthe grammaticality or ungrammaticality of various sentence-like strings, and concern-ing grammatical ambiguity or nonambiguity of various sentences. Such data aboutone’s own language is available from the armchair, in the form of introspectivelyaccessible, confidently held, first-person judgments about grammaticality. Althoughthese judgments presumably are generated in a manner consistent with the generalrules of syntax—so that the competent speaker possesses an implicit mastery of thoserules, whatever they are—grammatical competence does not generate explicit beliefswhose contents are the syntactic rules themselves.

At the second stage, the linguist proposes certain abstract general syntactic principles,hypothesizing that they are the rules of grammar for a given language. These proposalsare not themselves the direct deliverances of the linguist’s cognitive mechanisms ofgrammatical competence. Rather, the proposed rules of syntax are theoretical hypoth-eses about the language. They are put forth, tentatively, as candidates for justification

112 DAVID HENDERSON AND TERRY HORGAN

via abduction—i.e., justification via “inference to the best explanation” of pertinentdata (including the initial stage-one data).At the third stage, the explanatory power of the tentative hypotheses is further tested

in various ways. For instance, predictions are generated from the hypothesized rulesabout the grammaticality or ungrammaticality of various sentence-like strings that werenot in the original stage-one corpus, and the linguist then consults first-person gram-matical intuition as to whether or not such newly generated word-strings seem to havethe status attributed by the proposed rules. If apparent conflicts arise between generalhypothesis and concrete intuition—e.g., because the rules predict that the word-string‘Dogs dogs dog dog dogs’ is grammatical, whereas initially it intuitively seems ungram-matical—then the linguist considers whether the initially intuitive judgment mightplausibly be explained away as a performance error, a lapse in conceptual competence.(In the example at hand, it is well known among psychologists that center-embeddedgrammatical sentences are difficult for humans to process, as in the parallel sentence‘Cats dogs chase eat mice’. Also, the given word-string comes to seem intuitivelygrammatical when one bears in mind a suitable paraphrase, as in ‘Dogs who arevigorously pursued by other dogs are themselves vigorous pursuers of other dogs’.)Overall, the goal at stage three is to bring hypotheses and data (including new datagenerated in light of the hypotheses) into harmony, in such a way that the hypothesesend up doing well at explaining the data. The upshot, when all goes well, is that thehypotheses end up being epistemically well warranted abductively vis-a-vis the data,via inference to the best explanation.As is virtually always the case with inference to the best explanation, the evidential

considerations involved, in proffering the proposed rules of syntax as the putatively bestexplanation of the data, operate via wide reflective equilibrium. Considerations of fitvis-a-vis wider theory, and with various kinds of facts less directly connected to thematter at hand, are potentially relevant. Relevant considerations can include, forexample, the grammaticality intuitions of other people. They can also include theoriesand results in cognitive science concerning natural language processing, since ultim-ately an adequate formulation of the rules of syntax would have to mesh with a detailedcognitive-scientific account of how those rules get accommodated, either implicitly orexplicitly, by human language-processing mechanisms. Another important dimensionof the linguist’s wide-reflective-equilibriummethodology is the availability of plausiblesupplementary hypotheses for ‘explaining away’ any recalcitrant data—in particular,recalcitrant grammaticality judgments, or judgment-tendencies, that do not conformwith the proposed syntactic rules.The linguist’s armchair grammaticality-judgments operate epistemically under two

different guises, in the course of the three-stage abductive reasoning. When initiallyproduced, they are defeasibly presumed correct—as indeed they are correct, insofar asthey are the direct products of the linguist’s own grammatical competence.1 The contentsof the judgments—rather than empirical fact of their occurrence—are the data-points,

ON THE ARMCHAIR JUST IF ICATION 113

when the judgments are initially produced. (This is so in stage one, and also whenintuitive judgments about newly produced word-strings are initially made in stagethree.) But in the course of the overall abduction, when evaluating the explanatorystrength of proffered grammatical rules, the judgment-states also figure qua psychologicaloccurrences, which are empirical phenomena themselves. The linguist, when regardingthe judgments under the guise of empirical phenomena, makes the empirical assump-tion that most of them are the direct products of the linguist’s own grammaticalcompetence. (Some may be mistaken, however, and may turn out to be debunkinglyexplainable as performance errors—as in the case of the grammatical sentence ‘Dogsdogs dog dog dogs’.) So the judgment-states, qua empirical phenomena, become datafor an overall theory that must provide theoretical explanations of their status and theirsource. This theory will necessarily have dual dimensions—one regarding the rules ofthe grammar for the language, and the other regarding the nature and extent of theinformant’s (linguist’s) own grammatical competence. Accordingly, the overall three-stage reasoning process clearly constitutes an empirical abductive inference; it is not itselfeither the direct product of the linguist’s grammatical competence, or the product ofabductive reasoning that lacks empirical aspects.

Empirical factors and empirical assumptions of other kinds typically enter in at stagethree as well, over and above concrete grammaticality-judgments under the guise ofempirical occurrences. One is the issue of the availability and independent plausibilityof psychological debunking explanations of initial grammaticality-judgments thatconflict with the proposed rules. Another is the empirical assumption that the gram-maticality-judgments so far deployed are suitably representative of the aspects of Englishsyntax that the linguist seeks to explain via the proposed rules of grammar—therepresentativeness assumption, as we will call it. (If there are grammatical (or ungrammat-ical) strings of English words that the proposed rules would count as ungrammatical(grammatical)—word-strings possessing features not possessed by any of those thatthe linguist has formed judgments about—then the linguist’s data-set fails to satisfy therepresentativeness assumption.) Yet another is the empirical assumption that theproposed rules are psychologically real, in the sense that the psychological mechanismssubserving the grammatical competence of English speakers conform to those veryrules. (Such conformity could be implicit, however—i.e., not a matter of explicitlyrepresenting and consulting the rules unconsciously, but rather a matter of possessinglinguistic dispositions that accord with the rules.)

It bears emphasis that the three stages need not be temporally exclusive or tempor-ally sequential. They are best regarded as three abstractable aspects of the overall process ofabductive reasoning concerning rules of natural-language syntax. In practice, thesethree aspects often will be intimately intertwined temporally, in the ongoing process ofseeking out potential grammatical rules, garnering concrete grammaticality judgmentsto test those rules, weeding out likely performance errors, and in general looking for agood fit between theory and data.


2 Low-Grade A Priori ReasoningThe reflective investigation appropriate to many philosophically interesting and import-ant issues is best understood as having a three-stage structure that parallels the linguist’sreflective investigations of grammar. As in the case of linguistic investigation, philosoph-ical reflection draws on the scattered concrete direct deliverances of a cognitive compe-tence—conceptual competence, the analogue of grammatical competence. It seeks toarrive at conceptually grounded necessary truths that are general in scope—the analoguesof the rules of grammar the linguist seeks to discover. Furthermore, the relevant concep-tually grounded necessary truths are not reflectively generated in a direct way out of one’sconceptual competence; so the ensuing reflection will need to employ the more particu-lar concrete deliverances of conceptual competence as data for abductive inference. Thisreflective procedure has both psychological and conceptual faces, like investigations ofgrammar, and it is properly understood as having an empirical dimension.Although the linguistics analogy is helpful and suggestive, the kinds of armchair-

obtainable data that are pertinent to philosophical conceptual-analytic inquiry appearto be fairly diverse (more so than in the linguistics case), with some kinds being moredirectly analogous to grammaticality judgments than others. The types of data that canfigure in philosophical conceptual-analytic reflection include the following:

1. Intuitive judgments about what it is correct to say concerning various concretescenarios, actual or hypothetical.

2. Facts about conflicting judgments or judgment-tendencies, concerning the cor-rect use of certain concepts in various actual or hypothetical scenarios.

3. Facts about standardly employed warrant-criteria for the use of various concepts.4. Facts about the key purposes served by various terms and concepts.5. General background knowledge, including untendentious scientific knowledge.

Data of all these kinds can go into the hopper of wide reflective equilibrium wherebyconceptual-analytic claims are defended in philosophy. One makes a case for a certainconceptual-analytic hypothesis—for instance, the contention that the meaning ofnatural-kind terms depends on the language-users’ environment—by arguing that itdoes a better job, all things considered, of accommodating the relevant data than doany competing hypotheses. Such reasoning, we maintain, is broadly empirical: infer-ence to the best explanation, in which data of all the kinds 1–5 are potentially relevantand empirical elements of various kinds figure in various ways. Although it can beconducted from the armchair and it is aimed at discovering conceptually groundednecessary truths, it does not fully conform to the traditional conception of a priorireasoning. According to the traditional conception, a priori reasoning is uncontamin-ated by empirical considerations; this we call the “high-grade” a priori. By contrast,empirically informed armchair-abduction aimed at discerning conceptually groundednecessary truths is what we call the “low-grade” a priori.

ON THE ARMCHAIR JUST IF ICATION 115

While philosophical reflection is not devoid of empirical constraints—and thus wouldnot qualify as a priori in the high-grade form of much philosophical lore—it is nonethe-less a quite distinctive form of inquiry. (It is not distinctive in being deployed only byphilosophers, or by being all of philosophy.Rather, it is distinctive in that it draws heavilyon armchair soundings of conceptual competence, and it seeks to arrive at conceptuallygrounded necessary truths by inference from armchair-available data.) Such conceptual-competence-based reflection seems to us to be well characterized as a kind of a priorireasoning. It bears many of the marks commonly associated with the a priori—namely,aimed at discerning conceptually grounded necessary truths, practiced in the armchair,resulting from reflection, with limited reliance of empirical information.However, as wereadily note, it is not utterly free of reliance on empirical information. Thus we write oflow-grade a priori justification. Philosophers often havemisconstrued this form of inquiry:they have mistakenly claimed that empirical considerations do not enter into it.

Not only is the range of considerations operative here more diverse than thoseoperative in linguistic study of grammar, but what is to be sorted out is itself morediverse or multi-faceted. The grammarian seeks to sort out grammatical strings fromnon-grammatical ones (and also to identify grammatical ambiguities), while also sortingout competent performances, performance errors, and the character of one’s capacityto distinguish these. The philosopher seeks to distinguish which claims belong towhich of various epistemic and semantic statuses. Important and uncontroversialcontingent truths must be distinguished from metaphysically necessary truths—andin the course of doing this, it will be important to identify certain truths that aremetaphysically necessary without being conceptually grounded necessary truths (i.e.,without being analytic). That water is H2O, for example, is not merely an uncontro-versial empirical fact, it is metaphysically necessary.

Several points need to be borne well in mind concerning the nature of philosophicalreflection of the kind we are calling low-grade a priori. First, a prominent way ofdrawing upon one’s conceptual competence, conspicuously manifested in much well-received philosophical work, involves the generation of judgments about relativelyspecific and concrete scenarios. It is reasonable to suppose that human conceptualcompetence is particularly suited to the generation of responses to such concrete specificscenarios—and that it is at its most sure-footed here. After all, in everyday contexts,where conceptual competence serves in a largely unnoticed way, it would seem tofunction largely in the direct and automatic generation of applications of concepts.

Second, conceptual competence is much less steady and reliable when called upon todirectly generate conceptually grounded general truths. Perhaps conceptual competencecan generate general truths with reasonable success when they concern certain conceptsthat do not feature in philosophically interesting claims. But the track record ofphilosophers, particularly with respect to those concepts that have proved to be ofphilosophical moment, suggests that one’s conceptual competence may be much betterat directly delivering presumably veridical judgments regarding specific applications ofthose concepts than it is at directly generating veridical judgments regarding generalities.


Third, by drawing upon what conceptual competence does provide, one can manageto justifiably believe certain conceptually grounded generalities that are not themselvesthe direct deliverances of that competence. This is where abduction enters in, with thethree-stage structure lately described.Fourth, data of type 2 in the above list—namely, facts about conflicting judgments

or judgment-tendencies, concerning the correct use of certain concepts in variousactual or hypothetical scenarios—will often figure quite prominently in low-grade apriori reflection. After all, numerous philosophical problems involve certain intuitivelyfelt tensions that reflect such internal conflicts. (One familiar example, amongmany thatcould be cited, is the tension between the tendency to embrace a variety of everydayknowledge-claims and a tendency to feel the pull of radical external-world skepticismwhen contemplating hypothetical radical-deception scenarios.) Because this kind ofdata figures so importantly, such reflection is apt to incorporate proposed psychologicaldebunking-explanations fairly prominently—certainly more prominently than in thecase of a linguist’s armchair syntactic theorizing. The fact that low-grade a priorireflection will often rely fairly heavily on plausible debunking explanations, explan-ations that are suitably respectful of intuitive judgments and judgment-tendencies(rather than treating them as silly blunders), is a particularly striking way that thisform of abductive reasoning becomes empirically infected. (One familiar example,amongmany that could be cited, is the way that certain contextualists about the conceptof knowledge propose to respectfully debunk the tendency to feel the pull of radicalexternal-world skepticism: the claim is that one fails to notice (as is easy to do) thatthe concept of knowledge is governed by implicit contextual parameters, and that theposing of radical-deception scenarios tends to create a context where those parametersbecome vastly more demanding than usual.)Fifth, a question arises concerning the scope of the category ‘low-grade a priori’,

vis-a-vis the various aspects of philosophical armchair reflection. Which kinds ofconclusions that philosophers come to in their armchairs should be included underthis rubric, and why? (For instance, should psychological side-explanations of thedebunking kind count as low-grade a priori themselves, or not?) In our view this ispartly a terminological issue once one eschews the traditional high-grade conception ofthe a priori as a form of reasoning that is uncontaminated by the empirical. However,in order to retain a sufficiently robust conceptual link to the notion of the a priori astraditionally understood, we hereby stipulate that, strictly speaking, a piece ofreasoning only counts as low-grade a priori if it is an abductive argument, conductablein the armchair, in support of a hypothesis of the form “It is conceptually necessary thatP.” (Note that on this usage, an overall abductive argument that low-grade a prioriitself can appeal to the plausibility of certain considerations—e.g., certain psychologicaldebunking-explanations—that are not low-grade a priori themselves. Also, in somecases the modality of conceptual necessity may well be implicitly attributed in context,rather than being built explicitly into the articulated hypothesis that one embraces vialow-grade a priori abduction.)

ON THE ARMCHAIR JUSTIF ICATION 117

3 A Case Study with Some Striking MoralsOne goal in this section is to illustrate low-grade a priori reasoning in action, focusing on avery familiar example: the case of water. Another goal is to argue, in a low-grade a prioriway, for twomorals that are both apt to seem striking in the present philosophical climate:first, that certain facts about non-analytic necessity (e.g., the fact that it is metaphysicallynecessary that water = H20) are explainable in a way that adverts to more basic factsabout what is conceptually necessary; and second, that certain epistemic possibilities (e.g.,the epistemic possibility, given only what was known about water in 1750, that water ischemically simple) are not metaphysical possibilities. The first of these conclusions isstriking because it goes contrary to a currently widespread belief that in philosophy, non-analytic metaphysically necessary truths have largely displaced conceptually groundednecessary truths. The second conclusion is striking because it goes contrary to thecurrently widespread belief that epistemic modality is a species of metaphysical modality.

To begin with, let us consider some scenarios that we envision as eliciting particu-larly good data. There are several sorts of scenarios that we think are illustrative anduseful. Here is our variation on a sort of scenario that is familiar from Putnam (1975)and Kripke (1980): (Kripke 1980):

S1: Somewhere there is a community of language using agents inhabiting anenvironment (one might think of this environment as a planet). The peoplethere interact with a liquid substance that they take to be homogenous in itscharacter. (They may or may not have a theory about what features ultimatelymake for the homogeneity of a substance—but, if they do, they treat thistheory as revisable in light of further inquiry.) Their interaction with this stuffis characterized by an ability to identify (undisguised or relatively straightfor-ward) instances of it,2 and they adopt a term intended to refer to this substance,which is indeed of a homogeneous sort. The stuff in their samples, the instanceswith which they have interacted, is in the vast majority of cases H2O.

We will call scenarios such as S1, focused scenarios. Focused scenarios are descriptions ofepistemic possibilities that focus on an interaction of a community of language userswith a stuff, thing, or entity within their environment. S1 has an interesting feature: itdoes not itself use or refer to our concept of water, although it seems highly plausiblethat, in responding to it, one automatically draws of one’s sense for how concepts suchas water work. Thus, one draws on one’s conceptual competence with water.

In exploring the verdicts elicited by S1, one important question would be, “Wouldtheir concept refer to H2O?” This provokes the judgment or verdict that this conceptrefers to H2O.

One might explore how strongly one associates the scenario with our concept ofwater by posing the question: How does the concept that they form in S1 relate to ourconcept of water? We think that many would be likely to be inclined to say that theseare the same concept. This inclination could be strengthened by considering furtherscenarios. Here is one of special interest:


S2: Add to S1: They happen to choose the word ‘water’ to name the stuff andexpress the concept in S1. Also, elsewhere in their extended environment(perhaps on a nearby planet, or on a far continent), there is some stuff that theyhave not previously encountered. This stuff has surface properties that are verysimilar to the stuff with which they had been interacting when forming theirconcept expressed by ‘water’. However, this stuff has a very different under-lying character than the stuff involved in S1—it is composed of XYZ ratherthan H2O.

Now ask: Does their concept (that expressed by the word ‘water’ in their language)apply to this stuff—to XYZ? One tends to judge that it does not. Their concept refersto H2O and would not be satisfied by XYZ. If they have associated the word ‘water’with their concept arising by the interactions characterized in S1, then one judges thatit would be a mistake for them to use ‘water’ to refer to this stuff.It is worth noting that there are closely related questions to be asked about the

application of one’s own concepts to the stuffs featured in S1 and S2. To the question ofwhether it would be correct to say that the stuff at issue in S1 is water, one will judge thatit is. To the question of whether it would be correct to say that the stuff at issue in S2 iswater, one will likely say that it is not. Of course these judgments, and the ones made inconnection with S2 famously suggest that one’s concept of water is a rigid designator—and that one is thinking of the concept involved in S2 as working similarly.Consider now a different focused scenario:

S3: Somewhere there is a community of language-using agents inhabiting anenvironment (one might think of this environment as a planet). The peoplethere interact with a liquid substance that they take to be homogenous in itscharacter. (They may or may not have a theory about what features ultimatelymake for the homogeneity of a substance—but, if they do, they treat thistheory as revisable in light of further inquiry.) Their interaction with this stuffis characterized by an ability to identify (undisguised or relatively straightfor-ward) instances of it, and they adopt a term intended to refer to this substance,which is indeed of a homogeneous sort. The stuff in their samples has severalsignificant surface features in common with the stuff that we call water—however, there is no H2O in their immediate environment, so they havenever interacted with any H2O. In the vast majority of cases the stuff withwhich they have interacted is homogeneous XYZ.

Prompted by queries parallel to those mentioned, one would find that one is inclinedto judge that the concept here applies to XYZ. Further, one judges that their term‘water’ associated with the concept arising in this focused scenario applies to XYZ. Atthe same time, one judges that that stuff—XYZ—is not water. This pattern ofjudgments, prompted by the focused scenarios S1 and S3, strongly suggests that oneis thinking of their concept (in S3) as having its referent fixed by the sorts of interaction


featuring in these scenarios (S1 and S3)—they indicate that the referent of the conceptsis determined by the nature of the stuff with which the folk interact in coming to formtheir concept.3

Of course one might usefully think of a focused scenario that extends S3 in the waythat S2 elaborated S1:

S4: Add to S3: They happen to choose the word ‘water’ to name the stuff andexpress the concept in S1. Also, elsewhere in their extended environment,there is some stuff that they had not previously encountered. This stuff hassurface properties that are very similar to the stuff with which they had beeninteracting when forming their concept expressed by ‘water.’ However, thisstuff has a very different underlying character than the stuff involved in S3—itis composed of H2O rather than XYZ.

Does their concept (that expressed by the word ‘water’ in their language) apply to thisstuff—to H2O? One judges that it does not. Their concept refers to XYZ and wouldnot be satisfied by H2O. If they have associated the word ‘water’ with their conceptarising by the interactions characterized in S3, then one judges that it would be amistake for them to use ‘water’ to refer to this stuff. Even so, using our concept andEnglish, we note: it is water.

Intuitive verdicts like those elicited by the focused scenarios S1–S4 are examples ofthe kind of data that one garners, in the armchair, at stage one in the process of low-grade abduction. Such data suggests certain hypotheses, and setting forth these hypoth-eses constitutes stage two. Here are some plausible candidates inspired by the work ofKripke and Putnam, formulated in a way that explicitly builds in the modality ofconceptual necessity:

H1: It is conceptually necessary that for any substance s, if s is a natural kind and sis the referent of the concept water, then it is metaphysically necessary thatwater = s.

H2: It is conceptually necessary that for any substance s, if s is a natural kind and s isthe referent of the concept water, then water rigidly designates s.4

Note that H2, if true, directly explains H1: the reason why it is metaphysicallynecessary that water is identical to substance s (whatever natural kind that substancehappens to be) is that the concept water is a rigid designator.

Of course there is presumably nothing special or idiosyncratic about the conceptwater, as regards the potential for figuring in focused scenarios like S1–S4 and yieldingcomparable intuitive judgments. On the contrary, one can say much the same thing fornatural-kind concepts in general, and for names in general—as figures prominently inKripke’s landmark text Naming and Necessity. This fact yields analogues of hypothesesH1 and H2 for natural-kind terms generally, and for names generally.


Stage three can now commence: inquiring about the explanatory power of thestage-two hypotheses H1 and H2 (and their analogues). For starters, these twohypotheses fit well with the initial stage-one data: if H1 and H2 are true, then theirtruth entails the truth of the intuitive verdicts that were elicited by scenarios S1–S4. Inaddition, H1 and H2 entail the truth of certain further types of claims about concretescenarios like S1–S4. Some such claims are explicitly modal, like these:

(1) The substance referred to by the folks in scenario S1 could not fail to be water.(2) The substance referred to by the folks in scenario S2 could not be water.

Other such claims are explicitly semantic-conceptual, like these:

(3) Given that water is composed of H2O and that H2O is a natural kind, it wouldbe a violation of the meaning of the English word ‘water’ (and of the conceptthis word expresses) to deny that the substance referred to by the folks inscenario S1 is water.

(4) Given that water is composed of H2O and that H2O is a natural kind, it wouldbe a violation of the meaning of the English word ‘water’ (and of the conceptthis word expresses) to affirm that the substance referred to by the folks inscenario S2 is water.

So at stage three, one again consults one’s intuitions as a competent user of the conceptwater and the English word ‘water’: one asks oneself whether concrete-scenarioinvolving claims like (1)–(4) seem correct (and obviously so). Indeed they do, foreach of us co-authors—and we presume for you the reader, as well. This happyresult constitutes yet more data in support of the proposed hypotheses H1 and H2,over and above the data that constituted the scenario-judgments we started with.Thus far, then, H1 and H2 are faring well abductively. (Likewise, mutatis mutandis,for analogues of H1 and H2 regarding natural-kind terms in general and namesin general.)What about recalcitrant intuitive judgments or judgment-tendencies that do not

conform well with hypotheses H1 and H2, and that might therefore be subject to (dulyrespectful) debunking explanations? Here one does well to remember that before thepioneering work of Kripke and Putnam, it was very widely believed that identityclaims like ‘Water is H2O’ and ‘Hesperus is Phosphorus’ are contingent, not necessary.How might this belief be respectfully debunked? Here are some relevant remarks byKripke himself, from his Naming and Necessity:

There is a very strong feeling that leads one to think that, if you can’t know something by a prioriratiocination, then it’s got to be contingent: it might have turned out otherwise; but neverthelessI think this feeling is wrong . . . . [I]n a counterfactual world in which ‘Hesperus’ and ‘Phos-phorus’were not used in the way that we use them, as names of this planet, but as names of someother objects, one could have had qualitatively identical evidence and concluded that ‘Hesperus’and ‘Phosphorus’ named two different objects. But we, using the names as we do right now, can


say in advance that if Hesperus and Phosphorus are one and the same, then in no other possibleworld can they be different. (pp. 103–4)

And later in the same text Kripke adds this:

The inaccurate statement that Hesperus might have turned out not to be Phosphorus would bereplaced by the true contingency mentioned earlier in these lectures: two distinct bodies mighthave occupied in the morning and evening, respectively the very positions actually occupied byHesperus-Phosphorus-Venus. (pp. 104–5)

These remarks suggest a potential debunking explanation of the erstwhile intuitivebelief that identity statements like ‘Water is H2O’ and ‘Hesperus is Phosphorus’ areonly contingently true, which goes as follows. When one takes oneself to be envision-ing a metaphysically possible world in which Hesperus fails to be identical to Phos-phorus, one is envisioning a genuine metaphysically possible world but one ismisdescribing it in a subtle way: it is not a world in which Hesperus fails to be identicalto Phosphorus, but rather is a world very much like the actual world except that (a) twodistinct planetary bodies respectively occupy the morning and evening positions thatVenus actually occupies, and (b) the words ‘Phosphorus’ and ‘Hesperus’ respectivelyname those two bodies.

Does this approach adequately explain away intuitions of contingency that seem toconflict with hypotheses H1 and H2? In our view it does provide a partially adequatedebunking explanation, but not a complete one. Consider the following focusedscenario, which one might think of as reflecting the epistemic situation of folk prior tothe developments in chemistry associated with the discovery of oxygen (say that of folkin 1750, perhaps).5

S5: Our earthly community of language users interact and have interacted with aliquid substance that they take to be homogenous in its character, the sub-stance that English speakers call water. (They are at this point still seeking anunderstanding of what features ultimately make for the homogeneity of asubstance—some may think it is a simple, others may think it is homoge-neously composed of some homogeneous combination of other substances/elements/principles.) Their interaction with this stuff is characterized by anability to identify (undisguised or relatively straightforward) instances of it, and ineach of their languages they adopt a term (in English, ‘water’) intended to refer tothis substance, which is indeed of a homogeneous sort. The stuff in their samples,the instances with which they have interacted, is in the vast majority of cases H2O.

Now ask this question: Is it true, given the epistemic situation of the humans in thisfocused scenario, that water—that very stuff, rather than some qualitatively indistin-guishable stuff—could turn out to be some substance distinct from H2O? Our ownintuitive verdict is that the answer to this question is affirmative.


To further underscore this sort of intuitive verdict, consider the following. Supposethat one knows that Sam and Dave are related by the father-of (and thus by the son-of)relation, but that one does not have any information bearing on which is the father andwhich is the son. Suppose too that one sees two guys in the distance, both facing awayfrom oneself, and one knows that the guy on the left is Sam and the guy on the right isDave, and yet one’s view of them provides no evidence at all as to which one is thefather and which one the son. Now ask these two questions. First, could it turn out thatSam—that very guy on the left—is the father of Dave—that very guy on the right? Second,could it turn out that Sam—that very guy on the left—is the son of Dave—that very guy onthe right? Our own intuitive verdict is the answer to both questions is affirmative.(Nonetheless, we both find very plausible Kripke’s doctrine of the necessity of origin.)

Now, it seems clear that the intuitive, modal, concrete-scenario judgments latelynoted—on the one hand about how the very stuff that we humans call water couldhave turned out not to be H2O, and on the other hand that Sam could have turned outto be Dave’s father and that Dave could have turned out to be Sam’s father—are notamenable to the kind of debunking explanation that Kripke invokes. Suppose, forinstance, that (unbeknownst to the viewer who is contemplating Sam on the left andDave on the right) Sam is Dave’s father. When the viewer judges that Dave could turnout be Sam’s father, she not contemplating a metaphysically possible world in whichthe two guys are in switched positions (Dave on the left and Sam on the right) withswitched names (Dave named Sam and Sam named Dave) and then judging that for allshe knows, that metaphysically possible world is the actual world. No: she is contem-plating the possibility that Dave—that very guy she is now looking at there on theright—is the father of Sam—that very guy she is now looking at there on the left; andshe is judging that for all she knows, this possibility is actual. Likewise, mutatis mutandis,for one’s intuitive judgments concerning the focused scenario S5.On the other hand, it is not difficult to find a plausible way to treat these intuitive

scenario-verdicts that reveals the lack of any real conflict between them on one hand,and the hypotheses H1 and H2 on the other. One simply points out the following.First, H1 and H2 are conceptual-necessity claims pertaining to metaphysical possibility(and to how the reference of the concept water works across metaphysically possibleworlds), whereas the intuitive scenario-judgments now under consideration pertain toepistemic possibility. Second, some epistemic possibilities are just not metaphysicalpossibilities—e.g., (given that Sam is Dave’s father) the possibility that Dave is Sam’sfather. What gets debunked here is not the intuitive possibility-judgments themselves,but rather the thought (or suspicion, or tendency toward thinking) that these modaljudgments conflict with hypotheses H1 and H2. They don’t, because they concernepistemic possibilities some of which are not metaphysical possibilities at all. To be sure,this explanation requires embracing the claim that some epistemic possibilities are notmetaphysical possibilities; but that claim is enormously plausible on its face, as cases likethat of Sam and Dave make clear.


At any time, folk confront a number of open epistemic possibilities—some of theseare metaphysical necessities, others are epistemic possibilities that are metaphysicalimpossibilities. Judgments regarding how one would/should respond were variousepistemic possibilities to turn out to be actual, can be important data for reflection.Such data is richer than data that tracks merely metaphysical possibilities. Reflection onthis wider set of data, on what would have been right to say, provides informationabout aspects of concepts that would not be evident otherwise. In effect, S5 representsa variant on S3 that involves our concept water and that reflects the epistemic possibil-ities faced by folk prior to something like 1750. Such data can be particularly revealingregarding the kinds of claims featured in H3 and H4 (see note 4).

What we have been doing in this section is illustrating the three-stage process oflow-grade a priori abduction, vis-a-vis the two hypotheses H1 and H2. (And thediscussion is applicable, mutatis mutandis, to analogous hypotheses about names andnatural-kind terms generally.) Empirical claims and empirical side-explanations haveentered in at stage three, by way of assessing the explanatory fit between H1 and H2 onone hand, and pertinent data on the other hand. One empirical claim is that the long-held belief that identity claims like ‘Water is H2O’ are metaphysically contingent ismistaken, and constitutes a conceptual performance-error rather than an exercise ofconceptual competence. This claim is supported by a plausible-looking, and alsoempirical, debunking explanation—the kind given by Kripke in the passages quotedearlier. Another empirical claim is that the statement ‘Water could turn out to besimple (and thus composed of no more basic elements)’, as made by someone in 1750or so in a context where epistemic possibility is the salient modality, would constitutean exercise of conceptual competence, and hence would be correct. An associateddebunking explanation—again empirical—concerns the temptation to think that thetruth of this statement would conflict with hypotheses H1 and H2; the explanationappeals to the empirical claim that such a thought would rest on a conflation ofmetaphysical modality and epistemic modality, and/or on a failure to appreciate thatsome genuine epistemic possibilities simply are not metaphysical possibilities at all.

Our discussion in this section also reveals that two influential lines of thought inrecent philosophy are seriously misguided. One is the idea that putative conceptuallygrounded necessary truths no longer have a central role in philosophy, allegedly havingbeen displaced by non-analytic necessary truths. Timothy Williamson puts the ideathis way:

[The word] “analytic,” like “synonymous,” once was a central term in philosophical theorizing,notably in the work of logical positivists, such as Carnap, and of postwar linguistic philosophers,such as Strawson. The reason why it cannot recover that position lies not in Quine’s critique,which no longer seems compelling, but rather in Kripke’s widely accepted clarification of thedifferences between analyticitiy, apriority, and necessity. (Williamson 2007, p. 51)

But although Kripke’s clarification of those differences is certainly welcome andimportant, the fact remains that analyticity—i.e., conceptually grounded necessary


truth—very much still earns its keep in philosophy. One way it does so, inter alia—away that is quite ironic in light of Williamson’s remarks—is by playing a pivotal role inexplainingmany non-analytic necessity claims. Here is a sample such explanation, in theform of a deductive argument:

H1: It is conceptually necessary that for any substance s, if s is a natural kind and sis the referent of the concept water, then it is metaphysically necessary thatwater = s.

Empirical non-modal fact: H2O is a natural kind and is the referent of the conceptwater.

Therefore,

It is metaphysically necessary that water = H2O.

Another influential idea in recent philosophy concerns the nature of epistemic modal-ity. There is considerable enthusiasm these days for an idea expressible roughly asfollows (leaving aside, perhaps, domains like pure mathematics). An epistemic possibil-ity, for a cognitive agent A at a spatiotemporal location L, is just a metaphysicallypossible world w—or perhaps a “centered” possible world wc with a designated centercorresponding the actual-world location L of agent A—such that A does not know(at L) that w (or wc) is non-actual.

6 If the above discussion is correct, then this idea toois quite seriously wrongheaded.7

4 Comparisons on Two FrontsThere is certainly a large role for armchair methodology in philosophy, if we are rightin claiming that one important part of philosophy is seeking out conceptuallygrounded necessary truths via low-grade a priori abduction. But there are fans ofarmchair methodology who regard it differently than we do. Here we briefly considertwo alternative conceptions, one more “left wing” than ours and the other more “rightwing,” and we argue that our conception is preferable to each of these others.On our left is Timothy Williamson, whose conception of armchair methodology in

philosophy is articulated and defended in his recent book The Philosophy of Philosophy(2007). By his lights, armchair reasoning in philosophy is essentially just exercise of one’scapacity to evaluate the truth values of counterfactual conditionals. And, as the abovequotation indicates, he repudiates any significant role for analytic truths in philosophy.He does so, not because he finds Quine’s arguments against the analytic/syntheticdistinction persuasive, but rather because he thinks that there cannot be analytic truthsthat play the role(s) that philosophers have traditionally assigned to them. FollowingBoghossian (1977), Williamson distinguishes two kinds of conceptions of analyticity:“metaphysical” conceptions and “epistemological” conceptions. The core idea behindmetaphysical conceptions is this: “analytic sentences are true simply in virtue of theirmeaning, and analytic thoughts simply in virtue of their constituent concepts” (2007,


p. 52). As regards epistemological conceptions, he says, “the central idea behind suchconceptions of analyticity is that . . . failure to assent is not merely good evidence of failureto understand; it is constitutive of such failure” (2007, p. 73).Williamson allows that there are metaphysically analytic truths, but he offers two

reasons in support of the contention that such truths are not important in philosophy.First, he argues that they cannot play the distinctive epistemological role that he claimsthey have been assigned in philosophy—namely, being “insubstantial,” in the sensethat “knowing them poses no significant cognitive challenge” (2007, p. 54). Second, inlight of Kripke’s clarifications of the differences between analyticity, apriority, andnecessity, it turns out that “‘analytic’ does neither the purely epistemological work of ‘apriori’ nor the purely metaphysical work of ‘necessary’” (2007, p. 51).

As regards epistemological conceptions of analyticity, Williamson argues that theresimply are no epistemologically analytic truths. Even paradigmatic cases of putativeepistemological analyticity, he argues, are statements that one can sensibly refrain fromassenting to even if one understands such statements perfectly well. Concerning thestatement “Every vixen is a female fox,” for instance, one could sensibly withholdassent if one believed that this statement entails that there exist vixens and one wereuncertain as to the empirical question whether there currently exist any vixens. Or onecould sensibly withhold assent because one holds that some clearly female evolutionaryancestors of foxes are borderline cases for “fox” and therefore for “vixen,” and oneembraces a three-valued logic under which the earlier existence of such creaturesrenders the statement “Every vixen is a female fox” neither true nor false.

We ourselves are happy enough to concede that very probably, there are noepistemologically analytic truths; on this matter, we find Williamson quite persuasive.But of course we do maintain that certain claims of metaphysically analytic truth are veryimportant in philosophy—for instance, hypotheses H1 and H2, and their analogues fornames in general and natural-kind terms in general. It may well be the case, asWilliamson claims, that there are no philosophically important analytic truths thatare “insubstantial” in the sense that “knowing them poses no significant cognitivechallenge,” but no matter; insubstantiality is not what makes them important. Rather,their importance lies in what they reveal about the workings of philosophicallyinteresting concepts, and in their explanatory power. (As already noted, they helpexplain, among other things, why it is that certain statements that are not analytic arenevertheless necessary rather than contingent!) It is also true, as Williamson maintains,that ‘analytic’ does not itself do the work either of ‘a priori’ or of ‘necessary’; but it doesplenty of useful philosophical work nonetheless.

We should take the space to highlight several important points informing ourresponse here—explaining why we feel unthreatened by the repudiation of epistemicanalytic truths.

Central here is this: our own account of low-grade a priori reflection—which weargue is commonly pivotal in attaining philosophical understanding or perspective—makes it clear why truths arrived at in this way would not be epistemically analytic


truths, or “insubstantial,” as these matters have been characterized. It is worth noticingjust how demandingly insubstantial epistemically analytic truths would need to be—asreflected inWilliamson’s discussion. If a claim is to satisfy the epistemological concept ofanalyticity, anyone who understands the claim—anyone who nondeferentially graspsthe concepts (and responds fittingly to their mode of combination in that claim)—would be well epistemically situated with respect to that claim. Specifically if a claim isepistemically analytic, then all who understand it would be justified in believing it.Williamson registers this satisfaction condition in what he terms “understanding-justifi-cation links.” Here is one involving the candidate thought every vixen is a female fox:

(UJt) Nec, whoever grasps the thought every vixen is a female fox is justified inassenting to it.8

He argues by plausible counter-examples that there are no such claims. Parallelconsiderations extend to cases involving logical truths—as one can find plausiblefailures of understanding–justification links involving modus ponens and excludedmiddle, even non-contradiction. In each case Williamson can point to apparentlycompetent logicians and philosophers who understand the concepts involved, andyet are, on extended careful reflection, inclined not to flatly assent. Their reflection issuch that they would not be justified in believing some general formulation of theprinciple (modus ponens or excluded middle). Williamson’s point is not to be denied:there are no necessary, universally quantified truths connecting conceptual compe-tence and understanding with assent to generalities such as those at issue here—andthere is none connecting understanding with justification.Our own account of low-grade a priori reflection makes it clear why one should

expect no such necessary universal links—and we are ready to concede that what weterm conceptually grounded necessary truths are not, strictly, epistemically analytic.Central here is what we have described as the third stage of such reflection—the pointsat which the data afforded by responses to scenarios and the working hypothesesgenerated in connection with them, together with the range of data mentioned earlier,are fed into the hopper of wide reflective equilibrium. Such abductive processing willnecessarily be sensitive to much else that one justifiably holds or justifiably finds arelevant open question. Thus it is that one who envisions truth-value gaps, or robustvagueness, or paraconsistent logics—either as justified conclusions of reflection or as atleast open possibilities needing further investigation—will rightly be moved accord-ingly in reflecting abductively about putative conceptually grounded necessary truths.In keeping with this admission, we think that there is a plausible, but less demanding

idea that one can and should honor—granting that it falls short of the full idea of strictepistemic analyticity. Instead of thinking that one “exploit[s] whatever epistemic assetswe have simply in virtue of our linguistic and conceptual competence,”9 one canexploit these assets as integral components in a wider process of armchair-abductivereflection that admittedly has an empirical dimension as well. Possession of both theepistemic assets afforded by conceptual competence, and also the capacity for such


low-grade a priori inquiry, does not entail that the intelligent use of these epistemicresources will in all cases be sufficient to get one who understands the conceptuallygrounded necessary truth to have justified belief in it. Thus, our own account oflow-grade a priori reflection allows one to see that there can be conceptually groundednecessary truths, and hence thoughts that are metaphysically analytic, and howreflection can commonly yield justified beliefs in such truths—even though thosesame truths need not be epistemically analytic. One can understand them and yet notbe justified in believing them.

We should highlight a consideration on which we are in significant agreement withWilliamson: In his theorizing about the epistemology of metaphysical modality,Williamson embraces the constraint that this epistemology turns on cognitive capacitiesand learned skills that are rather common and importantly general. The present authorsalso have never found it plausible that folk are equipped with a special capacitydistinctive of philosophical insight or intuition—a capacity for metaphysical-neces-sity-tracking intuition. Rather, we have supposed it much more likely that philosoph-ical reflection (whether done by philosophers or by scientists and other laity, as it oftenis) deploys off-the-shelf hardware and software—it deploys cognitive capacities andlearned skills that are rather common and importantly general. In our discussion here,we have emphasized two capacities that find wide application and utility in humancognitive life: the conceptual competence that enables one to apply one’s concepts tothings in one’s environment, and the capacity for abductive inference. Indeed, in thefirst case, conceptual competence, we have insisted that it be relied on largely where itis clearly the most sure-footed—in application to concrete scenarios rather than in theattempt to deliver generalities straight-off. Williamson has emphasized anothercommon capacity, the capacity for evaluating counterfactual conditionals by deployingone’s cognitive processes “off-line”—and we are happy to acknowledge that we haverelied on this one as well.10 We think that it naturally looms large in those second-stagemoments at which one generates what function as working hypotheses for abductiveevaluation.

However, not all that can be generated in the armchair by such processes asWilliamson lauds will amount to conceptual or metaphysical necessities. Of course,some such statements will be formulations of metaphysical necessities, such as:

Necessarily, were something composed of water it would be composed of H2O.

But, as one reflects on scenarios, using the tools emphasized by Williamson, one willform verdicts that suggest counterfactuals that few would suppose to be necessary truthsat all. For example, consider one’s likely response to this scenario:

S6: A given agent is 80 years old and has troubling emphysema—and this agentattempted to run a mile yesterday.

Thinking from the armchair about this case, as it were just letting one’s cognitivemachinery loose to see what pops out, suppose that one entertains the question: Did


that individual run a sub-four-minute mile? One is likely to find oneself stronglyinclined toward the verdict that he did not. This allows one to generate a counter-factual conditional:

OLD-MAN: Were someone who is 80 years old and has troubling emphysema toattempt to run a mile, that person would not manage to then run afour-minute mile.

This is pretty clearly neither a conceptually grounded necessary truth nor a metaphysic-ally necessary truth. It is a contingent truth. So one would presumably not want toproceed to:

OLD-MANnec: Were someone who is 80 years old and has troubling emphysemato attempt to run a mile, that person would not manage to thenrun a four-minute mile.

Obviously, then, if one is to identify conceptually grounded necessary truths andmetaphysical necessary truths from the armchair (as ourselves we aspire to do) or justmetaphysically necessary truths (as Williamson aspires to do), one must have a methodthat can sort verdicts and conditionals into the various categories. One must have theepistemic capacity to sort the mixed-bag judgments that are the initial results ofarmchair reflection.The reason that verdicts produced in the armchair are a mixed bag in need of sorting

is not hard to appreciate. We have discussed it elsewhere (most fully in Henderson andHorgan 2011). It is that, as one settles down into one’s armchair, it is the wholecognitive/epistemic agent occupying the armchair. Sitting in the armchair is a critterwho has acquired and retained more information than whatever makes for its concep-tual competence—and, to some non-negligible extent, it is natural for the full range ofits information to condition or shape verdicts made in the armchair. That information isnot marked off or segregated from other information within the agent. Thus, as oneposes and responds to scenarios, one will doubtless have brought to bear a range ofinformation in addition to what features in one’s conceptual competence. The types ofinformation then put to work in the armchair, and the range of cognitive processes inplay, are likely to be diverse in character. Further, as we have argued elsewhere, some ofthe information deployed in human cognitive competence—including that deployed inone’s conceptually competent application of concepts to scenarios—is very likely totake the form of what we call “morphological content”—i.e., content or informationthat rightly conditions one’s cognitive processes without needing to be explicitlyand occurrently represented within the cognitive system (Henderson and Horgan1999, 2011). A similar idea is reflected in Williamson’s discussion of implicit infor-mation that shapes a cognitive system and its processing.11 In any case, while the verdictsarise quickly, and with some resonance of reasons, it is likely that they are a somewhatdiverse lot, and a somewhat undifferentiated mixed bag.


One thus needs a reflective process by which one can sort the various armchair-generated claims into their fitting classes—and understand the classes or “boxes” intowhich these are sorted. The account of philosophical reflection provided in this chapteranswers to this need. We have taken some inspiration from a form of inquiry inlinguistics that is plausibly parallel in that it begins with (putatively) competence-based judgments about concrete presented cases and treats these as data for abductivereflection. We suggest that Williamson should aspire to something along these lines.Indeed, we suggest that he has implicitly helped himself to as much when he borrowsresults from Kripke’s reflections on natural-kind substance-concepts—and borrowsfrom a range of related work. We have argued that such work affords a perspective onhow the relevant concepts operate—such as the perspective reflected in H1 and H2—and that these important results amount to conceptually grounded necessary truthsgotten by low-grade a priori reflection.We turn now from fans of philosophical armchair-methodology on our left wing

(notably Williamson), to our other flank. On our right are hypothetical philosophers,and perhaps some real ones too, who would hold out for the following kind of view. Itis indeed the case that one can come to know philosophically important analytic truthsvia a form of abduction that can be conducted effectively in the armchair. Also, it isindeed the case that this kind of abductive reasoning is quite common in philosophy.However, such armchair abduction is really a high-grade a priori procedure, rather thanbeing a procedure that inherently deploys empirical considerations. The inputs to thereasoning process are the contents of high-grade a priori judgments about focusedscenarios, and reasoning itself proceeds in accordance with non-empirical norms ofepistemologically proper abduction. It is true that the conclusions one reaches by suchhigh-grade a priori reasoning are themselves defeasible and fallible—and hence thatthese conclusions are subject to being overturned by subsequent empirical consider-ations. But no matter; armchair abduction aimed at generating defeasible knowledge ofconceptually grounded necessary truths is itself a high-grade a priori process.

Our response to such philosophers on our right wing is this. In practice, tentativehypotheses that are generated at stage two of the three-stage procedure we havedescribed are just that—epistemologically tentative, and not yet abductively wellwarranted. In practice, strong abductive warrant only emerges after various kinds ofempirical consideration have been given their due. Recall the remarks on theundifferentiated mixed bag that is the realistic ready output of armchair responses.Stage-one judgments don’t come pre-labeled as analytically necessary, non-analytic-ally necessary, and contingent; they need sorting into these categories as part of theoverall process of abduction, in order to properly assess their differential pertinenceto the abductive task at hand, and such sorting is apt to incorporate empiricalconsiderations. (Is it necessary or contingent, for example, that water is stuff thatis potable and fills lakes and streams and rains from the sky?) Such generalizationsas are readily gleaned from or suggested by simply conditionalizing on scenario-descriptions and verdicts are honestly as mixed a bag as are the responses to scenarios.


It is plausible that ready filters, such as the insistence that OLD-MAN cannot benecessary, and that OLD-MANnec is thus false, draws upon background sensibilitiesthat already reflect the kind of low-grade a priori results we recommend. Further,any candidate abductive generalization will likely confront challenges—e.g., judg-ment tendencies that are recalcitrant in not themselves supporting the generalizationin question. Recalcitrant judgments and judgment-tendencies concerning focusedscenarios need to receive plausible, empirically committal, debunking explanations.The question whether one’s data-set is sufficiently rich, sufficiently varied, andsufficiently representative should be addressed. Pertinent considerations from empir-ical disciplines like cognitive psychology and linguistics should be given their due.(Is contextualism about the concept of knowledge consonant with what cognitivepsychologists have to say about the cognitive mechanisms subserving the competentdeployment of this concept? What about evidence from linguistics?) Lots of give-and-take needs to occur when assessing the overall fit between data and hypothesis,and that give-and-take needs to be empirically informed in a variety of ways. Realarmchair abduction in philosophy cannot be high-grade a priori inference. To thinkotherwise is like thinking that baseball games are won in the third inning, but thewin is defeasible in the ensuing innings.12 We have argued that real armchairabduction in philosophy is a low-grade a priori process, inextricably intertwinedwith matters empirical.

Notes1. It is not implausible to suggest that these judgments, when they are the products of properly

exercised grammatical competence, themselves have the status that later we call “high-grade apriori,” being the product of the linguist’s implicit mastery of the concept grammatical in L(where L is the linguist’s own language). But this suggestion is not essential to the use ofarmchair syntactic theorizing in linguistics as a model for armchair philosophical inquiry inpursuit of conceptually grounded necessary truths.

2. One can and should allow for errors in difficult presentations of the stuff. Humans dooccasionally make such errors without it affecting the referent of their concepts.

3. Further scenarios would reveal that there is probably no nice compact way of formulatingexactly the character of the interactions that would make a difference to reference-fixing. If,for example, the folk were unknowingly envatted, yet in an extended way they interact withstuff in an environment through a vast computer interface, presenting them with thephenomenal world at hand, where the computer itself takes its cues from a model of theworld that is, strangely perhaps, the next planet over, which the computer has manipulatedvia remote robots, and interact with some stuff . . . if that, then one is probably not inclined tosuppose that that is to relevantly interact with H2O or XZY.

4. We might also mention a third generalization, one that is loosely parallel to what proponentsof two-dimensional modal semantics seem to have in mind when they write of “howreference is fixed, considering a centered possible world as actual.” This generalizationconditionalizes on central features of the epistemic situation envisioned in S1 and S3:


H3: As a matter of the way the spawned concepts work, when a community of languageusing agents inhabiting an environment interacts with a homogeneous substancethat they themselves take to be homogeneous, and they adopt a term intending it torefer to that substance, and they remain open to revising their conception of thatsubstance with ongoing inquiry, the concept associated with their term refers tostuff of the underlying physical kind that is instanced in the preponderance of thesamples with which they have interacted.

This matter of how the spawned concepts work would give rise to a class of conceptuallynecessary truths, including:

H4: It is conceptually necessary that for any homogeneous substance s, if folk haveinteracted with s, understanding it to be a homogeneous substance, and haveformed a concept, c, with the intention of referring to stuff of the underlyingphysical kind instanced in their samples, and stand open to accommodate results asto what this underlying stuff is, then s is the reference of their concept c.

5. This date is chosen to stay safely away from Joseph Priestly’s work on “airs” beginning in the1770s, which doubtless changed the epistemic situation in Europe by demonstrating thatwater was not a simple, and by isolating samples of substances from which it was apparentlycomposed.

6. See, for instance Chalmers (2004) and Stalnaker (2008). Chalmers embraces the claim thatall epistemic possibilities are metaphysical possibilities, and he invokes it in articulating oneof two different approaches he explores for applying the framework of two-dimensionalmodal semantics to epistemic modality. But Chalmers also acknowledges that the claim iscontroversial, and the other approach he explores is neutral about it. Stalnaker invokes theclaim in an unqualified way, which leads him (because of a metaphysically modal thesisabout phenomenal consciousness that he accepts but Chalmers does not) to embrace somevery implausible epistemological claims; see Horgan (in press).

7. It bears emphasis that although the wrongheaded idea is probably inspired in part byKripke’s remarks in Naming and Necessity like those quoted earlier, Kripke himself neversuggests that epistemic modals should be treated this way. Indeed, he explicitly repudiatesany such idea. See, e.g., his footnote 72, p. 143.

8. He explores a set of parallel understanding links—links to dispositions to assent and toknowledge as well as to justification:

(UAt) Nec, whoever grasps the thought every vixen is a female fox assents to it. (SeeHenderson and Horgan 2011, chapters 7–8)

(UKt) Nec, whoever grasps the thought every vixen is a female fox knows every vixen is afemale fox.

We focus on justification here, as we are interested in the justification afforded by low-grade a priori reflection.

9. This is how Williamson understands the “pivotal” aspiration associated with epistemicanalyticity (Williamson 2007, p. 77, our emphasis).

10. While we have not given this as high a billing as Williamson (2007, p. 73), it clearly can playa role in some of our account, and we have even discussed the place of simulation ininterpretation and explanation of others, see Henderson and Horgan (2000); Henderson(2011).


11. “However, in our imagination-based knowledge of counterfactuals, sense experience canplay a role that is neither strictly evidential nor purely enabling. For even without survivingas part of our total evidence, it can mold out habits of imagination and judgment in ways thatgo far beyond a merely enabling role” (Williamson 2007, p. 165).

12. We do realize that this analogy is imperfect. It is intended as an analogy for being objectivelyepistemically justified in embracing some philosophically interesting generalization of the sortsdiscussed here. Suppose one generates a few responses to scenarios, notes a plausible generaliza-tion on their conditionalization, and rests content—perhaps one judges that knowledge isjustified true belief, or that necessarily water is the potable stuff filling the lakes and rivers. Wethink that such a result would amount to a form of hasty generalization, because the range ofinquiry needed for fittingly reliable human belief formation has not been undergone. In thisrespect, it would be like declaring victory in the third inning of a baseball game. If this is not aprediction about how the game will turn out, if it is not merely a prediction that one willultimatelywin, the announcement of victory is false—one has not yetwon.Of course, onemightproject that your working hypothesis will turn out to be justified. But, for the sort of difficultconcepts that have been philosophical fare, there remains more to do—more innings to playbefore someone wins. Indeed, one should not be too surprised to see extra innings.

ReferencesBoghossian, P. (1977). “Analyticity.” A Companion to the Philosophy of Language. R. Hale and

C. Wright. Oxford, Blackwell.Chalmers, D. J. (2004). “Epistemic Two-Dimensional Semantics.” Philosophical Studies 118(1–2), 118–226.

Henderson, D. (2011). “Let’s be Flexible.” Journal of the Philosophy of History 5: 261–99.Henderson, D. and T. Horgan (1999). “What Is A Priori and What Is It Good For?” SouthernJournal of Philosophy 38 (Supplement): 51–86.

Henderson, D. and T. Horgan (2000). Simulation and Epistemic Competence. Empathy andAgency. H. Kogler and K. Steuber. Boulder, Westview Press: 119–43.

Henderson, D. and T. Horgan (2011). The Epistemological Spectrum. Oxford, Oxford UniversityPress.

Kripke, S. A. (1980). Naming and necessity. Cambridge, MA, Harvard University Press.Putnam, H. (1975). “The Meaning of Meaning.” Mind, Language, and Reality. H. Putnam.Cambridge Eng.; New York, Cambridge University Press: 215–71.

Stalnaker, R. (2008). Our Knowledge of the Internal World. Oxford; New York, Clarendon Press;Oxford University Press.

Williamson, T. (2007). The Philosophy of Philosophy. Malden, MA, Blackwell Publishing.


6

Concepts, Teleology, and RationalRevision*

Christopher S. Hill

1 IntroductionAre all of our beliefs susceptible to empirical evaluation, and therefore at risk, inprinciple at least, of being called into question by empirical evidence? That is, isevery belief fundamentally empirical in character? Or is there a special class of beliefsthat are immune to empirical revision?

As is well known, W. V. Quine (1980) argued forcefully in “Two Dogmas ofEmpiricism” for the first of these two options. According to him, our grounds for anybelief are mainly empirical, though supraempirical virtues like simplicity, fruitfulness,and continuity with tradition may also play a role in determining whether a propos-ition should be believed, at least in the case of highly theoretical hypotheses. Moreover,every belief is in principle revisable in the light of experience.

Although revolutionary when they were first put forward, these views of Quine’shave come to be widely accepted, even to the point of having some claim to beconsidered the received position. As I see it, however, there is still a great deal to be saidfor the other side. I will make a case for that side in the present paper. More specifically,I will argue for two main theses. One of these is the claim that a number of our beliefscannot be called into question by empirical evidence. This is true, for example, ofdefinitional beliefs, such as the belief that a fortnight is a period of fourteen days, andalso the laws of classical logic. Now it might seem at first that this claim is all that oneneeds to secure immunity to revision. But even if a belief is not at risk of empiricalrefutation, it might still be true that the motivation for holding the belief could beundermined by empirical evidence. Thus, for example, even if there cannot beevidence that disconfirms the laws of classical logic, empirical discoveries might stillprovide motivation for adopting an alternative to classical logic, such as quantum logic.Or so it might seem. My second thesis will challenge this view, maintaining that the

motivation for embracing certain empirically irrefutable beliefs is non-empirical, atleast in part, and that it therefore cannot be undermined or undercut by empiricaldiscoveries. In sum, I shall argue that certain of our beliefs are immune to empiricalrevision in two senses: they cannot be called into question by empirical evidence, andthey cannot be deprived of their cognitive value by such evidence.

2 Quinean SkepticismSince Quine’s attack on immunity to revision is widely regarded as decisive, it willbehoove us to take a look at it before proceeding to consider the prospects ofconstructing a positive account of this notion. I will also discuss Quine’s critiqueof analyticity, because, on some accounts of analyticity, analyticity and immunity torevision come to much the same thing. At one point Quine himself endorses such aview, saying that analytic statements “hold come what may.” (Quine 1980, p. 43)“Two Dogmas of Empiricism” begins with a critique of various traditional attempts

to explicate analyticity. Quine considers the following proposals:

a. A statement is analytic if it is true in all possible worlds.b. A statement is analytic if its denial is self-contradictory.c. A statement is analytic if its predicate attributes to its subject no more than is

already conceptually contained in the subject.d. A statement is analytic if it is true by virtue of meanings and independently of

fact.e. A statement is analytic if it is logically true or it can be turned into a logical truth

by putting synonyms for synonyms.f. A statement is analytic if it is true in virtue of definitions of its constituent terms.g. A statement is analytic if it is true in virtue of the semantical rules that govern the

use of its constituent terms.

Roughly speaking, Quine’s view is that the ideas that figure in these definitions—ideaslike meaning, synonymy, and definition—are highly problematic because they are notembedded in well-articulated theories with determinate predictive content. To put thepoint another way, although the ideas are all definable in terms of one another, they donot have any robust definitional or doctrinal links to concepts that have a clearpredictive or explanatory role. (Here and elsewhere in the present section, I followQuine in framing the discussion in terms of words and statements rather than conceptsand propositions. This will reduce possible distortions in reporting Quine’s views.I hope it will be clear that what is said in the present section can easily be reformulatedso as to apply to the universe of discourse of other sections.)According to a familiar interpretation of the argument (Grice and Strawson 1957), its

main premise is the claim that it is necessary to appeal to analyticity itself in explainingthe various notions that figure in the traditional definitions of analyticity. On thisinterpretation, Quine’s objection to analyticity boils down to the claim that circular

CONCEPTS, TELEOLOGY, AND RATIONAL REVIS ION 135

definitions are illegitimate. It is customary among those who read Quine in this way torespond by saying that definitional circles can be perfectly acceptable. They thenillustrate this claim with examples. Reflection shows this interpretation is quite mis-guided. Quine’s point is not that it is necessary to appeal to analyticity in explaining keyconcepts in the definitions, but that those concepts are as lacking in clarity as analyticityitself. They cannot be used to explain analyticity because they do not have clearlydefined roles in fruitful cognitive endeavors.

My objection to the argument is quite different than this standard objection. It is thatQuine’s list of explanations fails to acknowledge a number of promising proposals.Thus, the list makes no mention of proposals that make use of teleological notions,such as the proposal that a statement is analytic if acceptance of the statement isnecessary for one or more of its constituent terms to serve fundamental cognitiveinterests. Nor does it address proposals that invoke epistemic notions, such as theproposal that an analytic statement can be identified as one that we are justified inbelieving simply in virtue of the most basic dispositions or commitments governing theuse of the terms that serve as its constituents (cf. Boghossian 1997). Quine’s list also failsto take note of proposals that are based on the idea of explanation. An example is PaulHorwich’s proposal (Horwich 1999) that the analytic truths involving a term are theones that make a special, foundational contribution to the task of explaining the overalluse of the term. To give one more example, the list fails to acknowledge the possibilitythat analyticity can be explained in terms of the notion of correct use. According to thisidea, a statement S is analytic if S contains a term T such that an agent must accept S inorder to count as knowing how to use T correctly. To be sure, as things now stand, weare unable to say much in a theoretical vein about what mastery of use consists in; butwe have considerable knowledge of what correct use involves in specific cases.Generally speaking, we are able to recognize mastery of use when we see it. This is shownby the fact that we are able to correct language learners when they go astray, by the factthat we are able to compile dictionaries that command wide assent, and by the fact thatwe are in practice able to discriminate between linguistic differences and factualdisagreements. Mastery of use is an ability, and so is the capacity to recognize masteryof use. As with the ability to play the piano and the ability to recognize grammaticalsentences, it is possible in principle for there to be a scientific account of what theseabilities consist in.

After delivering his critique of traditional accounts of analyticity, Quine begins anextended argument for his revisability thesis, according to which it is in principle possiblefor any statement whatsoever, including even a law of logic or a law of mathematics, tobe called into question by empirical evidence. This argument then provides the basisfor a sustained attack on the notion of a priori knowledge. According to the maintraditional conception of the a priori, an agent A can be said to know a priori that astatement S is true just in case the following three conditions are satisfied: (i) A knowsthat S is true, and therefore has a true justified belief it is true; (ii) A’s justification forbelieving S is independent of empirical evidence; and (iii) A’s belief in S is immune to

136 CHRISTOPHER S. HILL

empirical revision, in the sense that the course of experience cannot provide A with anepistemic ground for rejecting S. Quine’s revisability thesis challenges clause (iii) of thisaccount, implying that there are no statements that satisfy it.While it is not altogether easy to see how Quine proposed to defend his revisability

thesis, I think that the preponderance of the evidence shows that he had two argumentsin mind.The first argument consists in an appeal to the history of science. Quine never makes

this argument fully explicit, but it is plausible that it can be formulated as follows:

First premise: If any proposition is empirically unrevisable, then the laws of logic are empiricallyunrevisable.Second premise: The laws of logic have in fact been called into question by empirical discoveries.Conclusion: No proposition is empirically unrevisable.

As a passage in “Two Dogmas of Empiricism” makes clear, Quine would havedefended the second premise by claiming that certain logical laws have been challengedby quantum mechanics. (See Quine 1980, p. 43. Quine 1986, pp. 85–6, offers a morenuanced treatment.)In thinking about this first argument, it is important to distinguish between the claim

that it is rational to consider replacing classical logic with quantum logic, and the claimthat it is rationally obligatory to make such a replacement. It appears that most contem-porary philosophers of science reject the second claim. Reflection shows, however,that Quine is only committed to the first claim. His revisability thesis just asserts that it isin principle possible for there to be evidence that makes it rational to revise logic. Thiscan be defended by simply pointing to rational debate among informed parties aboutwhether to undertake a revision. There is no need to point to cases in which a revisionhas actually occurred.Quine’s second argument for the revisability thesis derives from a certain conception

of rational revision—a conception that Quine takes to be implicit in scientific practice,and also to be defensible by a normative argument (not that he distinguishes sharplybetween these two endeavors). When a theory generates an experimental predictionthat turns out to be false, Quine points out, there are always a number of different waysof revising the theory so as to cancel the prediction. Thus, where X1, . . . , Xn is a list ofthe statements and rules of inference that play a role in the derivation of the prediction,one could in principle block the prediction by revising any one of them. Which Xiwillbe chosen? Quine’s answer is that the revision will be based, and should be based, onthree global properties of the theory in which the Xi’s are embedded—specifically,predictive power, simplicity, and continuity, where continuity is taken to be a matterof similarity between the pre-revision theory and the post-revision theory. One shouldtry to make a revision that will maximize a weighted average of these three variables.What exactly does this answer count against immunity to revision? Because nothinghas been said about preserving the Xi’s that belong to logic or mathematics, or thatcount intuitively as immune to revision for some other reason. If a particularXi belongs


to mathematics, say, it will be preserved just in case it is a component of the revisedtheory that best satisfies the three criteria. There is no guarantee that any givenstatement will pass this test. But is it really rational for us to allow logic and mathematicsto be susceptible to this sort of revision? This is just the question of whether it is reallytrue that predictive power, simplicity, and continuity are the only three desiderata thatwe are rationally required to consider. And Quine thinks we will in the end agree thatthey are. After all, he will say, the principal goal of cognition is the efficient predictionand control of nature. This is our primary cognitive interest. Moreover, he will add, itis clear that the three desiderata are the properties of theories that best promote thisinterest.

As I see it, then, Quine’s second argument for the revisability thesis is basedultimately on a doctrine that is sometimes associated with pragmatism—the doctrinethat efficient and accurate prediction is the main purpose of cognition. I believe that itis this doctrine that leads him to describe his position as involving a “shift towardpragmatism.”

Turning now to the question of whether the foregoing arguments are sound, I willargue that they both suffer from substantial flaws.

In the first place, it is far from clear that the example involving classical logic andquantum mechanics is a case of empirical revision in the relevant sense. A relevant casewould be one in which there were empirical grounds for rejecting a statement thatappears to be a priori. Instead of seeing quantum mechanics as providing a reason toreject classical logic, and to replace it with quantum logic, it is possible, and indeednatural, to see quantum mechanics as providing a reason to add the vocabulary ofquantum logic to our logical repertoire, and as providing a reason to use concepts fromthis new logic in formulating and developing microphysics, while continuing to useclassical logic in formulating and developing such branches of knowledge as math-ematics, philosophy, and law. In other words, instead of seeing quantum mechanics asproviding reasons for replacing an old logic with a new one, it is natural to see it asproviding reasons for supplementing an old logic with a new one, and for using these twologics in tandem. To be sure, it is not exactly a trivial matter to combine logics—asJ. H. Harris (1982) pointed out, it can happen that the wall between two logicscollapses when they are combined to form a single system. Accordingly, if one wishedto combine two logics, it would be necessary to impose restrictions on the use of eachof the logics to prevent them from collapsing into one another. It appears, however,that relatively weak restrictions would suffice, and more specifically, that it wouldsuffice to adopt restrictions that slightly modify the scope of classical logic withoutchanging its substance. (If classical logic and quantum logic are presented as naturaldeduction systems (i.e., as based on systems of rules of inference like the rules (i)–(iii)that are cited previously), it suffices to adopt a restriction forbidding one to apply rulesbelonging to one of the systems within derivations constructed in accordance withrules that belong to the other system. For discussion of the different case in which twologics are presented as axiomatic systems, see Schechter (forthcoming). Schechter


shares my view that it is possible to prevent collapse by adopting restrictions thatpreserve the basic identity of classical logic.)It seems, then, that there is a strategy for dealing with empirical challenges to logic that

has the effect of preserving the logical commitments that have traditionally been seen as apriori. Instead of advocating wholesale replacement, the strategy recommends that wesupplement familiar vocabularies and systems of belief with new ones, and continue touse the familiar items to serve the non-empirical purposes that they have always served. Itseems fair to say that any argument concerning the bearing of developments in thehistory of science on questions of immunity to revision has an obligation to take thisstrategy into account. But Quine’s brief discussions do not acknowledge it, nor does theformal argument that those discussions suggest. Accordingly, the argument cannot standon its own. To make it work, it would be necessary to show that the alternative strategyis unsatisfactory. It is far from clear that this could be done.There is also a problem with Quine’s second argument, which derives his revisability

thesis from the idea that the efficient prediction of experimental data is our dominantcognitive concern, and the only factor that should play a role in fixing belief. Reflectionshows that this idea is quite wrong.We have a number of cognitive interests that cannotbe reduced to efficient prediction. Thus, to give an especially simple example, we have alegitimate interest in introducing abbreviations for complex concepts. It is of course truethat concepts that serve this interest tend to simplify our predictive instruments andthereby make them more efficient; but our interest in abbreviations is not reducible toour empirical concerns. It would be reasonable to employ abbreviative concepts even ifwe were animated mainly by religious purposes, and had little interest in predictingnature or improving our understanding of it. For another example, consider ourmathematical beliefs. It is implausible that we hold these beliefs on the basis of empiricalevidence. Even so, however, it is evident that they are quite important to us. We regardit as entirely rational to seek to obtain more of them, even when doing so precludesone’s making additions to one’s stock of non-mathematical beliefs. Thus, for example, ifsomeone devotes her life to describing a mathematically interesting structure, we aremore likely to applaud her choice than to condemn it. Moreover, we are likely toapplaud the choice even if it prevents her from acquiring a certain range of predictiveabilities. Perhaps, if she had not been so preoccupied with pure mathematics, she couldhave made a major contribution to empirical science. Will this reflection cause us toshake our heads over the tragedy of a wasted life? Of course not! Obtaining knowledgeof mathematically interesting structures seems to us to be of fundamental importance.A third example is afforded by our modal beliefs. There are many questions of possibilityand necessity that we would like to be able to answer, even though the answers will dolittle if anything to augment the predictive power of empirical theories.It is clear, then, that we have some cognitive concerns that are independent of

our interest in the efficient prediction and control of nature. Since it fails to acknow-ledge this important fact, Quine’s second argument falls far short of establishing itsconclusion.


To summarize: Although Quine’s objections to analyticity and immunity to revisionare widely regarded as devastating, or at least as shifting the burden of proof toadvocates of these two notions, reflection shows that they actually contain quitesizeable holes. A positive account of analyticity has nothing to fear from them, nordoes a positive account of what it is for a proposition to be constitutive of a concept.Moreover, they leave defenders of the a priori with plenty of logical room in which tooperate.

3 Concepts and Cognitive InterestsAs the reader will recall, Wittgenstein’s Philosophical Investigations (Wittgenstein 2001,p. 6e) encourages us to think of linguistic expressions as having functions:

Think of the tools in a tool box: there is a hammer, pliers, a saw, a screw-driver, a rule, a glue-pot, glue, nails and screws. . . . The functions of words are as diverse as the functions of theseobjects.

Wittgenstein is here speaking about words rather than concepts, but what he says aboutwords seems to me to be true of concepts as well: they have functions, and thosefunctions are highly diverse, differing among themselves as much as the functions of ahammer, a saw, and glue.

It must be acknowledged that in many cases, it is very difficult to figure out whatcognitive interests a concept serves. Consider, for example, the concept of truth. It isarguable that we had no clue as to the cognitive function of this concept, despite manycenturies of speculation about the topic, until Quine pointed to its role in generalizedand indefinite assertion (Quine 1986, p. 12). Similarly, although great strides have beenmade in recent years, it is clear that many generations failed to appreciate the complexfunctions of conditional concepts. Even today we seem to lack fully adequate accountsof these concepts.

Still, the fact that cognitive functions are elusive in some cases should not blind us tothe fact that they are frequently transparent. Among other things, it is not hard to seethat perceptual concepts serve an interest in encoding information about such observ-able properties as sizes, shapes, and colors, that theoretical concepts serve an interest informulating laws of nature, and therefore an interest in explaining and predictingobservable phenomena, and that logical concepts serve an interest in framing propos-itions that can play certain roles in inference.

Moreover, in many cases, philosophical reflection has extended commonsenseknowledge of functions considerably. Consider, for example, the concept of disjunc-tion. Casual observation can lead to the conclusion that the value of disjunction derivesin part from its ability to participate in such inferences as reasoning by cases, disjunctivesyllogism, De Morgan’s laws, and the abbreviative inference that leads from pairs ofpropositions of the forms if p then r and if q then r to propositions of the form if p or q then r.But philosophical reflection has gone beyond these observations. Specifically, it has


shown that all of the useful deductive inferences involving disjunction can be capturedby rules of the following forms (together with appropriate rules for the other connect-ives):

(i) For any class of propositions �, if p is derivable from �, then (p or q) is derivablefrom � as well.

(ii) For any class of propositions �, if q is derivable from �, then (p or q) is derivablefrom � as well.

(iii) For any class of propositions �, if (p or q) is derivable from �, r is derivable fromp together with the members of �, and r is also derivable from q togetherwith the members of �, then r is derivable from �.

Since all of the useful deductive inferences involving disjunction follow from (i)–(iii), itis possible to identify the inferential interests that disjunction serves simply by enumer-ating these rules. But more: we also know that disjunction serves an epistemic interestthat is independent of our interest in having a concept that conforms to (i)–(iii). Morespecifically, it serves an epistemic interest in forming propositions that we can bejustified in believing even when we aren’t justified in believing either of their constitu-ents. Thus, for example, we can be justified in believing it will either rain tomorrow or snoweven if we are not justified in believing either of the propositional constituents of thisclaim. This much is apparent to common sense. Philosophical and mathematicalreflection have gone beyond this casual observation by providing a specific rule forassessing the degree to which epistemic support for simpler propositions is transferredto disjunctions that contain them. Expressed as a formula governing the probabilities ofdisjunctions, the rule comes to this:

(R) If Pr(e) > 0, then Pr(p or q/e) = Pr(p/e) + Pr(q/e) – Pr(p and q/e).

In combination with the other rules of the probability calculus that govern disjunction,this rule enables us to give a quite concrete account of the epistemic functions of theconcept. Accordingly, it is arguable that when (i)–(iii) are combined with the rules ofthe probability calculus that govern disjunction, we have a package that can account forall of the most fundamental functions of that concept. To put the point in somewhatdifferent terms, given that disjunction serves our interest in having a concept thatconforms to (i)–(iii) and rules like (R), we can see how it serves the full range of theinterests that are associated with disjunction, including, among many other things, ourinterest in having a concept that enables us to form new beliefs by disjunctive syllogism,and our interest in being able to form justified beliefs in complex propositions eventhough the available evidence falls short of justifying belief in their constituents.It appears, then, that by building on common sense, philosophy has made it plausible

that in a number of cases, at least, concepts are associated with sets of cognitive intereststhat are rich enough to account for their individuation.The view I wish particularly to recommend, however, is that questions about the

cognitive functions of concepts are empirical, and that science must therefore play the


dominant role in identifying such functions and describing their interconnections. Inconjunction with common sense, philosophy can throw light on global functions suchas contributing to predictive and explanatory success, and can also play a role inidentifying the more specific functions that are performed by especially simple con-cepts, like the logical constants. But the primary responsibility for developing a theoryof the cognitive functions of concepts must lie with scientific disciples such as psych-ology, anthropology, and the history of science. Questions about cognitive functionsare often too delicate, and too complex, to be settled by causal observation andarmchair reflection. Fortunately, science is already making lengthy strides forward inthis area, as can be seen by considering recent work in experimental philosophy, andrecent psychological research on such topics as categorization, cognitive development,and taxonomy. Thus, for example, in her recent book on conceptual development,Susan Carey describes empirical studies that answer a number of questions aboutthe cognitive functions of the concepts of infants and young children, includingtheir object concepts, their numerical concepts, and their concept of causation(Carey 2009). She also describes a number of findings that throw light on the cognitiveinterests served by a number of adult concepts, such as the notions of weight, mass,volume, and density.

Thus far we have seen that common sense and philosophy have succeeded inidentifying a number of the general functions that concepts perform, and have alsotaken note of some plausible claims about the concrete functions of specific concepts.We have also seen that psychology is often able to go beyond common sense andphilosophy in specifying functions. Indeed, given the successful empirical inquiries intofunctions that Carey reports, it is reasonable to expect that various scientific discipleswill someday converge on a general taxonomy of concepts, with positions in thetaxonomy being determined by cognitive functions.

In addition to maintaining that concepts serve cognitive interests, and that it is inprinciple possible for such interests to be identified, I wish to claim that concepts oftenserve cognitive interests that lie far afield from our empirical concerns. We canappreciate the merits of this third claim by considering the concepts that belong tologic and mathematics. It is true that concepts of these two kinds support empiricalinquiry in various ways, but it is plausible that they also serve non-empirical purposes.Thus, for example, it is plausible that the laws of classical logic play essential roles insuch non-empirical disciplines as law, mathematics, and theology, not to mention largeparts of philosophy. Because they serve non-empirical purposes, the laws of logic couldsurvive discoveries that called into question their utility in empirical endeavors.Euclidean geometry provides additional evidence for the third claim. In the earlytwentieth century empirical scientists gave up on the view that Euclidean geometryprovides an adequate model of physical space, but they continued to regard it as anexceptionally important piece of pure mathematics. Moreover, as a glance at themathematical offerings at any university will attest, it retains this status today. Thiscould not be true if the purposes that Euclidean geometry serves were purely empirical.


No doubt it is still taken seriously for a variety of reasons, but one of them is surely thatit articulates a structure that is exemplified in some possible worlds, and that humanbeings resonate to representations of that structure psychologically, independently ofwhether they believe it to provide a model of the physical world.Speaking more generally, it seems that mathematical concepts serve an interest in

articulating intuitions about structures that are possible and have a powerful aestheticappeal. This is not an especially informative characterization ofmathematics, and it certainlydoes not succeed in capturing all of the interests that lead mathematicians to pursue theirsubject, but it does seem to be true as far as it goes, and it suffices to make the point thatmathematics is not a slave to empirical purposes. I will elaborate a bit in Section 4.Now if the cognitive interests that serve as the raison d’etre of a concept are entirely

or largely non-empirical in character, then the commitments that govern the daily useof the concept can authorize beliefs and/or inferences that are non-empirical as well, inthe sense of authorizing us to hold beliefs and engage in inferences independently ofthe direction that the course of empirical evidence happens to take. Indeed, thecommitments that govern the use of the concept should authorize beliefs and/orinferences that are independent of the course of empirical evidence, because beliefsand/or inferences generally play essential roles in the proper functioning of concepts.A belief or inference that makes it possible for a concept to serve a non-empiricalcognitive interest should not be held hostage to the course of evidence.

4 Abbreviative DefinitionsThe purpose of an abbreviative concept is not to add to the predictions of an empiricaltheory, but to achieve economy of belief and reasoning. More specifically, an abbre-viative concept makes it possible to form propositions that (i) are shorter than counter-part propositions that contain a complex concept, and (ii) have the same epistemicproperties as their more complex counterparts (except for those arising from themereological structure of the abbreviated concept).It follows that when an agent introduces an abbreviative concept, he or she must

adopt rules of use which guarantee the coincidence of epistemic roles that is envisionedin (ii). In the case of the concept of a fortnight, an agent could do this by adopting (F1)and (F2):

(F1) Where P is any proposition containing one or more occurrences of theconcept period of fourteen days, P* is any proposition that is just like P exceptfor containing one or more occurrences of fortnight in place of occurrences ofperiod of fourteen days, and E is any body of evidence, the epistemic probabilityof P* relative to E is to be derivative from, and exactly identical with, theepistemic probability of P relative to E.

(F2) Where E is any body of evidence, the following proposition is to have themaximum epistemic probability relative to E:

(D) A fortnight is a period of fourteen days.


It would be possible to derive (F2) from (F1) and the assumption that for any body ofevidence E, every proposition that is a logical truth should have the maximumepistemic probability relative to E. This is so because the proposition a period of fourteendays is a period of fourteen days is a logical truth, and because proposition (D) is just likethat proposition except for having an occurrence of fortnight in place of an occurrenceof period of fourteen days. For present purposes, however, it is useful to be able to think of(F1) and (F2) as equally fundamental. (F2) plays a crucial part in fixing the epistemicrole of the concept of a fortnight, for it is that rule that authorizes us to take proposition(D) as a free premise in any context whatsoever (except perhaps for contexts involvingpropositional attitudes). In the interests of clarity, it is desirable to keep our discussionof abbreviative definitions independent of more complex questions about the epi-stemic role of logical concepts.

Is it legitimate for an agent to commit to using the concept of a fortnight inaccordance with (F1) and (F2)? Yes. When an agent acquires a new concept, it is upto him or her to determine how the concept will be used, and doing this clearlyinvolves determining which bodies of evidence will count as supporting propositionsthat contain the concept, and how much support each such body will provide. Wherecould the epistemic properties of a concept come from, if not from the rules of use thatthe agent adopts? To be sure, there are restrictions that an agent who is introducing anew concept must observe: the rules of evidence must not authorize contradictorybeliefs, and they must be internally coherent, in the sense that they must not implyconflicting claims concerning the evidential support concerning particular propos-itions. Moreover, in addition to these two general restrictions, there are others thatgovern rules of use of specific kinds. Of particular relevance here are two additionalrestrictions that the logical tradition has recognized as governing abbreviative defin-itions (Gupta 2008). One says that rules that introduce an abbreviative concept must beconservative, in the sense that they must not authorize us to form any new beliefs otherthan ones that involve the given concept. And the other says that an abbreviativeconcept must be eliminable—that is, it must be true that every proposition containingthe concept is equivalent to one that lacks the concept. These restrictions are import-ant. But as reflection will attest, they are easily satisfied by (F1) and (F2).

I have claimed that an agent commits to using the concept of a fortnight inaccordance with (F1) and (F2) in adding the concept to his or her conceptual lexicon.It is this commitment that makes it possible for the concept to serve the purpose ofabbreviating period of fourteen days. But what does it mean to say that an agent hasundertaken a commitment of this sort? There are two quite different ways of answeringthis question. One option is to say that the agent forms an intention to use the conceptin accordance with (F1) and (F2)—that is, an intention to use it in the ways that (F1)and (F2) specify. The other option is to say that the agent acquires a disposition to usethe concept in accordance with (F1) and (F2). I will not attempt to decide betweenthese options here. (On either option it will be true that the commitment serves acognitive interest of the agent.)


On the account I am proposing, the use of the concept of a fortnight is governed byrules of evidential support, but some philosophers would prefer to say that it isgoverned by rules that assign truth conditions to propositions containing the concept.More specifically, they would prefer to say that when we add the concept to ourlexicon, we in effect stipulate that any adequate interpretation of the lexicon mustconfer the same truth values on propositions containing fortnight as they confer on thecorresponding propositions containing period of fourteen days. In my view, this accountof the situation is confused. When we approach the task of introducing a new concept,we are not envisioning a range of possible interpretations of our conceptual scheme,and we do not have in mind a relational concept of truth. Indeed, as I see it, theconcept of truth under an interpretation is a late addition to our conceptual scheme,one that we acquire from mathematical logic long after the basic shape of the schemehas been determined. The content of the commonsense concept of truth is fixed by theclass of propositions that have the following form:

The proposition that P is true just in case P,

and we learn to wield the concept by acquiring a commitment (or a disposition) toaccept members of this set (Horwich 1998; Hill 2002). This view precludes the idea thatthe concept of truth is relational. Moreover, the view implies that propositions and theirconceptual constituents must have fully determinate natures before it is meaningful to talkof truth and truth conditions, for it implies that we presuppose propositions in explainingwhat it is for propositions to have truth values. It follows that the rules that fix the natureof concepts and propositions cannot take the form of assignments of truth conditions.To summarize, the main points so far are that the concept of a fortnight is used

mainly to abbreviate the more complex concept period of fourteen days, and that the wayto provide for this use is, first, to arrange for the former concept to have the sameepistemic properties as the latter, and second, to arrange for proposition (D) to have anepistemic status that makes it possible for us to use it as a free premise in any context.I will turn now to consider the implications of these views for questions aboutrevisability and apriority.It is clear that as long as (F1) and (F2) are in place, there will be an epistemic right to

believe proposition (D), because the proposition will be supported to the maximumdegree by every possible course of experience. It follows that the proposition is notvulnerable to disconfirmation by empirical evidence. But what about (F1) and (F2)themselves? Can empirical evidence call them into question? No, because they make noempirical claims. As we have already observed, they are conservative rules in the sense thatthey do not add to our epistemic commitments: it is possible to adopt them withoutbecoming committed to any proposition not involving fortnight to which one was notpreviously committed. This is true, in particular, of empirical propositions. (F1) and (F2)do not add to the empirical content of our theory of the world. They simply provide foralternative formulations of empirical commitments that we have already made.


I will not go into the question of whether proposition (D) is known a priori, becausethat would involve us in questions about the nature of knowledge, and also inquestions about permissible ways of arguing for the existence of knowledge, that arebest left aside in the present context. But we should consider the question of whetherour justification for believing (D) is a priori. A justification is a priori if it is independentof empirical evidence. Now reflection shows that “justified independently of empiricalevidence” can mean two quite different things. On the one hand, it can mean that ourjustification for believing (D) does not depend on any particular body of empiricalevidence—that it will be justified by the available evidence no matter what directionthe future course of experience happens to take. And on the other hand, it can meanthat our justification for believing (D) is independent of all empirical evidence. Now itis clear that if the foregoing reasoning is correct, (D) does count as having a non-empirical justification in the first sense. (F2) implies that (D) will be justified by anypossible future course of experience. But it also seems possible to argue that (D) has anon-empirical justification in the second sense. An agent who is conforming to (F2) isin a position to reflect that she will be justified in believing (D) no matter what formthe future course of experience will take. It seems natural to say that this reflection canprovide a justification for believing (D) that is prior to all future courses of experience,and that is therefore in an important sense independent of all of them.

I have maintained that (D) has a special cognitive and epistemic status in virtue of itsrelationship to the foundational rules of support concerning the concept of a fortnight.Moreover, I have maintained that (D) will retain this special status indefinitely—in fact,for as long as we continue to use the concept. This second view is challenged by thefollowing remarks by Quine, which urge that any special status that is originallypossessed by propositions like (D) is transitory, and that such propositions eventuallyacquire the status of empirical generalizations:

Legislative definition introduces a notion hitherto unused, or used only at variance with thepractice proposed, . . . so that a convention is wanted to settle the ambiguity. [“Legislative”] refersto the act [of defining a term], and not to its enduring consequence . . . . This is because we aretaking the notion of truth by convention fairly literally and simple-mindedly, for lack of anintelligible alternative. So conceived, conventionality is a passing trait, significant at the movingfront of science but useless in classifying the sentences behind the lines. It is a trait of events andnot of sentences. (Quine 1976, pp. 118–19)

According to Quine, then, while a proposition like (D) may enjoy a certain elevatedcognitive status at the moment at which it is initially adopted, this special status isnecessarily short-lived. Elaborating, we can perhaps say that according to Quine, one’soriginal intention to use a concept C* as an abbreviation for C is modified over time,and that it eventually evolves into a generalized intention to use C* to advance one’smain cognitive goals, such as predicting data, explaining data, and registering infor-mation. At this point the proposition a C* is a C, which was originally accepted on thebasis of the rules of evidence that establish an abbreviative practice, has no special status.


It is on a par with the other propositions in one’s system of beliefs, and as a result, likeany other statement, it can become a casualty of revisionary activity.This line of thought seems quite misguided to me. In particular, it seems quite

wrong to suppose that the intention to use C* as an abbreviation will be swamped oreroded by the desire to use C* to advance more important cognitive concerns. Thedesire to use C* as an abbreviation for C is perfectly compatible with the desire to use itin framing predictions and explanations, and also with the desire to use it in registeringinformation. Indeed, using C* as an abbreviation for C can be a way of furthering theseother concerns. When one introduces C* as an abbreviation for C, C already plays arole in various cognitive activities, including, we can suppose, the activities of predict-ing data, explaining data, and registering information. One introduces C* because onewants to have a more efficient way of conducting these activities, not because onewishes to conduct some other sort of business that is independent of them. It followsfrom this that one can continue to treat C* as an abbreviation for C across time,without encountering eroding pressures from other concerns. If it makes sense to useC* to streamline the business of predicting, etc. at one point of time, it can make senseto continue to do so. Indeed, there is no reason in principle why one might notcontinue to do so for an indefinitely long time.I wish to claim, then, that there is no natural limit to the lifespan of an intention to

use a concept as an abbreviation. Contrary to what Quine seems to have thought, thereis no tendency in the nature of things for intentions of this sort to be swept away by ourpursuit of more fundamental cognitive goals.Before moving on to consider other matters, I want to mention a distinction

between two quite different forms of revision. We have found reason to think that ifthe proposition a C* is a C is an abbreviative definition that introduces a concept C*,then it is just not possible, even in principle, to find empirical evidence that makes itappropriate for us to set the proposition aside as false. In this sense, there can be noempirically motivated revisions of definitions. Reflection shows, however, that there isanother sense in which it can be said that abbreviative definitions are subject toempirical assessment. This sense is brought to the fore by the following argument:“When we introduce a new concept C* to abbreviate a more complex concept C, it isgenerally true that C is a useful concept—an expression that plays a valuable role in oneor another of our cognitive practices. We introduce the defined concept simplybecause we would like to have a more efficient way of conducting the valuablebusiness in which C is involved. Now consider a case in which C is involved inempirical business. Suppose, in particular, that C enjoys a life in a branch of science.Further, let this case be one in which the theory in which C is involved eventuallybecomes discredited. Eventually we acquire evidence that leads us to lose our earlierconfidence in the theory, and we wind up with the view that C does not refer toanything. After this revision, we would no doubt progressively lose interest in C, andperhaps we would eventually reach a point at which it played only a very small role inour mental life. It could then easily happen that we ceased to make any use of the


abbreviative concept C*, and that we ceased to affirm the definition by which it wasoriginally introduced. The best way to describe this example is to say that it is one inwhich empirical evidence leads us to set a definition aside as useless. In other words,while it is not an example in which evidence calls the truth of a definition intoquestion, it does seem to be an example in which evidence undermines commitmentto a definition.”

Thus, in addition to revisions that are occasioned by empirical disconfirmation, thereare revisions that occur when the motivation for an abbreviative concept is undercutby empirical developments. It is clear that revisions of the second sort do sometimesoccur, but it can also happen that the motivation for an abbreviative concept cannotbe undercut in this way. To see this, it suffices to take another look at Euclideangeometry, which as standardly developed, contains many concepts that serve toabbreviate concepts that are more complex. In the early twentieth century physicistslost their motivation for deploying those abbreviative concepts in formulating laws ofnature, but the concepts have been retained by the mathematicians who continue tostudy Euclidean structures. In general, the motivation for using abbreviative concepts isoften quite complex, involving a priori components as well as empirical ones. To theextent that this is true, discoveries that undercut the empirical motivation for abbrevi-ations have no tendency to reduce our conceptual lexicon.

In sum, it appears that in at least one case, that of abbreviative definitions, it isappropriate to view propositions as having an interesting form of immunity to empir-ical revision, and that it is also possible to view them as justified a priori.

5 Logical ConceptsAccording to the picture I sketched earlier, logical concepts play two main roles in thecognitive life of an agent—an inferential role and an epistemic role. I will not be able todiscuss this claim in full generality here. Rather I will have to focus on a single example.The natural choice is classical disjunction, for the stage has already been set for adiscussion of that concept. I will focus on the question of whether the rules that governthe use of classical disjunction, and the logical beliefs to which they give rise, areimmune to empirical revision. As we will see, there is reason to answer this question ina way that is more in line with traditional rationalist doctrines than with Quineanempiricism. I hope it will be clear that what I will say about this one case can be adaptedso as to apply to others.

Now as we saw earlier, it is possible to account for all of the inferential properties ofclassical disjunction by supposing that it is governed by the introduction and elimin-ation rules of classical logic (i.e., rules (i)–(iii) of Section 3). As for the epistemic role ofthe concept, its most important feature is that it enables us to form justified beliefs incomplex propositions in circumstances in which we would not be justified in believingsimpler ones. To be more specific, it is part this role that we are in some circumstancesjustified in forming a belief of the form P or Q even though we are not justified in


believing P or in believing Q. It is also part of the role that we are in some circum-stances justified in believing a proposition of the form if P then Q or R in circumstancesin which we are not justified in believing if P then Q or in believing if P then R. It ispossible to account for these epistemic properties of classical disjunction by supposingthat it is governed by the following epistemic rule:

(R) If Pr(e) > 0, then Pr(p or q/e) = Pr(p/e) + Pr(q/e) – Pr(p and q/e).

The importance of this rule, and the disjunctive beliefs that it authorizes us to form,become evident when the beliefs in question are considered in relation to the inferen-tial role of disjunction. (R) gives us the right to form beliefs that we can then exploit invarious ways in the inferences that are directly authorized by rules (i)–(iii), and in theinferences that are authorized by (i)–(iii) in combination with the rules that govern theother connectives. (The beliefs with which (R) is concerned come in degrees—theyare graded beliefs rather than the “full-strength” or “binary” beliefs. The question ofhow graded beliefs are related to binary beliefs is complex, and I cannot address it here,though I should explicitly acknowledge that I am assuming that there are principlesthat permit transitions from the former to the latter.)To summarize: if it is true, as I contend, that the main cognitive functions of classical

disjunction are inferential and epistemic, it is possible to describe the main contribu-tions that classical disjunction makes to cognition by listing the classical introductionand elimination rules and the laws of the probability calculus that govern disjunction,including principally (R).Is the epistemic role of classical disjunction on a par with its inferential role, or is one

more fundamental than the other? Is it possible to explain its inferential properties interms of its epistemic properties, or the latter in terms of the former? No. There is nochance of a reductive explanation in either direction, for deductive reasoning makesuse of different abilities than the numerical computations that are required for probabil-istic updating. It should be noted, however, that there is an important relationship ofanother kind between the roles of inference for disjunction and the laws of epistemicprobability.Thus, to put the point somewhat impressionistically, it can be shown that the

probability that the conclusion of a disjunctive inference is false can never be greaterthan the probability that at least one of the premises is false. To be more precise, it ispossible to establish (1)–(3):

(1) For any two propositions p and q, the probability that (p or q) is false can never begreater than the probability that p is false.

(2) For any two propositions p and q, the probability that (p or q) is false can never begreater than the probability that q is false.

(3) For any three propositions p, q, and r, the probability that r is false cannot begreater than the probability that at least one of the following propositions is false:(p or q), (p ! r), and (q ! r).


(Here! is the material conditional and (p! r) and (q! r) represent the inferences p ├ rand q ├ r respectively. I am simplifying by focusing on the case in which the premises ofdisjunctive inferences are single propositions rather than classes of propositions, but theargument can easily be extended to cover the latter case involving finite classes.)

Each of (1)–(3) is a special case of the general rule that the probability of the conclusionof a classical argument is false is always less than or equal to the probability of at least oneof the premises is false. (For a proof, see Bennett (2003), sections 21 and 53.)

Is it legitimate for us to adopt the package of rules that consists of the introduction andelimination rules and the family of probabilistic rules that includes (R)? Yes. As weobserved earlier, it is permissible to adopt whatever rules of use onewants in introducinga new concept, provided that the relevant restrictions are satisfied. And the relevantrestrictions are satisfied in the present case. Thus, it is clear that the package does notauthorize contradictory beliefs, or assign conflicting properties to individual propos-itions. Moreover, in keeping with the requirements of the logical tradition, the rules ofinference for classical disjunction are conservative, in the sense that they do not authorizeus to believe any new propositions other than ones that contain disjunction. That is,apart from propositions that contain disjunction, our doxastic state after adopting theintroduction and elimination rules will be exactly the same as it was before.

To be sure, when the introduction and elimination rules are combined with (R), weacquire an epistemic right to hold new beliefs that do not involve disjunction. Thus, (R)authorizes us to form new disjunctive beliefs, and the rules of inference for disjunctionauthorize us to derive non-disjunctive conclusions from disjunctive premises. Inevit-ably, when these rules are combined, they authorize us to form new non-disjunctivebeliefs. To see this, suppose that r is a proposition that does not contain disjunction, andsuppose also that in combinationwith (R), the available evidence entitles us to believe (por q). Suppose further that r is derivable from p and is also derivable from q. In thesecircumstances, the elimination rule for classical disjunction authorizes us to infer thenon-disjunctive proposition r. It appears, then, that even though the rules of inferencefor classical disjunction are conservative when taken by themselves, the package con-sisting of the rules together with (R) is not conservative. Is this a problem?

No. There is nothing intrinsically wrong with rules that authorize new beliefs. Wesubscribe to many such rules. Moreover, the package consisting of the rules of inferencefor classical disjunction and (R) is particularly innocuous, As we have already seen, if aninference is performed in accordance with one of the classical rules for disjunction, thenthe probability that the conclusion is false cannot be greater than the probability that atleast one of the premises is false. It follows that when we obtain the non-disjunctiveproposition r in the way described in the previous paragraph, we are not taking any newepistemic risks. We will never be in a situation in which our evidence requires us toaccept all of (p or q), (p ! r), and (p ! r) while demanding that we reject r.

There is, however, a question pertinent to revisability that remains to be addressed. It isclearly possible for there to be an empirical theory that enjoys great predictive success onthe whole, but that in some cases leads via the classical rules of inference to conclusions


that are false. Now if the theory is sufficiently successful, it seems that it might be rational,in some reasonably robust sense, for the relevant cognoscenti to consider changing thetheory by replacing the classical connectives with connectives that are governed by lesspowerful rules of inference—that is, by rules that do not permit one to deduce the falseconclusions. Arguably this is what happened at one point in the debate about the properformulation of quantum mechanics. Birkhoff and von Neumann proposed a change inthe underlying logic, and Putnam and various others endorsed their proposal. Morerecently, a number of philosophers have objected to this approach. It appears that allparties to this dispute were rational, perhaps ideally so. Doesn’t this show that the classicalrules of inference are in principle vulnerable to revision?Although I cannot hope to do full justice to this question here, I wish to suggest that

we have already taken note of an adequate strategy for dealing with situations like theone presented by quantum mechanics. The main idea is that such situations can alwaysbe handled by supplementing classical logic with an alternative. There is no need toreplace it. To be sure, this strategy presupposes that there will always be a reason forhanging on to classical logic. But the presupposition can be defended. It is recognized onall sides that classical logic gives us inferential options that such alternatives as intuitio-nistic logic and quantum logic cannot provide. The alternatives to classical logic areconstitutionally weaker, which is why the question of replacing classical logic arises inthe first place—since classical logic is more powerful, theories framed in terms of it canhave false consequences while theories based on the alternatives do not. Because of thisextra strength, classical logic serves more cognitive interests than the alternatives. It hasadditional cognitive value. It would be rational for us to want to continue to avail ourselvesof this additional value in reasoning about non-empirical domains, even if it provedinconvenient to use it in formulating and developing empirical theories.It may be useful to consider this situation in somewhat greater detail. We should

begin by taking note of a small significant difference between the foregoing rule forclassical disjunction and the corresponding rules for quantum disjunction, which I willrepresent by quor. Here again are the classical rules:

(i) For any class of propositions �, if p is derivable from �, then (p or q) is derivablefrom � as well.

(ii) For any class of propositions �, if q is derivable from �, then (p or q) is derivablefrom � as well.

(iii) For any class of propositions �, if (p or q) is derivable from �, r is derivable from ptogether with the members of �, and r is also derivable from q together with themembers of �, then r is derivable from �.

And here are the corresponding rules for quantum disjunction:

(iv) For any class of propositions �, if p is derivable from �, then (p quor q ) isderivable from � as well. (Here and elsewhere, I use quor to express quantumdisjunction.)


(v) For any class of propositions �, if q is derivable from �, then (p quor q) isderivable from � as well.

(vi) For any class of propositions �, if (p quor q) is derivable from �, r is derivablefrom p, and r is also derivable from q, then r is derivable from �.

Now inspection reveals a difference between these two sets of rules. (iii) allows that thederivations running from p to r and q to rmay make use of statements that belong to �,but (vi) authorizes no such use of �. To put the point in positive terms, (vi) requires thatthe derivation running from p to r make use only of p itself, and that the derivationsrunning from q to r make use only of q itself. It is easy to miss this difference in the twoformulations, but it amounts to an immense difference in generative power. It is preciselybecause the rules differ in this way that it is possible to use the rules of inference governingclassical disjunction to prove the distribution law (namely, (p and (q or r)) iff ((p and q) or(p and r))), but it is not possible to prove a distribution law using the rules that governquantum disjunction. (For discussion of these points see Gibbins (1987), chapter 9.)

Thus far we have seen that classical disjunction and quantum disjunction playdifferent roles in inference, and we have also seen that this difference is captured by,and in fact derives from, the difference between (i)–(iii) and (iv)–(vi). I would like nowto emphasize that the difference between classical disjunction and quantum disjunctionis important. Classical disjunction has inferential powers that quantum disjunctionlacks, and as a result, the former serves cognitive interests that the latter cannot serve.

Suppose you have adequate grounds for believing a proposition p and also forbelieving a disjunctive proposition (q or r). Suppose also that you do not have anyreason that is independent of your reasons for believing p and believing (q or r) forbelieving the complex proposition ((p and q) or (p and r)). Suppose further that anotherproposition s can be derived from each of (p and q) and (p and r), but that it cannot bederived from any one of p, q, and r in isolation from the others. Suppose finally that youwould like to have a right to believe s because doing so would enable you to settle aquestion that is of some interest. Clearly this is a possible epistemic situation. Now inthese circumstances, classical logic authorizes you to infer ((p and q) or (p and r)) fromyour beliefs p and (q or r), and it also authorizes you to infer s from ((p and q) or (p and r)).Thus, it gives you the right you seek—the right to believe s. On the other hand, sincequantum logic does not authorize you to infer ((p and q) quor (p and r)) from p and (q quorr), it does not give you the right to believe s.

It is easy to think of concrete examples of this scenario. Suppose you believe that Edwill come to Lee’s party, and also that either Suzy or Carly will come as well. Supposefurther that you believe both of the following conditional propositions: if Ed and Suzycome to the party, then everyone at the party will have fun; and if Ed and Carly cometo the party, then everyone at the party will have fun. Given that or here expressesclassical disjunction, you have the right to infer from these beliefs that you will have funif you go to Lee’s party. This could be a very useful thing to know! It is a mark againstquantum logic that it would prevent us from deriving it. (I have given an empirical


example here in the interests of concreteness, but it is easy to see that there are countlesssimilar cases in which premises and conclusions are concerned with non-empiricaldomains.)It appears, then, that classical logic provides advantages that quantum logic cannot

match. It may be the case that large portions of our conceptual scheme can be recast inquantum logical terms without significant loss, but there are definitely situations inwhich a thoroughgoing acceptance of quantum logic would place us at a disadvantage.We could not formulate beliefs that have the same logical consequences as our classicalbeliefs, and accordingly, many rational beliefs would be lost to us. Accordingly, it seemsadvisable to go forward with a lexicon that contains two concepts of disjunction, onethat we use in formulating the laws of quantum mechanics, and another that we use inother domains.I have maintained that the rules governing the use of classical disjunction are

immune to empirical revision, and also that classical disjunction serves purposes thatare (i) important and (ii) independent of its role in empirical theories. I turn now brieflyto the case of logical laws that involve disjunction. There are two points that arerelevant to our present concerns. First, the laws in question are immune to empiricaldisconfirmation. This is a special case of the following general principle:

(E) Where p is any law of classical logic, and e is any body of evidence, Pr( p/e) = 1.

As with (R), I am thinking of (E) as a fundamental rule concerning the allocation andevaluation of evidence. In effect, it helps to define the classical connectives bystipulating that certain propositions containing them will never be challenged byempirical discoveries. Second, there is no risk that the motivation for embracing theclassical connectives will be undercut by empirical discoveries, such as the discovery thatit would be desirable to use different connectives in formulating a particular empiricaltheory. By the same token, there is no risk that we will be led by an empirical discoveryto abandon the classical laws, and allow principle (E) to lapse into irrelevance. This istrue for the same reason that empirical discoveries could not undercut our motivationfor the classical rules of inference. Even if they were to cease to guide us in ourreasoning about an empirical domain, there would be other domains in which theycontinued to play a useful role. That is, like the classical rules of inference, they wouldearn their keep by contributing to non-empirical endeavors. Hence, they are immuneto empirical revision in the strongest possible sense. They will be available to guidereasoning for as long as human beings continue to think.

6 Additional Examples of ImmunityWe have found that a number of propositions and rules of inference are immune toempirical revision in two senses. They cannot be called into question by empiricalevidence, and the motivation for accepting them can survive changes, including evenradical changes, in empirical theories. The first form of immunity is epistemic immunity,


and the second form is teleological immunity. According to our earlier reflections, both ofthese forms of immunity are possessed by abbreviative definitions, and also by the lawsand rules of inference of classical logic. (This isn’t true of all abbreviative definitions,but as the example of Euclidean geometry shows, it is true of some.) I will now askwhether there are propositions or rules beyond the ones that we have been consideringthat are either epistemically or teleologically immune to revision. What are the scopesof these two conceptions of immunity? Are they narrow or broad?

Arguably, they include the laws of mathematics. In many cases, our reasons foraccepting such laws include the fact that they are empirically fruitful, but there aregrounds for thinking that there are other, deeper reasons for accepting them. At thedeepest level, it seems, mathematicians believe the laws of mathematics because it isaesthetically reinforcing to do so—because in accepting the laws they gain access to arealm of contemplation and activity that is aesthetically rewarding. It also seems likelythat mathematicians are interested in their subject because they believe that it describesnon-empirical entities of some sort—perhaps systems of abstract objects, or perhapspatterns or properties that are instantiated in other possible worlds even if not in ours.These two non-empirical interests are intimately related. The activity of describingabstract entities would have little appeal if it lacked aesthetic value, and the aestheticvalue of mathematical activity would be lessened, perhaps considerably, if mathemat-icians became convinced that there are no entities answering to their descriptions. Thispicture of the reasons for being interested in mathematics receives strong support fromtwo considerations. First, as we noticed earlier, mathematicians continue to be inter-ested in Euclidean geometry despite the fact that it is no longer seen as a description ofphysical space. And second, many mathematical systems have been invented andpursued without there being any empirical applications in sight. This is true, forexample, of hyperbolic geometries. In view of these considerations, it seems appropri-ate to conclude that acceptance of mathematical axioms is due at the most fundamentallevel to a combination of aesthetic reinforcement and a sense that mathematicsdescribes a non-empirical domain, and therefore, that our interest in maintainingthem cannot be undercut by empirical discoveries. By the same token, it seemsappropriate to say that they are immune to teleological revision.

Moreover, a case can be made that the laws of mathematics are immune to epistemicrevision. Given that there are non-empirical reasons for accepting them, we would notwant to adopt rules of evidence which implied that their acceptability is conditional onour having certain types of evidence and not others. Accordingly, we would want ouracceptance of them to be governed by a principle of evidence like the foregoingprinciple (E) concerning the laws of classical logic:

(E*) Where p is any law of mathematics, and e is any body of evidence, Pr(p/e) = 1.

This implies that the laws of mathematics are characterized by epistemic immunity.When people assert (E*), it is usually because it follows from the probability calculus

in combination with the assumption that the laws of mathematics are necessary. Here,


however, I am viewing it in a quite different way—specifically, as representing a free-standing disposition or commitment concerning the evaluation of mathematical prop-ositions in relation to evidence. Since Quine was a staunch opponent of modalassumptions, it would beg the question against him to base a case for (E*) on anassumption about necessity. There is no such problem, however, with the idea that itserves to fix the epistemic role of mathematical concepts.(E*) implies that empirical discoveries have no bearing on the acceptability of

mathematical theories. Given that we rely heavily on mathematics in generatingempirical predictions, it might seem puzzling that we are free to adopt such a commit-ment. But our puzzlement dissipates when we recall that the applicability of mathemat-ics to the world depends on what is known as a representation theorem. To secure theapplicability of a mathematical theory M to an empirical domain D, it is necessary toprove a theorem which establishes that there is a structure-preserving function of somesort—usually an isomorphism—linking D to preferred models of M. One this repre-sentation theorem has been proved, it is possible to combine M with a theory of D togenerate predictions. Suppose now that some of these predictions turn out to be false. Inthis case, we would be compelled to reject the representation theorem, but notM itself.This is because the theorem depends on assumptions of two kinds—the axioms ofM, and empirical hypotheses about D. Given that there is an independent, aestheticmotivation for accepting the axioms, it would preferable as well as possible to respond tothe predictive failure by changing the empirical hypotheses and devising a new math-ematical theory that was better suited to the resulting empirical picture ofD. (Of course,if the sole reason for introducing this new mathematical theory was to provide a modelof D, we would have no reason to hang onto it if the predictions supported by itsrepresentation theorem also turned out to be erroneous.)Are there other examples of immunity to revision? Perhaps. To illustrate the range

of possibilities here, I will discuss a family of examples that lies far afield from the laws ofmathematics.Descriptional theories of meaning claim that concepts of various types are equivalent

with definite descriptions. If any of these theories are true, then propositions givingexpression to the posited equivalences are at least immune to evidential revision. Thus,for example, according to the descriptional theory of kind concepts, a kind concept Kis more or less synonymous with a description that invokes the perceptually salientcharacteristics that are possessed by normal members of K in the actual world. Views ofthis sort met with powerful objections in the 1970s, but they have evolved in ways thatat least deprive those objections of much of their force. (See, e.g., Jackson 2010.) Tothe extent that the descriptional theory can be defended, there is room for an argumentthat our use of kind concepts is guided by commitments concerning the allocation andevaluation of evidence that provide an a priori guarantee that certain propositionslinking kind concepts to descriptions will be supported by evidence. Similar remarksapply to certain propositions that link proper names to descriptions, such as theproposition that Willard Van Orman Quine is the person x such that information


about x is actually transmitted by testimony involving the name “Willard Van OrmanQuine.” Further, as is shown in a classic paper by David Lewis (Lewis 1970), there aregrounds for thinking that theoretical concepts come to the same things as descriptionsbased on the Ramsey sentences of the theories in which they occur. Insofar as this istrue, it is possible to argue that a range of propositions equating theoretical conceptswith Ramsified descriptions are guaranteed to have evidential support.

Descriptional propositions of these kinds are in effect abbreviative definitions, andare therefore immune to evidential revision for the same reason as the proposition thata fortnight is a period of fourteen days is immune. Do they have any claim to beconsidered immune to teleological revision as well? Possibly so, for it might be that wewould wish to think about kinds and theoretical entities as possibilities even ifwe became convinced that they do not actually exist. Moreover, we might wish tocontinue to attribute beliefs and other attitudes involving the relevant concepts toactual and possible agents. We have lost interest in using the notion of phlogiston infirst order descriptions of the world, but we retain an interest in characterizing thebeliefs of the proto-chemists who accepted the theory of combustion and rusting inwhich phlogiston played a role. Perhaps we also wish to reserve the right to use theconcept in characterizing the attitudes of imaginary characters.

It is arguable, then, that our two notions of immunity have spheres of applicationthat are both large and highly variegated. At all events, they seem robust enough tomerit philosophical attention.

7 AnalyticityQuine’s critique of immunity to revision was developed in concert with a systematicattack on analyticity. Now that the former notion has been somewhat rehabilitated, itis natural to inquire whether it might be used to illuminate the latter notion, perhaps byadopting the theory that consists of the following two claims: First, a statement isanalytic just in case it expresses a proposition p such that (i) p is immune to revision, and(ii) p carries no non-logical existential commitments. And second, given that there arethree types of immunity to revision (epistemic immunity, teleological immunity, andthe combination of the two), analyticity comes in three different flavors.

I put this theory forward in a tentative spirit, for it is beyond the scope of the presentchapter to determine how well it squares with our pre-theoretical, intuitive ascriptionsof analyticity. It is clear at the outset, however, that the theory has certain virtues. Forinstance, since it is empirically assessable whether a proposition would be rejected invarious evidential circumstances, the theory implies that ascriptions of analyticity havetestable implications. This is a big plus in the present dialectical context, for as we saw,it was one of Quine’s main complaints about traditional conceptions of analyticity thatthey are empirically empty. I also note that Quine himself held that analyticity squaresup with unrevisability pretty well. Thus, as noticed earlier, he says that an analyticstatement is one that “holds come what may.”


Note* Earlier versions of this chapter were presented to a discussion group at Australian NationalUniversity in 2007, a conference honoring Ruth Millikan at the University of Connecticut in2008, a conference on thought experiments and the a priori in Forteleza, Brazil in 2009, aconference on imagination and the a priori at the University of St. Andrews in 2010, and ameeting of the Society for Philosophy and Psychology in 2011. I learned a great deal from thediscussions on all those occasions. I have also been helped considerably by the comments ofAlex Byrne on one of the earlier versions, and by advice from Jamie Dreier, Ivan Fox, HilaryPutnam, and Joshua Schechter. My deepest debts are to David Christensen, Anil Gupta, andVann McGee, who have placed their very considerable abilities at my disposal on a number ofoccasions.

ReferencesBennett, Jonathan. (2003). A Philosophical Guide to Conditionals. Oxford: Clarendon Press.Boghossian, Paul. (1997). “Analyticity,” in Bob Hale and Crispin Wright (eds), A Companion tothe Philosophy of Language. Oxford: Blackwell Publishers Ltd.

Carey, Susan. (2009). The Origin of Concepts. Oxford: Oxford University Press.Gibbins, Peter. (1987). Particles and Paradoxes. Cambridge: Cambridge University Press.Grice, H. Paul and Peter F. Strawson. (1957). “In Defence of a Dogma,” The Philosophical Review

65: 141–58.Gupta, Anil. (2008). “Definitions,” The Stanford Encyclopedia of Philosophy. <http://www.plato.stanford.edu/>.

Harris, John. H. (1982). “What’s so Logical about the ‘Logical’ Axioms?” Studia Logica 41: 159–71.Hill, Christopher S. (2002). Thought and World. Cambridge: Cambridge University Press.Horwich, Paul. (1998). Truth, 2nd ed. Oxford: Oxford University Press.Horwich, Paul. (1999). Meaning. Oxford: Oxford University Press.Jackson, Frank. (2010). Language, Names, and Information. Oxford: Wiley-Blackwell.Lewis, David K. (1970). “How to Define Theoretical Terms,” Journal of Philosophy 67: 427–46.Quine,Willard Van Orman. (1976). “Carnap and Logical Truth,” in Quine, TheWays of Paradoxand Other Essays, 2nd ed. Cambridge, MA: Harvard University Press.

Quine, Willard Van Orman. (1980). “Two Dogmas of Empiricism,” in Quine, From a LogicalPoint of View, 2nd ed. Cambridge, MA: Harvard University Press.

Quine, Willard Van Orman. (1986). Philosophy of Logic, 2nd ed. Cambridge, MA: HarvardUniversity Press.

Schechter, Joshua. (forthcoming). “Juxtaposition: A New Way to Combine Logics,” Review ofSymbolic Logic.

Wittgenstein, Ludwig. (2001). The Philosophical Investigations, 3rd ed. Oxford: Blackwell.


http://www.plato.stanford.edu/

http://www.plato.stanford.edu/

7

A Priori Testimony Revisited

Anna-Sara Malmgren

Let ‘A Priori Testimony’ be the thesis, introduced and defended by Tyler Burge (1993,1997, 1999), that there’s a priori knowledge by testimony—more precisely: that at leastsome of our actual testimonial knowledge is a priori. How does Burge argue this? Inlarge part, by giving a positive account of testimonial warrant and knowledge thatsustains it: an account that implies the thesis, and purports to explain how it could betrue. Burge’s case for A Priori Testimony, then, rests heavily on the plausibility of hispositive account.

In previous work, I contested the plausibility of that account (Malmgren 2006). Oneof the problems I identified is specific to it; one is more general—it arises, it seems, on anymodel of the epistemology of testimony on which A Priori Testimony comes out true.

Ram Neta has recently responded to the part of my criticism that targets Burge’saccount in particular (Neta 2010). In this chapter I argue that Neta’s response isinadequate. I also explore my second, more general, criticism a little further, andI discuss the prospects for vindicating A Priori Testimony in a different way—a waythat, at least at first sight, looks more promising than Burge’s: with appeal to (whatI call) the ‘Fast Track Model’ of testimony. I argue that this strategy faces difficultchallenges of its own.

The chapter is organized as follows: in }1, I restate the key components of Burge’saccount. In }2, I summarize—and further clarify—my previous objections to it. In }3,I reply to Neta. In }4, I consider a few other possible counter-arguments, andI introduce and examine the Fast Track Model.

11

1.1 The Proof Case

Two suggestive analogies may explain why A Priori Testimony even seems worthtaking seriously; here, let me just mention one: the analogy with relatively uncontro-versial cases of ‘perception-triggered’ a priori warrant or knowledge—cases where suchwarrant or knowledge is acquired by means of diagrams, formulae, or proofs (cf. 1993,

480; 1997, 37).2 Your specific perceptual experience of, say, a written representation ofa certain logical proof may help you access a warrant—an independently existing, apriori warrant—for believing a certain theorem; perhaps a warrant that you wouldn’t,even couldn’t, otherwise come to have. The experience can do this without (as it were)contaminating the a priori status of that warrant. If the warrant is strong enough, andother things are equal, you may even come to know the theorem—a priori—as aresult.This kind of case provides one of the clearest illustrations there are, of the triggering,

merely causal, or warrant-enabling role—as opposed to the justifying, or warrant-conferring role—of perception. It’s hard to give a principled account of the distinctionbetween these two roles, but at least we have a paradigm, in cases like this—a paradigmthat we think we understand.3 By assimilating the acquisition of testimonial warrantand knowledge to this paradigm, Burge may tempt us to think that we understand howit, too, could be a priori. Perhaps the perceptual states and processes that are involved inthe production (or sustenance) of testimonial beliefs only play a triggering role, in atleast some cases: they just facilitate our access to independently existing, a prioriwarrants. It’s a suggestive thought.In the end, Burge’s model of the epistemology of testimony doesn’t quite back up

the suggestive thought (that certain testimony cases are like that). As we will see, there’sa sense in which the model ‘gives us more than we bargained for’—if what webargained for was that thought. The admission that specific perceptual experiencescan play a triggering role, in cases like that of the proof, doesn’t commit us toexpanding the domain of the a priori beyond standard limits. (Indeed, it’s compatiblewith highly restrictive conceptions of that domain—say, as including only analyticallytrue propositions.) On Burge’s account of testimony, however, the domain is radicallyexpanded. Granted: the domain of a priori knowledge stays the same. But the domain ofa priori warrant now includes that zebras are larger than red poppies, that Bettie likes chocolatecake, that many children have imaginary friends—and any number of other, ‘deeply’contingent, propositions (cf. Evans 1985). This isn’t yet an objection to the view,but it’s worth taking note of.

1.2 Burge’s account

Burge is a pluralist about epistemic warrant: he holds that it comes in (at least) twovarieties. ‘Justification’ is necessarily cognitively accessible to the agent—perhaps evenas such: as warrant—whereas ‘entitlement’ needn’t be (1993, 458–9; 2003, 504).4 Thea priori/a posteriori distinction applies to both types of warrant, and a warrant is a prioriif “neither sense experience nor perceptual belief constitute nor enhance its [warrant-ing] force” (1997, 21). Knowledge is a priori, in turn, when it’s “underwritten by an apriori justification or entitlement that needs no further justification or entitlement tomake it knowledge” (1993, 458). And warrant or knowledge that’s not a priori is aposteriori. Burge also holds that at least some a priori warrant and knowledge isempirically defeasible (1993, 460–1). As will emerge, two warrants of this type—two

A PRIORI TESTIMONY REVIS ITED 159

a priori, empirically defeasible, warrants—take center stage in his vindication ofA Priori Testimony: the recipient’s default pro tanto entitlements to ‘rely on herunderstanding’, and to ‘rely on the rationality of her source’.5

In testimony, a propositional content is asserted—or in some other way ‘presented-as-true’—by a source, to a recipient. (There may of course be multiple sources and/orrecipients, but nothing here turns on this.) As Burge conceives of assertion, it requires alinguistic action—namely the spoken or written utterance of a suitable declarativesentence—but it’s unclear whether that goes for all presentations-as-true (or whether,say, the shaking of one’s head, or the drawing of a figure in the sand, qualifies in certaincircumstances). More generally, it’s unclear how broad the category of presentations-as-true is supposed to be: Burge explicitly includes obvious presuppositions andconventional implicatures, but leaves it open how to generalize from there (1993,482, fn. 20, 21). For much of what follows, however, we can ignore this complication,and focus on testimonial acts that do involve assertions—assertions that, let’s grantBurge, are all expressed by means of declarative sentences. (I’ll use ‘saying’, ‘telling’ and‘testifying’—in lieu of ‘presentation-as-true—as umbrella terms: to designate anycommunicative act that’s a potential source of testimonial warrant and knowledge.)6

A source may or may not have warrant to believe the content she asserts—if shedoes, and if her attempt to testify is otherwise successful, that warrant is passed on to therecipient: it’s ‘preserved’ in the testimony (1993, 466, 486). But it’s not required that thesource has warrant, and that this warrant is preserved, for the recipient to havewarrant—or warranted belief—by that source’s testimony. For the recipient to know,the source must have warrant: indeed, she must in turn know (what she asserts)—andin some other way than by testimony. (At least there must be a source with non-testimonial knowledge somewhere down the chain; 1993, 485–6, fn. 24. I’ll treat thisqualification as understood.) But all it takes, for the recipient to have testimonialwarrant, is that her two default pro tanto entitlements are in place: her entitlement torely on her understanding—henceforth: the ‘u-entitlement’—and her entitlement torely on the rationality of her source—the ‘r-entitlement’. And all it takes for thoseentitlements to be in place is that she lacks (undefeated) defeaters.7

The two entitlements can be in place—remain undefeated—even if what the sourceasserts is distinct from what it seems to the recipient that she asserts. (In such cases, thetestimonial warrant is a warrant to believe the ‘seeming’ content.) On Burge’s picture,then, testimonial warrant doesn’t require correct identification, by the recipient, of theasserted content: it’s compatible with her mishearing or misunderstanding what thesource said. In fact, it doesn’t even require that a content—any content—is (or was)asserted, by the ostensible source or by anyone else: a vivid enough illusion orhallucination, as of someone saying that p, may other things equal yield testimonialwarrant to believe that p (1997, 24–6).

Testimonial knowledge is more demanding, in these respects as well: it does requirethere to be a source, and it requires the recipient to correctly identify both content andforce of that source’s speech-act—say, its being an assertion that p (1993, 480, fn. 19). It

160 ANNA-SARA MALMGREN

doesn’t, however, require her to correctly identify the words that figure in the source’sutterance. And it’s unclear whether it requires her to identify the source (either as hersource, or as anything else—under any other mode of presentation). It’s also unclearwhat the precise relation is between the correct identification—or, in Burge’s preferredphrase, ‘correct understanding’—of a given token speech-act, and the knowledge orwarranted belief that a speech-act with that content and force is (or was) performed.We’ll return to this.The two default entitlements form part of the recipient’s ‘proprietary’ warrant to

believe a given proposition. As already indicated, they’re not only jointly sufficient butindividually necessary for (her to have) testimonial warrant; hence they’re necessary fortestimonial knowledge as well. (In cases where both entitlements are indeed in place,and the source in turn has warrant, the source’s warrant and the recipient’s proprietarywarrant together combine to form the recipient’s ‘extended’ warrant; 1993, 485–6.)Next, for the recipient to have a priori testimonial warrant or knowledge, her

entitlements must not just be in place—they must be a priori. But Burge arguesthat they’re indeed a priori, provided that two further conditions are met: first, therecipient doesn’t depend on any empirical defeater-defeaters (for the entitlements tobe in place)—either she has no defeaters, or her defeaters are defeated by a prioridefeater-defeaters—and, second, her understanding of the target utterance is intellectual(1997, 22).What’s intellectual understanding? Burge tells us that:

[ . . . ] it is understanding whose exercise in particular instances does not require in those instancesperceptual warrant for the application of what is understood. A first approximation elaboration isthat it is conceptual understanding that does not require, in thinking and understanding anintentional content, perceptual warrant for the de re application of some aspect of the content.(1997, 21, italics original)

Elsewhere it’s further explicated as understanding that only depends on our “standinglinguistic abilities” or “standing linguistic competence” (1999, 234, italics in original).8

And Burge takes a very liberal view on what that competence is able to deliver. Itdoesn’t, he grants, suffice to understand utterances of sentences that contain certaincontext-sensitive expressions: expressions whose understanding (on given occasions ofuse) require perceptual identification of their referents—paradigmatically: perceptualdemonstratives and pure indexicals. Nor does it suffice to recover certain indirectspeech-acts: speech-acts that differ in content and/or force from those that constitutethe ‘default readings’ of the uttered sentences—paradigmatically: conversational im-plicatures (1993, 483; 1997, 22; 1999, 234, 245). But, with these two caveats, Burgeclaims that “[t]he standing mechanism yields an enormous amount of understanding”(1999, 235).Importantly, it doesn’t just yield understanding of content, but also, frequently, of force

(1993, 482; 1997, 22, 44, fn. 3). In particular, our understanding of assertoric force isfrequently intellectual. That, we’re told, is because there’s a ‘conceptual connection’


between assertoric force and declarative mood (1993, 482). And keep in mind thatwhat’s at issue here isn’t just our understanding of sentence-types—but of specificdated utterances (or sentence-tokens). The claim is that your grasp of both content andforce of the speech-act expressed by a specific dated utterance of, say, ‘zebras are largerthan red poppies’ may well be intellectual—that your standing linguistic competencemay suffice to identify that particular speech-act as an assertion that zebras are larger thanred poppies. (This example is Burge’s own; 1997, 22.) Likewise—I presume—for yourunderstanding of a specific dated utterance of, say, ‘16 x 78 = 1248’, ‘Norway wasoccupied duringWWII’, ‘Bettie likes chocolate cake’, or ‘many children have imaginaryfriends.’

For concreteness, let ‘AR’ (for ‘a priori recipient’) be a recipient who has what ittakes to be a priori u-entitled, and a priori r-entitled: AR’s understanding of hersource’s utterance (or apparent utterance, or apparent source’s utterance . . . ) is intel-lectual, and AR has no undefeated or empirically defeated defeaters. Note that that’salready enough to ensure that AR has some a priori warrant to believe the content ofthat utterance—say, that Bettie likes chocolate cake. Whether AR has all-things-considered a priori warrant partly depends on the status of the (eventual) warrant thatis passed on from her source. It also depends on the status of AR’s supplementarywarrant, if any: if AR has no supplementary warrant (to believe that content), or if shedoes but it’s a priori as well, the status of AR’s all-things-considered warrant is a priori.But if AR has a posteriori supplementary warrant, then, of course, her all-things-considered warrant is a posteriori too.

Suppose AR falls in the first mentioned category: she has no supplementary warrantat all to believe that Bettie likes chocolate cake. Then the only remaining variable is thestatus of her source’s warrant: if the source has a posteriori warrant, then, again, AR’sall-things-considered warrant is a posteriori. But if the source has a priori warrant, ARis in the clear: her all-things-considered warrant is a priori. Likewise if the source has nowarrant—in that case too, AR is in the clear, since her all-things-considered warrant isexhausted by her default a priori entitlements.

Thus there are—crudely put—at least two different ways to come by all-things-considered a priori warrant by testimony: one is to use a source with a priori warrant;another is to use a source without warrant. (A third is to use a merely apparent source,and/or a merely apparent utterance, since, in such cases too, no warrant at all—henceno a posteriori warrant—is preserved in the testimony.) But, Burge tells us, there’s onlyone way to come by a priori knowledge by testimony: for that, one needs a source whoherself has—not just warrant, or even a priori warrant—but a priori knowledge of thecontent in question: a priori knowledge that’s not, in turn, testimonial (1993, 486–7).9

AR has a priori knowledge by testimony, then, just in case there’s indeed a source,the content that AR takes that source to assert is the very same content that the sourceasserts, and the source has a priori, non-testimonial knowledge of that content. Thefirst two conditions, recall, are requirements on testimonial knowledge in general. Andthe third condition has an interesting consequence: that what we can know a priori by


testimony comes apart from what we can have a priori testimonial warrant to believe(cf. 1997, 44 fn. 2). Presumably, our sample content is a case in point: that Bettie likeschocolate cake can’t be known a priori by testimony, since it can’t be known a priori inany other—any non-testimonial—way. So, given that this is by hypothesis the contentthat’s at issue here,AR doesn’t have a priori testimonial knowledge, however else we fillout the details of the case. At most AR has all-things-considered a priori (doxastic)warrant.But the immediate focus in what follows is on warrant—not knowledge. If, as

I maintain, it’s not even possible to be in AR’s shoes,10 then, of course, it’s not possibleto have a priori knowledge by testimony.

22.1 Objections (first round)

In Malmgren 2006, I argued that there’s no plausible conception, available to Burge, ofthe object of the first default entitlement—no plausible answer to the question “whatdoes the u-entitlement entitle the recipient to do?” My argument was, in effect, anargument by elimination: I considered every (at least minimally feasible) candidateanswer I could think of, and criticized each in turn (Malmgren 2006, 212–24). Two ofthe candidates will be repeated shortly.Next, I argued that Burge fails to accommodate the presumption that our

knowledge of what’s said (‘of the specific type’)—say, your knowledge that Maxjust said that Bettie likes chocolate cake—plays an epistemic role, vis-a-vis ourtestimonial knowledge, and that the failure to do so commits him to a peculiarpicture of the recipient’s motivational psychology (Malmgren 2006, 224–30).(But the fundamental point concerns warranted belief—not knowledge. Seebelow.) The second objection is largely independent of the first. It’s alsomore general—on the face of it, it arises on any account that sustains A PrioriTestimony.Burge is clear about what the recipient’s second default entitlement—the r-entitle-

ment—is a warrant to do (Burge 1993, 467–73; 1997, 21): it’s a warrant to accept astrue the content of an arbitrary source’s assertion (or apparent assertion, etc.11). Moresimply: it’s a warrant for her to believe what she was told, by a given source on a givenoccasion—say, that Bettie likes chocolate cake. That the recipient must have warrant tobelieve that content, in order to have testimonial warrant/knowledge, is obviously notcontroversial. It’s not controversial since that’s precisely the content she has testimonialwarrant to believe, or testimonial knowledge of, if she does.12

It’s tempting to suppose that the first default entitlement—the u-entitlement—islikewise a warrant for the recipient to believe something; specifically: a warrant tobelieve that the given content is (or was) indeed asserted, by the given source on thegiven occasion. It’s tempting to suppose so, because it’s independent plausible that


testimonial warrant and knowledge requires warrant to believe a claim of this sort—roughly: that you can’t have testimonial warrant to believe what you were told (say,that Bettie likes chocolate cake) unless you have warrant to believe that you were toldit, by so-and-so then-and-there (say, that Max just said that Bettie likes chocolatecake). So this interpretation makes sense of the alleged necessity of the u-entitlement. Italso looks—in at least some places—like it captures Burge’s intent (e.g. Burge 1993,476; 1999, 235).

The main problem is that it’s very hard to see how, if this reading were correct, theu-entitlement could ever be a priori: how anyone, otherwise relevantly like us (cf.footnote 10), could have a priori warrant for believing a content such as that Max justsaid that Bettie likes chocolate cake. On the face of it, this is no more plausible when, byBurge’s criteria, the target utterance is capable of being intellectually understood, thanwhen it isn’t (cf. Malmgren 2006, 216–17). And note that the point doesn’t rest on anydemanding conceptual requirements on believing—and, derivatively, on having warrantto believe—the content in question: requirements to the effect that one be able to referto the time and place of the given assertion under any other mode of presentation than,say, sometime recently someplace not too far away, or right here now, or refer to the sourceunder any other mode of presentation than, say, my source, or the speaker, or whoeverwrote this book.

Another option—perhaps, in the end, Burge’s preferred option—is that what the u-entitlement entitles the recipient to do is to believe that the given content is (or was)asserted by some source or other, at some point or other. More simply: to believe that she wastold it, but without any commitment at all as to where, when and by whom—say, thatsomeone or other, at some point or other, and in some way or other, said that Bettielikes chocolate cake. (Cf. “What one is entitled to on intellectual grounds is merely,prima facie, that a given content is presented as true. One gets nothing about the time,form, or circumstances of the assertion” Burge 1993, 483; second emphasis added.)

However, this suggestion is hardly an improvement on the first: it’s no easier to seehow anyone could have a priori warrant to believe a content about what’s said of this—less ‘committal’ or specific—type, than to believe a content about what’s said of themore specific type. There are of course several ways, easily available to a typicalrecipient of testimony, to come by warrant to believe it—perhaps most obviously: bycompetent deductive inference from (warranted belief in) a content of the morespecific type—but none that, it seems, would or could deliver a priori warrant for herto believe it.13

The problem here isn’t, or not just, that Burge himself fails to give an adequateexplanation of how the testimonial recipient could have a priori warrant to believewhat—on either of these readings—she sometimes, supposedly, has a priori warrant tobelieve. (Although that, too, is the case in my view.14) It’s that there is no adequateexplanation in the offing: indeed, we don’t even know how to begin to explain whatneeds explaining. That, at any rate, is what I argued in my previous piece (Malmgren215–17).


But, as we’ll see, this is the point at which I’m being challenged by Neta. Accordingto him, I overlook15 a ‘perfectly good’ explanation of how it’s possible for a recipient tohave a priori warrant to believe a content about what’s said of the less specific type—schematically: that someone at some point said that p.16

2.2 Objections (second round)

Let a ‘successful recipient’ be an agent who at least has doxastic testimonial warrant: anagent who is warranted in believing, and/or knows, that p by testimony. My secondobjection to Burge starts with the thought that any such agent is (also) warranted inbelieving a content about what’s said of the more specific type—schematically: that hersource said that p.17 But her warrant for believing that her source said that p is inevitably aposteriori.The next step is that the successful recipient’s warranted belief that her source said

that p plays a causal-explanatory role in the formation of her testimonial belief that p.More simply put: that she believes what she’s been told at least in part because shebelieves that she’s been told it, by so-and-so then-and-there.)18

Third, her belief that her source said that p isn’t ‘any old cause’: it has the right kindof content to provide her with a (good, pro tanto) reason—a warrant, or partialwarrant—to believe that p. There’s obviously no deductive relationship between thecontents of these two beliefs, but I take it that that’s not required for a proposition to bea reason to believe another. And, for all I’ve said, there are numerous contexts wherethe content that her source said that p isn’t a reason (or is a fully defeated reason) for agiven agent to believe that p—but when that agent is a successful recipient, with respectto p, she’s in a context where it is a reason.19

Fourth: that her source said that p is a reason that the successful recipient has: it’s areason of hers to believe that p. Perhaps the fact that she believes it, or that she haswarrant to do so, is already enough to ensure this. But it also comes out true on morestringent views of what it takes to have a reason: views that require warranted belief. (Ifshe knows what’s said, it comes out on even more stringent views.20)A natural way to pull all of this together is by saying that the proposition that her

source said that p is among the successful recipient’s reasons for believing that p—heroperative reasons: the reasons-for-which she believes that p—or, equivalently, that herbelief that p is at least partly based on her belief that the source said that p. This, ofcourse, fits nicely with the observation that a typical recipient, if she’s the leastreflective, may well cite the fact that her source said that p as her reason, or one ofher reasons, for believing that p—say, she may respond that Max just told her so, whenasked how she knows that Bettie likes chocolate cake (or why, for what reason, or onwhat grounds, she believes it). On this way of pulling things together, she simply comesout right.But Burge can’t avail himself of this, natural picture: it’s in tension with A Priori

Testimony. If the recipient’s belief that p is based on her empirically warranted beliefthat her source said that p, then her belief that p is empirically warranted too (hence, at


best, constitutes a posteriori knowledge). So Burge had better deny that her one beliefis based on the other—or, at least: deny that that’s the case across the board.One option here is to grant that what I called the ‘natural picture’ accurately captures

the motivational psychology of the canonical or typical (successful) recipient, but arguethat there are exceptions: that at least some successful recipients—importantly: recipi-ents whose testimonial beliefs constitute knowledge—differ from the norm in thisrespect. Another option is to deny that the natural picture captures the structure ofthe typical—or even any!—successful recipients’ reasons and beliefs.Either way, however, Burge is now in an uncomfortable spot. Take a specific,

successful recipient whose belief that her source said that p is a cause of her (warranted)belief that p, and provides a reason for her to believe that p—a recipient who has thatreason, but whose belief that p isn’t based on her belief that her source said that p.What’s the explanation? Why is her testimonial belief not based on her belief aboutwhat’s said? Burge—and, more generally, the A Priori Testimony advocate—owes usan answer to this question. Let me clarify why it’s a difficult burden to discharge.

The case as described is structurally similar to the familiar counter-examples—abundant in the literature—to a straightforward causal analysis of the basing-relation(and/or of operative reasons): cases of ‘reason-independent causation.’ Various com-peting diagnoses have been offered of what exactly goes on, or goes wrong, in cases likethese—cases where a suitable belief (or other intentional state) figures in the causalexplanation of another belief (or other state, or action) without being part of its base.One broad family of diagnoses appeal to the notion of deviant causation; another familyintroduces further, or substitute, conditions on basing.21

My challenge to the advocate of A Priori Testimony isn’t just a challenge to take astand in this debate; that would be relatively uninteresting—not because the debateconcerns an uninteresting problem, but because it concerns a problem that, at least onthe face of it, can be expected to have a solution. Rather, the challenge is to make goodon the claim that some cases of testimony—indeed, of testimonial knowledge—are casesof reason-independent causation, whatever the correct analysis of that phenomenon is.The assimilation has little face-value appeal, as a quick glance at some of the familiarexamples should make clear.22 And it doesn’t increase in appeal on any of the extant,available diagnoses.

Suppose the first broad family of diagnoses—diagnoses that appeal to causal devi-ance—is on the right track; then, to a first approximation, what makes all the familiarcases cases of reason-independent causation is the presence of some kind of interferencemechanism: a mechanism that interferes with a salient subset of the agent’s normalcausal–psychological processes, and that’s crucially involved in the production of thefeatured effect.23 In this framework, the suggestion—the suggestion in need of furtherdefense—would be that some such mechanism (or set of mechanisms) is involved,across the relevant range of testimony cases: an interference mechanism that explainswhy the recipient’s belief that p isn’t based on, although caused by, her belief that hersource said that p—a mechanism that, crucially, doesn’t prevent her from knowing that


p by that source’s testimony. What would the mechanism be? At the very least, weneed to be provided with more detail. And note that—even at this level of abstrac-tion—the suggestion is quite odd: that a prerequisite for a priori warrant and know-ledge by testimony is that the testimonial belief be partly explained by the workings ofan interference mechanism. Of course, matters are worse if the scope of the suggestionis typical testimony.A competing suggestion, belonging to the other family of views, would be that some

higher-level requirement or other on basing has been violated: that, in the cases ofinterest, the recipient lacks the appropriate attitude—perhaps awareness, belief, orwarranted belief—to the status of the claim that her source said that p: its being areason (for her) to believe that p. But this suggestion too is, at best, peculiar: that theonly recipient who’s able to have a priori warrant and knowledge by testimony is therecipient who, loosely put, gets her reasons wrong—who fails to realize that a certainreason to believe that p, indeed one of her reasons to believe that p, is indeed a reasonfor her to do so. (Or, if the requisite attitude is warranted belief, who fails to rationallyrealize this: who lacks warrant altogether to believe that p, or who does have warrantbut believes it on other grounds.) And, again, matters are worse if the scope of thesuggestion is typical testimony.Matters are worse if that’s the scope since it’s very implausible that the typical,

successful recipient fits either of the above (minimal, theoretical) descriptions.24 At amore intuitive level: it’s implausible that that recipient’s motivational psychology isrelevantly similar to that of agents featured in stock examples of reason-independentcausation. The principal problem, however, applies regardless of scope: it’s that thesuggestion commits us to there being at least some testimonial recipients whosemotivational psychology is relevantly similar to that of such agents—but who never-theless know the target proposition (by testimony). It’s by no means obvious how tomake good on this suggestion, given that the agents in the stock examples don’t evenhave doxastic warrant (where applicable): their beliefs aren’t properly based, and so notwarranted, even when all other requirements on warranted belief are met.25

I return to these considerations in }4.

33.1 Neta’s proposal

Neta doesn’t address my second objection (just rehearsed). But he offers an interestingreply to the first—to my objection that Burge has no plausible conception available ofthe object of the u-entitlement: no good answer to the question what it’s a warrant todo.Neta argues that I’m too quick in rejecting the suggestion that it’s a warrant to believe

a content about what’s said of the less specific type: that someone at some point saidthat p. (Or, in the locution he borrows from Burge: ‘that a rational source presented


is as true that p’.26) In response to my main complaint about that option, Neta argues asfollows:

[w]hat I’d like to focus on is Malmgren’s suggestion [ . . . ] that we have no way to explain how itis possible for someone to whom it seems that a rational source has presented it as true that p to bea priori warranted in believing the contingent proposition that a rational source has presented it astrue that p. It seems to me that, on the contrary, we have a perfectly good explanation of this,which I’ll now outline by means of an example.Suppose that I present it as true that, say, Washington, DC is the capital of the United States. If

I do this under normal circumstances, then, while I might not be a priori warranted in anyparticular beliefs concerning the medium by which I presented this proposition as true [ . . . ],I can at least be a priori, defeasibly warranted in thinking that I have (somehow or other)presented it as true that Washington, DC is the capital of the United States: my warrant forbelieving this contingent proposition concerning what I have presented as true is of the same sortas my warrant for believing other contingent propositions concerning my communicativeintentions, and it is not typically constituted by my sensory or perceptual experiences. It iscrucial to recall here that for an agent to present something as true does not require that the agentmake any noises or bodily movements: Burge leaves open that an agent can present something astrue by means of extra-sensory perception. Indeed, in at least one passage he suggests that judgingthat p is a way of presenting it as true that p: “I use the term ‘presentation as true’ to cover morethan assertions and judgments” (Burge 1993, 482, emphasis added). If this is right, then I canpresent something to myself as true simply by thinking of it as true, i.e., judging that it is true orotherwise affirming its truth in thought.Now, if I can be a priori warranted in thinking that I have presented this proposition as true,

then I can also be a prioriwarranted in thinking that I have been given to understand this proposition astrue. But the following conditional is a priori: if I have been given to understand something as true,then a rational source (maybe me, maybe another rational source) has given it to be understood astrue, or in other words, has presented it as true. (Neta 2010, 209–10, italics in original.)

The passage continues with the attribution of some further reasoning to the (putative)recipient—reasoning that’s required on Neta’s reconstruction of Burge’s account. Butthere’s no need to rehearse that here (and nothing turns on the exegetical disagree-ment). The quoted passage appears to already contain what I’d asked for: an explan-ation—or at least an explanation sketch—of how anyone (relevantly like us) couldhave a priori warrant to believe that someone at some point said that p. But is it aplausible sketch? And is it applicable to the case of interest? I’ll consider these questionsin turn.

On the face of it, the force of Neta’s case essentially depends on the fact that theagent who has the a priori warrant is the very same agent who performs the assertionthat makes the target claim true (or, at least, that this seems to her to be the case):the same agent whose ‘saying’—here: judging—that p makes it true that someone atsome point said that p. It’s because, but only because, she has warrant for believing thatshe said that p, that she has warrant for believing the more general claim. Moreprecisely: the case seems to be a standard case of inferential warrant, where the agent’s


warrant to believe one proposition (here: that someone at some point said that p) partlyrests on, or derives from, her warrant to believe another (here: that she said that p)—aproposition that’s a reason for her to believe the first. On usual assumptions aboutinferential warrant, her warrant to believe the more general proposition is a priori onlyif her warrant to believe the more specific proposition is. And, now, it’s not implausiblethat the latter warrant is a priori—provided that introspectivewarrant is a priori, at least inthe case at hand.So let’s grant that it is. To further simplify: let’s grant that introspective warrant is

a priori across the board, and that one can have such warrant for believing that oneis currently ‘thinking of p as true’ or ‘judging that p.’27 (I take these phrases to beequivalent to ‘occurrently believing that p’.) If, as Neta suggests, thinking of p as true isjust another way of saying that p, one thereby has a priori warrant for believing that oneis saying that p, which transparently entails that someone at some point did. Presum-ably, this little argument is able to constitute a ‘complete’ inferential justification; i.e.one doesn’t need warrant—a priori or a posteriori—for believing any further propos-itions, to have warrant for believing its conclusion (given warrant for believing thepremise).28 But then, if other things are equal, no a posteriori element is introduced,and so our agent—the agent who has (introspective) a priori warrant for believing thepremise—has a priori warrant for believing the conclusion too: that someone at somepoint said that p.It’s unclear whether introspection affords any other suitable a priori route to the

desired destination. Perhaps it’s possible to have introspective warrant to believe thatone, just a moment ago, intended to say that p. Although intending to say somethingisn’t itself a way of saying it—not even on the present, highly permissive conception ofsayings—this, too, could plausibly constitute the first step in an (inferential) warrant tobelieve that someone at some point said that p. What’s doubtful, however, is that itcould constitute the only step—in particular, that there’s a way to ‘move’ from it, to(warrant to believe) the conclusion, without the benefit of any auxiliary empiricalwarrant. (E.g. to believe that one usually says what one intends to say? That onefrequently did in the past?) The proposed warrant-structure can be elaborated in avariety of ways, consistent with the rationality of the corresponding transition. But it’svery controversial that it—in contrast to the previous structure—needs no elaborationat all: that it’s complete as it stands. And, on all the obvious ways of elaborating it, theagent has warrant to believe further propositions: warrant that could only be empirical.Likewise for routes from (introspective warrant to believe) that one now intends to

say that p, that one remembers that one said that p, that one remembers that someone atsome point said that p, that one believes that someone at some point said that p—and soon.29 In each case, considerable further work is needed to support the claim that it’spossible to rationally move from the suggested starting-point, to the desired end-point,without any auxiliary a posteriori warrant on the agent’s part.As I read it, Neta’s example does not introduce this complication. It does however

depend on the permissive conception of sayings. But I’ll go along with that for now.30


Next: is the proffered kind of a priori warrant, to believe that someone at some pointsaid that p, available to a testimonial recipient—specifically: to a successful recipient, ofthe testimony that p? It seems clear that it is. We can stipulate that the character inNeta’s case has testimonial knowledge or warrant for believing the target proposition(that Washington, DC, is the capital of the US); on the face of it, that presents noobstacle at all to her also having a priori warrant, of the sort we’ve sketched, forbelieving that someone at some point said it.

The crucial question, however, is whether that a priori warrant would or could beexplanatory of her testimonial warrant/knowledge: whether it’s possible to be a suc-cessful recipient of the testimony that p partly because of, or in virtue of, the fact that onehas the proffered a priori warrant for believing that someone at some point said this.And the answer to that question seems to be ‘no.’ Not because we can’t testify toourselves—we can—but, first, because the putative recipient’s (a priori) warrant tobelieve that someone at some point said that p would at best be idle in the explanation:all the work would be done by her (a priori) warrant to believe that she, or her source,said that p. Bracketing exegetical concerns, that’s all well and good; we could simplyregard Neta’s proposal as an answer to the question how the u-entitlement could be apriori—if it’s a warrant to believe a proposition about what’s said of the specific type:that one’s source said that p. But, second, it’s not clear that what would be explained, inthe most favorable case, is testimonial warrant or knowledge: it lacks a certain pattern ofundermining defeat that’s arguably distinctive of testimony.

3.2 Testifying to oneself

Suppose you dust off my old diary, and read what I wrote twenty years ago. Presum-ably you could, by doing so, acquire testimonial knowledge of—or warrant forbelieving—any number of things about my past. Likewise, it seems, if I dust off myold diary, and read what I wrote back then. I mustn’t mistake the diary for a piece offiction, have doubts or reasons to doubt the author’s sincerity, etc. But all of that, ofcourse, goes for you as well (in the scenario where it’s you who reads it). The fact thatI’m the author doesn’t, by itself, seem to make a relevant (negative) difference; nordoes it make a difference that—as we may further suppose—I believe, or even know,that I am.

For another prima facie example (of what I’ll call ‘self-testimony’), consider a case ofthe familiar sort in which you consult the to-do-list, the recorded message on yourphone, or the notes in your agenda, that your past self composed to aid your memory:to help you remember that p. On the face of it, there are circumstances where this,too—the consultation of your own reminder—provides you with knowledge orwarranted belief that p: knowledge or warrant by (your own) testimony.31

There’s no need to suppose that the agent, in a case like this, has a priori warrant forbelieving that someone at some point said that p—or, for that matter, that she inparticular did. It’s perfectly possible that she has some such warrant (on the assumptionsmade above). But we needn’t imagine the example in that way, to make it plausible


that she has knowledge or warranted belief that p, and that that knowledge/warrant istestimonial. It does, indeed, seem necessary to imagine it in such a way that she haswarrant—somewarrant—for believing that her source said that p. On the obvious readingof the case, however, she has that: she has a posteriori warrant of the usual kind—thekind that’s available to potential recipients other than herself. It’s a warrant that owessome of its ‘justificatory force’ to her experience of a specific utterance; here, an utterancemade (or apparently made) by her past self. The exact details do not matter.But perhaps there are other cases—other prima facie instances of warrant or knowledge

by self-testimony—in which the corresponding a priori warrant does play a crucial role?Here is a candidate. Suppose you forgot all (external) reminders at home, but that a littleintrospection reveals that you’re thinking of it as true—that you occurrently believe—that p. (To screen off noise, we may also suppose that you lack salient independentreason to believe that p.) Could this, too—the ‘inner’ consultation of your own belief—result in knowledge of, or warranted belief, that p? Could your introspective evidencethat you believe that p be a reason for you to (continue) to believe it? It does in fact seemplausible that it could. Of course, various additional requirements must be satisfied—here, as in the other examples—but even on fairly demanding views of what thoserequirements are, it looks like the case could be realized in such a way that they’re met.32

(E.g. perhaps you have, and/or have evidence that you have, an excellent relevant trackrecord, or perhaps p’s truth is part of the best available explanation of why you believe it.)But would the warrant or knowledge that you’d have, in a suitable realization of the

case, be testimonial? (And if not, why not?)The act of saying, telling, or testifying—the type of communicative act that’s

supposed to be the source of distinctively testimonial warrant/knowledge—oftenreceives an initial gloss that precludes this: a gloss that stipulates it to involve linguisticor other overt behavior. (Burge is a notable exception here.) Occasionally it’s also partof the initial gloss that source and recipient are distinct (and/or take one another to bedistinct); if that’s correct, even seemingly straightforward examples of self-testimony—like the diary- and reminder-case—are in question.33 However, it’s unclear whetherany such gloss deserves to be treated as final word. Indeed, it’s unclear how to approachthe matter—what the proper methodology is. I take it that it shouldn’t be settled byfiat; substantial issues might hinge on the classification of our three examples (and,more generally, on how we demarcate the target domain). Perhaps the best theoreticalframework—our best, future, overall epistemology—will assimilate all three to canon-ical cases of warrant/knowledge by testimony; perhaps only some, or none. Perhapsit’ll offer alternate groupings, depending on the questions asked. (There might, ofcourse, be reasons to close off some options in advance. At present I’m just questioningthe advisability of doing it by stipulation.)So, how to proceed? One approach is to directly investigate the nature or ‘meta-

physics’ of testimony, and of what it is to testify, in abstraction from any epistemo-logical concerns—perhaps by reflecting on a wide range of examples of differentcommunicative acts and behaviors, and consider which of them intuitively count as


testimony. (There are some attempts at this in the literature.34) But I doubt that any,suitably general, questions about the metaphysics of testimony are fruitfully pursued inisolation from questions about its epistemology—in particular, that we can makeprogress with the classification of hard cases in this way. The research project (at leastas carried out to date) appears to presuppose that there’s a determinate notion oftestimony—or, more precisely, of ‘natural’ or ‘informal’ testimony (cf. Coady 1992;Lackey 2006)—a notion that we have a firm, independent grasp on, and that maps on tothe right domain: testimony as a source of a distinctive type of warrant and knowledge.But why think so?35

What I propose, instead, is to focus on paradigm cases of testimonial warrant/knowledge, and to seize on a notable—recognizably epistemic—feature that they allseem to share: a feature that can be acknowledged without appeal or commitment toany specific theory of the epistemology of testimony. The specific feature that I have inmind is a certain defeater-structure: a characteristic pattern among the first-orderconsiderations that are able to constitute (normative) ‘underminers’.36 On the workinghypothesis that this isn’t an accident—but that the shared pattern is, or reflects someproperty that is, part of what sets testimonial warrant/knowledge apart from warrant/knowledge of other types—we can get a little further with the present issue.37

Canonical testimonial warrant and knowledge is potentially undermined by at leastthree distinct kinds of considerations: evidence that the source is incompetent (e.g. thather belief in the content she asserts is false, and/or unreliably generated or sustained);evidence that she’s insincere (e.g. that she’s lying or acting); and evidence of relevantcommunication-failure (e.g. of misperception or misunderstanding, on the recipient’spart). These considerations interact in interesting ways, come in varying degrees ofgenerality, and may take somewhat different form, depending on whether it’s warrantor knowledge that’s at issue—but none of that matters here.38 What matters is that, inparadigm cases, warrant and knowledge by testimony is vulnerable to all three, broadvarieties of undermining defeat. That shouldn’t be controversial: everyone—regardlessof their theoretical views on the epistemology of testimony—can agree on it.

Now, it seems clear that this goes for the diary-reader’s warrant (or knowledge) aswell, in our first prima facie case of self-testimony: her warrant is potentially underminedby evidence of incompetence, insincerity, and communication-failure—even thoughshe is, and/or thinks that she is, her own source. It’s not hard to specify what wouldconstitute such evidence, or to describe specific circumstances where her warrant isindeed defeated by it. (Suppose that, on second read, a crucial phrase seems to havebeen used ironically, or that she recalls a tendency to brag on certain matters in herentries . . . ) Likewise for our second example—where the agent consults her ownreminder—and, indeed, for all others I can think of where the putative recipient’swarrant to believe that her source said that p is empirical warrant of the usual kind.The only case that differs is the one that Neta needs: the case where the correspond-

ing warrant is introspective and a priori. Arguably, it doesn’t differ with respect toevidence of incompetence, in the (putative) source; here, manifested as evidence that


some or all of one’s introspectively accessible beliefs are, for instance, the product ofwishful thinking. But there seems to be no room for defeaters of the second and thirdvariety. Insincerity and communication-failure requires a certain vehicle that just isn’tthere—nor is rationally taken to be there (with one qualification, to come).What would count as evidence that the source is (or was) insincere, in testifying that

p, in the case where ex hypothesis she testifies simply by judging—occurrently believ-ing—that p? A belief cannot be insincere. It can’t even coherently appear insincere (tothe first person, or anyone else): only the expression of belief can. Attitudes of othertypes can misleadingly appear to be beliefs (and conversely)—but evidence that somesuch possibility is realized isn’t yet evidence of insincerity in a belief.Next, what would count as evidence of relevant communication-failure? Evidence

that threatens to defeat (here: undermine or rebut) an agent’s introspective warrant tobelieve that she ‘said’ that p.39 Canonical testimonial warrant/knowledge is vulnerableto defeat by evidence that one, say, misheard or misinterpreted the utterance by whichthe source said, or seemed to say, that p. But, in the case of interest, there is no suchutterance; nor is there a gesture, a drawing—or, it seems, any other suitable middleman.

To clarify: the very possibility of relevant communication-failure, here, seems tostand or fall with a certain controversial view of introspection: a view on which ourintrospective access to (the occurrence, content, and attitude-type of) our mental statesis mediated by access to representational vehicles—vehicles that, importantly, requireinterpretation, by the first person, as well as ‘inner’ perception (or independent apprehen-sion of some other kind). Otherwise they couldn’t be misperceived, misunderstood, or, itseems, admit of any other recognizable form of communication-failure.That view has very little prima facie plausibility. Of course we sometimes ‘think in

words’ (or, for that matter: in pictures, sounds, or tangibles), but it’s unclear whetherthis phenomenon is of significance for the epistemology of introspection—and, even ifit is, it doesn’t seem possible to fail to understand those words: to misidentify the contentor force of an utterance in inner speech.40 At most, one can misidentify the words, orother symbols, that figure in one’s (inner) utterance—as when, upon speaking out loudone realizes that one wasn’t thinking in the language one took oneself to be thinkingin. (Note also that evidence of such misidentification doesn’t defeat one’s introspectivewarrant to believe that one thought such-and-such.)Misleading evidence can undermine too. But once again: what would it look like?

On the face of it, any such evidence would have to include—or operate against thebackdrop of—evidence that the controversial view of introspection just outlined iscorrect. Let’s give our agent that; then perhaps, in light of further details, it could berational for her to think that relevant communication-failure has indeed occurred—say, that she’s misunderstood a crucial inner utterance. (Apparent expert testimonymight suffice.) That, in turn, might undermine her introspective warrant to believe thatshe said that p—and, derivatively, her warrant to believe that p.41

But if that’s what it takes, it takes too much. It is not enough to point to cases wherethere’s reason to believe a substantive, controversial claim about the etiology of


introspection—even granted such cases are possible. They’re too cheap. To a firstapproximation: for any class of beliefs c and method m, such that it’s possible to have(sufficient) evidence that one’s c-beliefs are formed or sustained at least partly by m,there’s a set of possible underminers—a set of first-order considerations that, for anyonewho has that evidence, are able to constitute underminers (of whatever warrant her c-beliefs do, or would, enjoy). So, to make good on the present suggestion, we need tocome up with a cleaner example: a cleaner example where the purportedly testimonialwarrant/knowledge is vulnerable to defeat by evidence of communication-failure (aswell as by evidence of insincerity and of incompetence). I can’t think of one.

I’ve given Neta’s proposal my best shot. The most it can do is help explain a certainkind of warrant/knowledge that arguably doesn’t qualify as testimonial. And if it weretestimonial, it would be self-testimonial. The proposal could still be used in a vindica-tion of A Priori Testimony—but it would be a vindication with the somewhatdisappointing rider that what makes the thesis true is that a certain kind of self-testimony is sometimes a priori. Furthermore, what does the explaining of the agent’swarrant/knowledge, in the most favorable case, is her a priori warrant to believe thatshe, or her source, said that p—at no point need her (derivative) a priori warrant tobelieve that someone at some point said that p be mentioned. I haven’t even tried tomake good on the claim that there are (at least some, actual) cases where she only,somehow, has the latter warrant—and partly on that basis knows that p.42

44.1 Enablers and explainers

Let’s return to the argument rehearsed in }2.2 (my second objection to Burge).I argued that a successful recipient’s empirically warranted belief that her source saidthat p plays a certain epistemic role vis-a-vis her testimonial belief that p—that of aground, or base—a role that, on the face of it, can’t be reconciled with A PrioriTestimony. Pending a reconciliatory story, we are owed a plausible description of acase where it doesn’t play that role—a case that, nevertheless, instantiates testimonialknowledge. It doesn’t look promising to model the case on stock examples of reason-independent causation. But it’s not clear what the alternative is, once we grant that therecipient’s belief about what’s said both provides a reason for her to believe that p, andpartly causally explains that belief.

As stated, the objection assumes that any successful recipient is warranted in believingthat her source said that p. Can it proceed on the weaker assumption that she has warrantto believe or for believing this (i.e. ‘propositional’ warrant)? On the weaker assumption,it’s left open whether her belief about what’s said explains her testimonial belief (since it’sleft open whether she has the former).Would it, still, be dubious to suppose that there are


cases where the proposition that her source said that p isn’t among the reasons for whichshe believes that p?Arguably not—at least not on the grounds that it forces a peculiar picture of what

goes on in her mind. For all that’s been said, these are simply cases where the causal-explanatory constraint on basing is violated in a straightforward way: there’s noneed to invoke anything like deviance. If there’s no such constraint on basing, moremust be said of course. But there seems to be nothing inherently problematic aboutcases of roughly this shape: an agent knows or has warranted belief that p; she also has areason to believe that q; q is (or would be) a reason for her to believe that p; q isn’tamong the reasons for which she believes that p. Perhaps the critical type of testimonycase is just another case like that?Fair enough. The argument is conditional on the stronger assumption (which,

importantly, I have yet to see a compelling counter-example or objection to43).However, there’s a residual issue that cries out for explanation—one that can’t beavoided by retreating to the weaker assumption—namely: why does testimonialknowledge/warranted belief that p require warrant to believe that one’s source saidthat p, if not because it rests, in part, on that warrant? (And, if it does, how could it everbe a priori?) If there’s another–better–explanation available, it would be good tosee what it looks like. Now, if both assumptions—strong and weak—are abandoned,that pressure goes away as well. But another emerges: to explain the three-prongdefeater-pattern that seems to be characteristic of testimonial warrant and knowledge;in particular, the vulnerability to defeat by evidence of insincerity and communication-failure. The assumption—either strong or weak—points to a straightforward explan-ation of that vulnerability.44

Someone who grants the stronger assumption might object at a different point: tothe claim that the recipient’s belief about what’s said plays a causal-explanatory—asopposed to causal-enabling—role vis-a-vis her testimonial belief. This distinction (too)is hard to explicate, but not to illustrate. For instance, that Bettie has a head is plausiblya causal enabling condition, relative to a specific ordinary context, and the explanandathat she likes chocolate cake: it’s part of the general background conditions that makethis possible, but it doesn’t explain, even in part, why or how it happens. (Perhaps itcould explain it, in other—rather peculiar—circumstances; that doesn’t affect thepoint.)This distinction is different from—and, it seems, crosscuts—that between warrant-

enablers (states/conditions in the warrant-enabling role) and warrant-conferrers (states/conditions in the warrant-conferring role).45 Enablers, in the second sense, can becausal explainers, and so far I’ve been assuming that that’s precisely the status that theA Priori Testimony advocate would assign to the successful recipient’s belief that hersource said that p, relative to her belief that p.Note that it’s indeed what we should say, if we wish to sustain the analogy with the

case of the logic proof. (Unlike, say, the rotation of the earth, and the absence of toomuch arsenic in your system, your perception of the representation of the proof partly


causally explains your knowledge of the theorem.) Furthermore and importantly, thealternative has no intuitive support.

A plausible intuitive mark of genuine causal enablers is that they’re legitimately takenfor granted in a satisfying, causal or other, explanation of the given effects (cf. Bermùdez1995). We can give a complete (correct) account of why Bettie likes chocolate cake,without at any point citing her having a head.46 Another intuitive mark is that causalenablers tend to stay constant across a large amount of (actual and counterfactual)variation in the effects they enable—presumably that’s why they’re legitimatelyignored, in the explanation of these effects. (And why they have little, if any, predictiveforce on their own.) Other things equal Bettie would still have a head, whether or notshe liked chocolate cake—indeed, whether she had any food preferences at all.

But the successful recipient’s belief that her source said that p isn’t legitimatelyignored in the explanation of her believing that p by testimony (under that—therelevant—description). How would the explanation go? It might of course be over-determined that a given recipient believes that p, or even that she believes that p bytestimony, but that’s beside the point: for each testimonial route, the fact that shebelieves that her source said that p is a crucial part of the explanation. Relatedly, there’slittle reason to think that her belief about what’s said exhibits the constancy of a causalenabler (vis-a-vis significant variation in the purported ‘enablee’): that, other thingsequal, she’d still believe that her source said that p, whether or not she believed that p—rather than not-p, or q, or nothing at all—by that source’s testimony. Again, obviousenablers (of her believing that p by testimony) include her having a head, and theabsence of too much arsenic in her system: these conditions do pass the test.47

4.2 Fast track warrant preservation

Here’s a final suggestion (on my opponent’s behalf).

Let’s concede that any successful recipient of testimony that p is warranted—empirically war-ranted—in believing that her source said that p, and on that basis empirically warranted inbelieving that p. At least some such recipients are also warranted in believing that p on independ-ent, a priori grounds. What grounds? Those of her source. The source’s warrant sometimestakes the ‘fast track’, alongside the recipient’s empirical grounds for believing the same propos-ition—it’s robustly preserved in the testimony, status and strength intact. If it’s a priori, ofsufficient strength, and not itself testimonial, the recipient can even gain a priori knowledgethat p in this way.

In the abstract, this model—the ‘Fast Track Model’—is at least not wildly implaus-ible. It can be regarded as a (substantial but) friendly amendment to Burge’s account, oras part of a competitor; this does not matter. (It’s clearly compatible with manydifferent views of the ‘slow’ track.) It’s a model on which all-things-considered testimo-nial warrant is never a priori, but that’s not a problem per se: doxastic warrant, evenknowledge, could be over-determined in the critical cases. The model gives theadvocate of A Priori Testimony recourse to the natural picture of the recipient’s


psychology outlined in }2.2—with the added proviso that the natural picture isn’talways the whole picture. Note also that the model doesn’t give us ‘more than webargained for’ (cf. }1.2): testimony only issues a priori warrant when the source (or,perhaps, the source’s source . . . ) has a priori, non-testimonial warrant. And in theabsence of a source, or a source with warrant, there’ll be no a priori warrant by testimony,since there’s no warrant (of any kind) to be preserved.48

However, a proponent of the Fast Track Model faces two challenging, preliminarytasks: to clarify the intended warrant-preservation thesis, and to articulate the mechan-ism that implements it—the separate mechanism that explains how warrant is indeedpreserved, in the intended sense, if/when it is. (The preceding arguments should makeclear why a separate mechanism is called for. But, surprisingly, that need seems to havebeen overlooked so far.49) I’ll focus on the former task, since it seems prudent toaddress that first (and precisify the job description for the mechanism). Moreover, wedon’t in fact have a serious proposal on the table until the warrant-preservation thesis atissue has been clarified.There’s a range of warrant-preservation (or ‘transmission’) theses in the general

vicinity that don’t meet the Fast Track Model’s needs—including variations on eachof the following:

A recipient has warrant for believing that p(a) if and only if the source has warrant for believing that p(b) only if the source has warrant for believing that p(c) if the source has warrant for believing that p(d) if and because the source has warrant for believing that p(e) in some cases where the source has warrant for believing that p.

Options (a) through (d) are too weak and too strong—at any rate: considerably strongerthan the model needs, and arguably too strong to be plausible—whereas (e) is just tooweak. Each option is too weak, in that it leaves it entirely open whether the recipient’swarrant shares any significant features with the source’s warrant—notably: that of beinga priori (or a posteriori, as applicable).What the Fast Track Model needs is a thesis of roughly the following form:

A successful recipient has warrant of type w for believing that p(f) if (alt. if and because) the source has warrant of type w for believing that p(g) in some cases where the source has warrant of type w for believing that p.

The restriction to successful recipients allows us to say that the other—the unavoidablyempirical—strand of any all-things-considered testimonial warrant doesn’t stand or fallwith the source’s warrant. Furthermore, it’s built into each of (f) and (g) that, whenwarrant is preserved, the recipient has the same warrant (type) as the source. On theintended interpretation, warrants are rather finely individuated, but not too finely: notso finely that two distinct subjects are unable to share warrant type, and not so broadlythat two warrants of the same type can differ in their status as a priori or as a posteriori.


(Nor in strength—at least not too much—lest the explanation of a priori testimonialknowledge fall through.) Now: what’s that way? Can anything more informative be saidabout it?

Perhaps, on the intended reading, warrants are just carved in whatever way they’recarved for it to be possible to pass on one’s warrant to another in the usual, seeminglyun-mysterious, way: by facilitating her access to it. I can share my (a priori) warrant forbelieving a certain logical theorem with you, by guiding you through the proof—or,on suitable assumptions about your motivation and ability, simply by informing youthat there is a proof, and leaving you to it. Likewise, I can share my (a posteriori)warrant for believing that Bettie likes chocolate cake, by telling you that she eats itevery day and claims to like it—or perhaps by pointing over at Bettie (fully occupiedeating cake).

Whatever the precise structure of my warrant, and however exactly that structurediffers from that of relevantly different warrant types, there’s an intuitive sense in whichI’m able to pass you my warrant—and share it with you—in examples like these. Andwhen a warrant is shared in this familiar way, then, other things equal, it doesn’t lose itsa priori/a posteriori status ‘in transfer’. Provided you pick up on the right warrant—thewarrant I attempt to share—and you lack defeaters (and status-affecting defeater-defeaters), the warrant you have as a result of our interaction is a priori if mine is,and a posteriori if mine is. The same goes, it seems, for relative strength: other thingsequal, our warrants will match in that respect too. (The status and strength of our all-things-considered warrants may, of course, differ—but that, too, is as it should be.)

Let’s call this phenomenon ‘facilitating warrant-sharing.’ It might be tempting to useit as the blueprint—the blueprint for further explication of the warrant-preservationthesis at the heart of the Fast Track Model. Since we have an intuitive, independentgrasp of facilitating warrant-sharing—and so, presumably, of the individuation condi-tions for warrant types that’s implicit in it—it might be tempting to invoke it here.Note also how plausible it is that, in facilitating warrant-sharing, one’s experiences ofthe other agent’s utterance, of specific features of the conversational context, etc., justplay a triggering role (vis-a-vis the warrant one acquires). Perhaps, then, the possibilityof a priori testimonial warrant can, after all, be explained with appeal to the proof caseanalogy?

However, there’s a problem right upfront: in facilitating warrant-sharing, too manyfeatures of the original warrant seem to survive the transfer from one agent to thenext—too many for it to be a suitable blueprint. In particular: actual and potentialdefeaters survive—including underminers (at all levels of grain). Other things equal,the warrant you have, after I guide you through the proof, or alert you to cake-eatingBettie, is vulnerable to the same undermining defeaters as my warrant is: if evidencethat Bettie suffers from bulimia nervosa potentially undermines my warrant for believingthat she likes chocolate cake, it potentially undermines yours; if evidence thatthe ostensible proof we worked through contains a serious but elusive fallacy poten-


tially undermines my warrant for believing the theorem, it potentially underminesyours—and so on.But these defeater-patterns don’t survive the transfer of warrant that supposedly takes

place in testimony. To see this, suppose you and I are warranted in believing the sameproposition—say, that many children have imaginary friends—by the testimony of twodifferent sources, Max and Bettie, respectively. They’re warranted, in turn, but bydifferent methods: Max’s warrant partly rests on his warrant to believe that this wasconfirmed by a recent global study; Bettie’s warrant partly rests on her warrant tobelieve that Where the Wild Things Are is hugely popular.Presumably, the case can be elaborated in such a way that the epistemic situations of

the recipients (i.e. you and me) don’t differ in any (other) relevant respects—say, withrespect to what we believe, or have reason to believe, about our sources’ warrants.

On the proposed gloss, the warrant-preservation thesis seems to imply that ourrespective, testimonial warrants would, nevertheless, have starkly different defeater-structures: not just in that yours (and not mine) is potentially undermined by evidencethatMax’s utterance was insincere, thatMax is incompetent, etc.—that’s how it shouldbe; but in that yours (and not mine) is potentially undermined by evidence that thestudyMax relies on has methodological flaws, or has been disconfirmed by other studiesin the interim. Conversely, of course, for my warrant and Bettie’s: other things equal,my warrant inherits the defeater-pattern associated with hers. But that’s very implaus-ible.50 (And the choice between (f ) and (g) makes no relevant difference. Against (g):just add whatever’s supposed to be missing, for the case to be one where warrant isindeed preserved; then, again, consider the upshot for the distribution of defeaters.)If this is right, then warrants, as they figure in the warrant-preservation thesis that the

Fast Track Model needs, aren’t individuated in whatever way warrants are individuatedin facilitating warrant-sharing. (Or they are, and the view is a non-starter.) But how,then, are they individuated? Without an informative answer to that question, thewarrant-preservation thesis—and with it the Fast Track Model—is simply too pro-grammatic to deserve our confidence.51

Notes1. Unless indicated, all the references in this section are to works by Burge.2. See my 2006, }9, for some discussion of the other analogy—with so-called ‘preservative

memory’. (See also Christensen and Kornblith 1997; Lawlor 2002.)3. If Kitcher 2000 or Williamson 2007 is right, we may not. But I won’t engage their arguments

here. (See my 2011, }9 for some discussion of Williamson’s.)4. Nothing here turns on how exactly this distinction is precisified. For discussion, see Silins

2006; Casullo 2007a.5. Burge uses ‘a priori’ and ‘non-empirical’ interchangeably; likewise for ‘a posteriori’ and

‘empirical’. I follow that usage. And I treat ‘warrant’, ‘evidence’, and (good, epistemic)‘reason’ as equivalent. When nothing hinges on the difference, I move freely between ‘S


has warrant to believe/for believing that p’ (propositional warrant) and ‘S is warranted inbelieving/has warranted belief that p’ (doxastic warrant). Most of the time I write as if belief,as well as warrant, is an all-or-nothing affair—this, too, is a simplification that nothing turns on.

6. In section 3 we return to the question what, besides assertion, might fall in the umbrellacategory.

7. A little more precisely: all it takes is that she’s a linguistically competent, rational agent, wholacks defeaters. According to Burge, the source of the u-entitlement is the reliability of ourlinguistic competence, whereas the r-entitlement ‘comes with being a rational agent’; 1993,467; 1997, 31; 1999, 233. (For a fuller picture, cf. 2003.)

8. What Burge means by ‘standing linguistic competence’ is at least roughly equivalent to whatsome have called ‘semantic (or narrow semantic) competence’, ‘knowledge of the meaningof expression-types’, or ‘knowledge of context-invariant meaning.’ (In 1999, ‘comprehen-sion’ is used in lieu of ‘intellectual understanding’.)

9. Perhaps it’s enough that there’s a priori non-testimonial knowledge somewhere in the chain.(Either way, it’s unclear whether this condition is well motivated, given that Burge’s officialreason for requiring knowledge by the source is to preclude Gettier-cases; 1993, 486, fn. 24.For related discussion, see Casullo 2007b.)

10. At least for agents relevantly like us—without telepathic capacities, ‘thought-injection’ skills(cf. 1999, 240), or the like. (Cf. the curious example in Burge 1999, 240.) But it’s not clear tome that our notion of testimonial warrant/knowledge is even applicable to agents with suchradically different means of communication.

11. This qualification will henceforth be treated as understood.12. What is controversial is that at least part of her overall warrant to believe that content is

default warrant—and, of course, that it’s sometimes a priori.13. Another way is by testimony, but I trust that it’s clear why appeal to that route would be a

non-starter here.14. See Malmgren, 2006, 217, fn. 40. Note also that Burge’s argument for the claim that

perception sometimes plays a triggering role, vis-a-vis the u-entitlement, doesn’t beardirectly on this concern: even if that argument were successful—as I argue that it’s not(2006, 201–2)—the most it would show is that the u-entitlement is sometimes a priori. Thequestion how it could be remains.

15. But see my 2006, 217–18 (esp. fn. 41).16. The remaining options that I offer Burge—that the u-entitlement (just like the r-entitle-

ment) is a warrant to believe that p, or that it’s a warrant to make a transition—are criticizedin my 2006, }5, and there’s no need to go over them here.

17. It’s plausible that the requirement is conditional on the possession of certain concepts (cf. my2006, 236, n. 79). And that what in the first instance matters is the apparent saying and source.

18. I write as if what does the causal work is strictly speaking her beliefs, rather than her warrantedbeliefs or knowledge, but nothing crucial turns on this. Likewise for various general issues inthe metaphysics of causation—e.g. whether mental states of any kind (as opposed to, say,facts or events) are fit to be causal relata—and in the metaphysics of reasons/warrants. Theargument can be restated in your framework of choice.

19. I don’t think it matters whether we think of the relevant contextual factors as includingauxiliary reasons, or only warrant-enablers—conditions that enable the given proposition toconstitute a (‘complete’) reason.


20. E.g. Williamson 2000. (But, by my lights, knowledge of what’s said is at most a plausiblerequirement on testimonial knowledge.)

21. For a helpful overview, see Korcz 2010.22. See e.g. Davidson 1963; Chisholm 1966; Lennon 1990; Plantinga 1993.23. This gloss—in particular the notion of normality involved—obviously needs precisification.

But it’s not meant to rule out the possibility that, for some agents and effects, interferencemechanisms are statistically dominant.

24. I leave counterfactual analyses of the basing-relation out of consideration (see e.g. Swain1979) since, on the face of it, they have nothing at all to say about the present case: allrelevant counterfactuals hold.

25. At least: not warranted by the route that’s explicit in the description of each case. (Cf. }4.)26. The difference in formulation makes no difference; in particular, ‘rational’ here contrasts

with ‘arational’—not ‘irrational’. (But see Fricker 2006 for a case that Burge in factequivocates on the notion of rationality at play.)

27. Nothing controversial is intended by ‘introspective’—all we need is that there’s at least onetype of warrant, that’s available to the first person (only), that’s a priori, and that at least hasthe scope indicated in the text. (And we can ignore the possibility that whether it’s availabledepends on the value of p.)

28. I.e., that a ‘liberal’ treatment of the warrant-structure is correct. Strictly speaking, theargument only requires that no additional, empirical warrant is called for (and the brand of‘conservatism’ on which that’s true is radical indeed—given the simplicity of the argumentunder consideration, it’s unclear if it can recognize any inferential a priori warrant at all).

29. In correspondence Neta suggests a route from it introspectively ‘seeming’ that someone atsome point said that p (to having a priori warrant to believe that someone did). It’s not clearhow exactly the purported seeming-state is supposed to differ from a belief (or memory)state with the same content, but in any case this suggestion falls in the same category as thoselisted in the text: in effect, it just transforms my original request—for an explanation of howit’s possible to have a priori warrant to believe the target claim—to a request for anexplanation for how the proposed transition could be a priori (in the terminology of my2006, 220–4).

30. It’s also worth noting that the kind of a priori warrant that the agent has, on my reading ofthe case—a partly introspective, inferential, warrant—is most likely not what Burge himselfhad in mind. (For one thing, it makes the restriction to contents that can be ‘intellectuallyunderstood’ completely mysterious. For another, it jars with his repeated insistence that thewarrant at issue is non-inferential—perhaps even with it being an entitlement! Cf. Burge1997, 30.) But I’m interested in the proposal as a suggestion on Burge’s behalf: one thatmight be used to amend or modify his account, with a view to vindicating A PrioriTestimony.

31. I gloss over differences in the requirements for (testimonial) warrant and knowledge.Nothing turns on this.

32. Pace some versions of the ‘reflection principle’ (van Fraasen 1984). The intuitive consider-ations that I offer here don't yet amount to—and aren’t meant to be—an argument againstthat principle. (But see e.g. Christensen 1991; Arntzenius 2003).

33. Certain permutations may survive, of course. (This won’t matter here.) Examples of eachtype of gloss can be found in Matilal and Chakrabarti 1994; Lackey and Sosa 2006.


34. E.g. Graham 1997; Lackey 2006; Cullison 2010. (Contrast the approach of Fricker 2006a.)35. The notion wouldn’t have to be pre-theoretical—and I’m not assuming that. Why think

that there is a determinate notion even within philosophical discourse that meets thesedesiderata?

36. Roughly: defeaters that provide evidence that the method whereby the agent formed, orwould form, her belief that p isn’t a good method—e.g. that it’s unreliable. ‘Rebutters,’ incontrast, provide independent reason to disbelieve p. (This is rough indeed, but shouldsuffice for our purposes. Cf. Pollock 1986; Casullo 2003).

37. The hypothesis isn’t, implausibly, that it’s impossible for non-testimonial warrant/knowledgeto exhibit the relevant defeater-pattern; but that doing so is necessary to count as testimonialwarrant/knowledge in particular—that it’s part of the nature or essence of that epistemickind.

38. E.g., for each defeater-type, the evidence may pertain to a single case, to any possible case,or anything in between. And evidence concerning the de facto source arguably only under-mines (testimonial) knowledge, whereas evidence concerning the apparent source underminesboth warrant and knowledge. (‘Trustworthiness’ is sometimes used as an umbrella term forsincerity and competence.)

39. I’m assuming that introspective warrant is defeasible in the first place. (For dissenting voices,see e.g. Alston 1972; Burge 1996.)

40. But cf. Byrne 2005. (The point is orthogonal to whether we—also, always—think in alanguage-of-thought.)

41. A parallel move might be available for the second type of underminer.42. Neta also addresses another argument of mine, strictly aimed at the suggestion that the

u-entitlement (just like the r-entitlement) is a warrant to believe that p (my 2006, 210–11).That suggestion implies that a recipient can have a priori warrant for believing contents thather source doesn’t—indeed can’t—have a priori warrant to believe. That struck me asabsurd. I first illustrated the perceived absurdity with a case featuring the content that it’sraining. Neta complains that that content isn’t capable of being intellectually understood (cf.}1.2). But I explicitly considered that possibility, later in the text, and went on to give otherexamples (my 2006, 224). Neta’s main point, however, is that he simply doesn’t find theimplication in question absurd. And that’s fair enough: as I also acknowledged (ibid. 225),it’s in any case a consequence of Burge’s overall view—it’s what explains the expansion ofthe domain of a priori warrant—and one might just regard it as another (perhaps surprisingor interesting) upshot of that view. In this chapter, I’ve tried to articulate at least my ownunease with this consequence a little better (}1.2). And fortunately not much hinges ontreating it as absurd; there are other reasons to reject the suggestion this argument was aimed at(my 2006, 218–19).

43. There are many variables, and it’s hard to construct a clear test case (cf. my 2006, 229–30,esp. n. 66). But extant attempts at counter-examples tend to assume a very stringentconception of belief, or of the psychological processes involved (e.g. Burge 1993; Recanati2002; Audi 2006). Goldberg 2010 assumes an implausible account of the warrant structure(68). Gilbert 1991 argues, on partly empirical grounds, that it’s impossible to “understand aproposition without representing it as true” (114). Taken at face value, that’s obviouslyfalse, since there are world-to-mind directed propositional attitudes (e.g. desires). Takenmore charitably—as the claim that belief necessarily precedes disbelief (temporally and/or


explanatorily)—it’s compatible with everything I’m saying. (At most, however, Gilbert’sarguments support the even weaker claim that people are rather gullible.)

44. The significance of incompetence is also easily explained, on the overall model I favor(my 2006, }8).

45. The standard terminology here is confusing—the same phrase, ‘causal enabler’, is sometimesused to designate enabler-as-opposed-to-justifier, and other times to designate enabler-as-opposed-to-explainer. (And neither distinction maps neatly on to the efficacy/relevancedistinction.)

46. Perhaps there are several correct explanations, at different levels of description; nothing hereturns on this.

47. Interestingly: her beliefs that there are other minds, that she’s not a brain-in-a-vat, that she’snot now dreaming, etc., seem to pass the test too—given some obvious restrictions on thevalue of p.

48. The model also avoids the undesirable consequence that, unless there’s a source, and thatsource has warrant, there’s no testimonial warrant at all—a consequence that seemingly can’tbe avoided if all testimonial warrant is explained in terms of warrant-preservation. (For abold endorsement, see Owens 2006, 16.)

49. Not just by Burge, but e.g. by Faulkner 2000, Moran 2006, Owens 2006, Schmitt 2006—each of whom advocates a warrant-preservation thesis of roughly the kind the modelrequires. Lawlor 2002 notes the need, as well as the corresponding need in an account ofwarrant-preservation by memory; she argues that it can be met in the latter case, but not in theformer. (However, the mechanism Lawlor describes at most explains the preservation ofcontent, by memory—but that’s not sufficient for warrant-preservation, and arguably notnecessary either.) Goldberg 2010 specifies a mechanism of sorts. But see my 2012 forobjections.

50. If there’s a strength-asymmetry between our sources’s warrants, the thesis also implies thatwe’re warranted to different degrees. That, too, seems to me unacceptable. But it’sa consequence that the Fast Track Model is supposed to have. If anything, then, this isthe basis for a different objection. (And intuitions about warrant-strength seem to varyconsiderably.) For examples similar to mine, see McDowell 1994; Gerken 2011.

51. Many thanks to audiences at Amherst College, Harvard, Northern Institute of Philosophy atAberdeen, the Carolina Metaphysics Workshop, and to Mikkel Gerken, Mike Raven, andRam Neta.

ReferencesArtzenius, Frank, 2003, “Some Problems for Conditionalization and Reflection,” Journal of

Philosophy, 100 (7), 356–70.Audi, Robert, 2006, “Testimony as an A Priori Basis of Acceptance: Problems and Prospects,”

Philosophica, 78, 85–104.Bermùdez, Jose Luis, 1995, “Nonconceptual Content: From Perceptual Experience to Sub-personal Computational States,” Mind and Language 10 (4), 333–69.

Burge, Tyler, 1993, “Content Preservation,” The Philosophical Review, 102, 457–88.Burge, Tyler, 1997, “Interlocution, Perception, and Memory,” Philosophical Studies, 86, 21–47.


Burge, Tyler, 1999, “Comprehension and Interpretation,” The Library of Living Philosophers, 27,The Philosophy of Donald Davidson, ed. Lewis E. Hahn, 229–54. Chicago: Open Court.

Burge, Tyler, 2003, “Perceptual Entitlement,” Philosophy and Phenomenological Research, 67,503–48.

Byrne, Alex, 2005, “Introspection,” Philosophical Topics, 33 (1), 79–104.Byrne, Alex, “Knowing That I Am Thinking,” forthcoming, Self-Knowledge, ed. A. Hatzimoysis.Oxford: Oxford University Press.

Casullo, Albert, 2003, A Priori Justification. New York: Oxford University Press.Casullo, Albert, 2007a, “What is Entitlement?” Acta Analytica 22 (4), 267–79.Casullo, Albert, 2007b, “Testimony and A Priori Knowledge,” Episteme 4 (2007): 322–34.Chisholm, Roderick, 1966, Theory of Knowledge. Englewood Cliffs, N.J.: Prentice-Hall.Christensen, D., 1991, “Clever Bookies and Coherent Beliefs,” The Philosophical Review, 100,229–47.

Christensen, David, and Hilary Kornblith, 1997, “Testimony, Memory, and the Limits of the APriori,” Philosophical Studies, 86: 1–20.

Coady, C. A. J., 1992, Testimony; A Philosophical Study. Oxford: Clarendon Press.Cullison, Andrew, 2010, “On the Nature of Testimony,” Episteme, 7 (2), 114–27.Davidson, Donald, 1963, “Actions, Reasons, and Causes,” Journal of Philosophy, 60.Evans, Gareth, 1985 [1979], “Reference and Contingency.” In Collected Papers. Oxford: Clar-

endon.Faulkner, Paul, 2000, “The Social Character of Testimonial Knowledge,” Journal of Philosophy,

97, 581–601.Fricker, Elizabeth, 2006, “Second-Hand Knowledge,” Philosophy and Phenomenological Research,73 (3), 592–618.

Gerken, Mikkel, 2011, “Internalism and Externalism in the Epistemology of Testimony,”Philosophy and Phenomenological Research (DOI 10.1111/j.1933-1592.2011.00532.x).

Goldberg, Sanford, 2010, Relying on Others. Oxford: Oxford University Press.Graham, Peter J., 1997, “What is Testimony?”, The Philosophical Quarterly, 47, 227–32.Green, Christopher R., 2008, “Epistemology of Testimony,” Internet Encyclopedia of Philosophy,

ed. Dowden and Fieser.Kitcher, Philip, 2000, “A Priori Knowledge Revisited,” New Essays on the A Priori, ed. Boghos-sian and Peacocke, 65–91. New York: Oxford University Press.

Korcz, Keith Allen, 2010, “The Epistemic Basing Relation,” The Stanford Encyclopedia ofPhilosophy, ed. Zalta.

Lackey, Jennifer, 2006, “The Nature of Testimony,” Pacific Philosophical Quarterly, 87, 177–97.Jennifer Lackey and Ernest Sosa (eds), 2006, The Epistemology of Testimony. Oxford: Oxford

University Press.Lawlor, Krista, 2002, “Memory, Anaphora and Content Preservation,” Philosophical Studies, 109:

97–119.Lennon, Katherine, 1990, Explaining Human Action. London: Duckworth.Malmgren, Anna-Sara, 2006, “Is There A Priori Knowledge by Testimony?”, The PhilosophicalReview, 115: 2, 119–241.

Malmgren, Anna-Sara, 2011, “Rationalism and the Content of Intuitive Judgments,”Mind 120:263–327.


Matilal, Bimal Krishna and Chakrabarti, Arindam, 1994, Knowing from Words: Western and IndianPhilosophical Analysis of Understanding and Testimony. Dordrecht and Boston: Kluwer Academic.

McDowell, John, 1994, “Knowledge By Hearsay,” in Matilal and Chakrabarti, 1994.Moran, Richard, 2006, “Getting Told and Being Believed,” in Lackey and Sosa (eds), TheEpistemology of Testimony, Philosophers’ Imprint 5: 4.

Neta, Ram, 2010, “Can A Priori Entitlement be Preserved by Testimony?” in Haddock, Millar,and Pritchard (eds) Social Epistemology. Oxford: Oxford University Press, 194–215.

Owens, David, 2006, “Testimony and Assertion,” Philosophical Studies, 130: 105–29.Pollock, John, 1986, Contemporary Theories of Knowledge. Savage, MD: Rowman and Littlefield.Recanati, F., 2002, “Does Linguistic Communication Rest on Inference?” Mind and Language,

17 (1–2): 105–12.Schmitt, Frederick F., 2006, “Testimonial Justification and Transindividual Reasons,” in Lackey

and Sosa, 2006.Silins, Nicholas, 2012, “Explaining Perceptual Entitlement,” Erkenntnis, 76 (2): 243–61 (DOI

10.1007/s10670-011-9304-0).Swain, M., 1979, “Justification and the Basis of Belief,” in Pappas (ed.), Justification and Knowledge,

Dordrecht: D. Reidel.Van Fraassen, B. C., 1984, “Belief and the Will,” The Journal of Philosophy, 81, 235–56.Williamson, Timothy, 2000, Knowledge and Its Limits. Oxford: Oxford University Press.Williamson, Timothy, 2007, The Philosophy of Philosophy. Oxford: Blackwell.


8

Intuitions and FoundationsThe Relevance of Moore and Wittgenstein

Ernest Sosa

Philosophers want to know what in general makes our attitudes justified, if and whenthey are. Traditional epistemology in particular wants to know what it takes for ourbeliefs to be justified. Skeptics have of course argued that our beliefs generally are notjustified. But even such skeptical arguments are based on assumptions, howeverimplicit, as to what it takes for a belief to be justified.

Some of our contingent beliefs are justified based on further beliefs. A further beliefprovides justification, however, only if it is itself justified. This sets up the familiarregress/circle/foundations problematic.

One’s body of beliefs is rationally structured. Here two metaphors present as manyoptions. One is the traditional foundationalist pyramid. The other, the coherentist raft,reflects the fact that rational basing is not asymmetric: a belief can provide a rationalbasis for another even while the latter concurrently returns the favor. For the coher-entist, our belief system is a structure floating free, each belief deriving all its epistemicjustification from its place in the structure. It is of course good for our body of beliefs tobe thus integrated, but it is not so good for it to be causally and explanatorily detachedfrom the surrounding world. Such beliefs would be lacking, epistemically lacking. Nor isit enough for beliefs to be true accidentally, as might be the vast bulk of a demon victim’sfree-floating beliefs, regardless of how intricately they interlock. Mutual basing cannotplausibly exhaust all there is to epistemic justification.

We can of course define a kind of internal, coherential justification that the demon’svictim does enjoy. This seems harmless, but epistemic evaluation involves more thansuch purely internal status, even without considering Gettier issues. Intellectual com-petence goes beyond mere free-floating coherence. Beyond such coherence, we areinterested in a competence that more fully helps constitute knowledge.1 What rendersa belief thus rationally competent? It might be the belief ’s being based rationally onother beliefs already competently formed in their own right. But that can’t go onforever, nor can each of a set of beliefs be competent by being based rationally on theothers, and nothing more. Some beliefs must be competent through something otherthan rational grounding on other beliefs. What might this other source be?2

Perception and introspection might give us what we seek, since neither need involverational grounding on other beliefs, but can work instead off the presentational given.Perceptual and introspective beliefs can be based rationally on states that lie beyondjustification and unjustification, such as a sensory experience. An experience canground either or both of the following: (i) a perceptual belief about the surroundings;(ii) an introspective belief about the presence of that very experience. Such perceptualand introspective beliefs are thus foundationally rational, by involving conceptual“moves” evaluable as rationally justified or not, moves based on other mental states.These other states are regress-stoppers, providers of foundational justification, by havingonly one foot in the space of reasons. They provide justification without in turnrequiring it, or even sensibly allowing it.What relation must an experiential state E bear to a corresponding belief B in order

for E to (help) justify B? Suppose B is the belief B(p), whose propositional content is theproposition .3 What relation must B(p) bear to E in order to be justified by beingrationally based on E? This question has a twofold answer.Take first a state E with a corresponding content . B(p) is said to derive rational

justification from being based rationally on E(p), as when B(p) is the belief that one seesthat such and such, and E(p) is a visual experience as if one sees that such and such.For another thing, E might itself amount to the fact that p, as an ache suffered by

S might amount to the fact that S aches (in a certain way).Our twofold answer faces the Speckled Hen problem, however, wherein the subject

inappropriately believes that his conscious experience is of a certain sort, or that hissurroundings have a certain perceptible character, while his visual experience is of thatsort, and does have the content attributed to the surroundings. This all can happen wellbeyond the subject’s ability to subitize, unfortunately, which is usually limited tocomplexity of degree four or so, at most.4 Thus, if one has an experience as of seeingeight dots, one might believe one sees eight dots on the seen surface, or in one’ssubjective visual field, and one might be right just by accident. Neither the perceptualnor the introspective belief may then be so much as competently formed, because thenumber of dots exceeds one’s subitizing limits.Even for proper understanding of perceptual and introspective foundations, then,

one needs to invoke the subject’s competence. Here again the competence is reason-based. When the number of dots is within his subitizing limits, the subject again basesperceptual and/or introspective beliefs on corresponding experience, but now thebeliefs are rationally justified. The difference is that beliefs so formed are reliably true.The subject is able competently to discern whether there are exactly three visible dotson the seen surface right before him, or in his visual field.

Must all foundational competence be thus reason-based? That is doubtful given thepossibility of reliable blindsight. This is not the blindsight discovered by vision scientistsin our actual world. Without much of an imaginative leap, however, we reach worldswhere blindsight ability becomes prowess. What are we to say about such blindsightbeliefs? Are they all rationally incompetent? They involve conceptual moves, contentful

INTUIT IONS AND FOUNDATIONS 187

seemings, and beliefs. Are we to dismiss these moves as epistemically deplorable withoutexception, even despite the fact that they are nearly infallible?

That would seem especially ill-advised about our most basic arithmetical, geomet-rical, and logical beliefs and seemings. No basis is here in evidence for our belief, noreven for our seeming, our inclination or attraction to assent, nothing sufficiently likethe visual sensory experience that mediates between a visible fact and our perception ofthat fact.5

Thus are we led to a conception of intuitions as seemings, as inclinations orattractions to assent that are not reason-based, seemings not based rationally onanything beyond sheer understanding of the proposition involved. Such seemings areconceptual, since only given understanding can our assent be attracted. This distin-guishes seemings from experiences. One can experience contents of experience thatone could not even fully represent, for lack of the required concepts. In contrast,seemings are conceptual deployments, and rationally evaluable as such.6

What distinguishes foundational seemings that are justified? Shall we adopt a latitu-dinarian position according to which all intuitive seemings are ipso facto justified? Thiswould include all those with no basis beyond the subject’s understanding of theproposition involved, except possibly for a presentationally given state beyond justifi-cation and unjustification. From this it would follow that biases or superstitions arerationally justified, however, so long as they are imbibed from one’s culture with norational basis, which is how such beliefs are too often acquired. They are absorbedthrough mechanisms of enculturation that need not include explicit verbal instruction.They enter rather through the osmosis of body language, tone of voice, the perceivedbehavior first of elders and later of peer trend-setters, etc.

That poses the problem of justified intuition: What distinguishes intuitions that areepistemically justified from those that are not? The answer may not be far to seek oncewe recall that even the presentational given can provide justification only through thesubject’s relevant competence. Perhaps we need only invoke a competence that is notreason-based, one that sub-personally enables reliable discernment of the true from thefalse in the relevant class of beliefs.

What distinguishes rationally justified intuitions from intuitive though irrationalbiases and superstitions? In order to be justified, intuitions must manifest an epistemiccompetence, one seated in the individual or in his culture or subgroup.

We are helped to appreciate the full scope of such intuitions by Moore’s defense ofcommon sense, and Wittgenstein’s response. Let us look more closely at the Moore/Wittgenstein exchange.7

Prominent among the beliefs defended by Moore in his “Defense of CommonSense” are those voiced as follows, all of which he takes to constitute certain knowledge:

I now have an existing, living body that has existed continuously since its birth some time ago.During that time it has been in contact with or some distance from many other bodies with shapeand size. These include living human bodies that had been born some time earlier and had existed

188 ERNEST SOSA

continuously since then. I have perceived things and observed facts about them, and have also beenaware of unobserved facts, such as that my body existed yesterday. And the like is true of manyother people. Finally, many other people also know corresponding things about themselves.

That is lifted or paraphrased from the beginning paragraphs of his celebrated paper.8 Forconvenience, let’s call such beliefs “core components of common sense.” In “Proof ofan External World,”Moore then argues that the physical bodies accepted by commonsense are “external,” i.e., metaphysically independent from all humanminds. It is on thisbasis that he offers his proof that there is an anthropically “external” world. (Note,parenthetically, that this does not directly address the traditional skeptical concern, forwhich it does not matter whether the world external to the subject is (a) anthropicallyconstructed or (b) “external” even relative to humanity at large. Regardless of whethersuch a world is of sort (a) or is rather of sort (b), the traditional skeptic will worry equallywhether he can justify belief in its existence on a basis provided by what is given to hisown subjective experience. In his article, Moore is hence obviously not directly concernedwith traditional skepticism!)Contained in basic common sense are hence general propositions like the following:

M1. There are and have been physical, spatially related bodies.M2. Some of these are living bodies.M3. Among these are people.M4. People perceive things and observe facts.M5. People are also aware of unobserved facts.M6. People know these things.

Having postulated such certainly known truisms definitory of common sense, andbeyond any reasonable skeptical doubt, Moore leaves room aplenty for philosophicalaction. For one thing, the analytic philosopher can provide an analysis of these knownfacts, of how they are metaphysically constituted. That is a task for analytic metaphys-ics. For a second thing, we can try to understand just howwe know these things that wetake ourselves so obviously to know. This is a task for epistemology. And the two tasksmay intertwine.In the course of his long career, Moore tries to carry out the epistemological task, in

repeated attempts that culminate in his papers “Certainty” and “Four Forms of Skepti-cism.” These adopt a form of indirect realism, according to which we know aboutexternal, mind-independent objects through inductive or analogical inference fromwhat we know immediately. As far as contingent reality goes, what we know immedi-ately seems limited to knowledge restrictedly about ourselves, including knowledge ofour own sense experiences. Knowledge of perceived physical objects and of other mindsis inferential, based on reasons provided ultimately bywhat we know immediately. Eventhe knowledge that one is awake and not just dreaming is based on reasons constituted byour concurrent and recent experiences; these reasons can be conclusive even if wecannot list them fully, which would be required for a proof useable against the skeptic.


What about our “core components of Common Sense,” the propositions thataccording to Moore constitute basic theses of our ordinary view concerning theworld and our place in it?

Would the core certainties of common sense also acquire their status through reasons providedby what we know immediately?

They would indeed, or so we can plausibly infer from the concluding paragraphs of“Four Forms of Skepticism.”Much earlier, in “Hume’s Philosophy,” Moore already tackles our indented ques-

tion directly, and concludes with a related question:

Must we admit a class of propositions that assert “matters of fact” and are known to be true“ . . . neither by direct observation nor by memory, nor yet as a result of previous observations”?

In that 1909 paper he remains agnostic. He says only that there “may” be suchpropositions.

By the time of his “Defense of Common Sense” in 1925, Moore is ready to saymore. About his core components of Common Sense, he now asserts that

. . . in the case of most of them, I do not know them directly: that is to say, I know them onlybecause, in the past, I have known to be true other propositions which were evidence for them.

This appears to be his final view on the epistemic status of core components ofcommon sense. No deviation from it is discernible in his later work.

That sums up Moore’s Common Sense philosophy, as laid out in a paper consideredby Wittgenstein to be Moore’s best, one that stimulated his own On Certainty. In thatbook Wittgenstein repudiates Moore’s epistemology and offers a radically differentalternative. I will here present the gist of that alternative, and invite the reader tocompare my account with the passages cited in my supportive footnote.

Wittgenstein juggles many issues in On Certainty. He is concerned with languagegames, with pragmatics of language use, with dialectical interplay, with what it is properto say to someone, with the effects of context on all this.Whether we agree with his pre-Gricean take on such issues, much in his text is detachable from that.Much concerns justbelief, knowledge, and certainty, regardless of how this may or may not relate tolanguage use and dialectical context. Moore, for his part, in his relevant epistemologyis largely unconcerned with such dialectical, linguistic, and contextual issues (whatevermay be true of his thoughts on Moore’s paradox). In what follows I too will abstractalmost wholly from them, while remaining neutral on their substance and on theirrelation to our more purely epistemological issues. Here then is how I see the positionthat Wittgenstein lays out in opposition to Moore. It contains four theses.

W1. We do not learn the core components of common sense through ratiocin-ation or evidence. Rather we “acquire” them as part of a coherent system:“light dawns gradually on the whole.”

190 ERNEST SOSA

W2. We do not strictly know these components. Rather they “stand fast” for us,and it is upon such a riverbed that our retail beliefs flow; it is by turning onsuch hinges that doors open to our retail knowledge.

W3. Core beliefs of common sense generally require no conscious assent. Manyare manifest rather in how we act.

W4. We can grasp the core propositions that give content to such beliefs onlythrough believing enough of them (along with other related propositions withwhich they form one’s individual system of beliefs). We might doubt any oneof them, but we could not possibly doubt them all (and retain our grasp).9

Wittgenstein thus rejects Moore’s view that we know core components of commonsense on the rational basis of other things we know or have known. In fact we know nosuch components, nor of course do we learn them based on evidence grasped throughother beliefs already in place.It seems to me that Wittgenstein is misled by his focus on language games and

dialectical linguistic interplay into denying that we know core components of commonsense.10 Even if he were right about that, moreover, the following question would stillremain: If a set of core commitments “stand fast” for a community in a way similar to theway our set of commonsense core commitments stand fast for us, is that enough torender the beliefs of that community epistemically proper, and indeed to make themknowledge (at least modulo Gettier issues)? Does the community’s sheer adherencesuffice to make it true that their interlocking system of beliefs has core components that“stand fast” while the rest constitute knowledge? Is it not conceivable that there be acommunity with a crazy enough system to which they nonetheless adhere as firmly andpervasively as we adhere to ours? Are there not indeed actual communities with weirdbelief systems kept in place through groupthink or other mechanisms of social cohesion?There are signs that Wittgenstein may have rejected the search for a philosophical

explanation of how our common sense knowledge might come about. Among thethings that philosophy ponders are some I would put as follows:

What sort of thing is such knowledge, and how might it come about? Can we answer thisquestion so as to enable understanding of how we might have as much of it as we do? Might weeven explain with philosophical generality how we do have as much of it as we do?

Even if it is as obvious as it is to Moore that we do enjoy such knowledge, the questionremains as to just how in general it comes about. Moore’s answer to this philosophicalquestion is his indirect realism in terms of immediate knowledge, inferential know-ledge, and so on.11

Wittgenstein thinks he can avoid our philosophical question by denying that wereally know all the things that Moore thinks we know. Some things we know are saidto be based not on things we know but rather on things that “stand fast” for us. Thethought seems to be that if we are to grasp the relevant propositions, then enough of

INTUITIONS AND FOUNDATIONS 191

the things that stand fast for us must so stand. Given their understanding-endowingstatus, such propositions “stand fast” for us rather than being things we “know.”

In fact Wittgenstein seems thereby misled into a wildly implausible stance. Even ifwe choose to join him in his linguistic deviance, moreover, this question still remains:Does everything that “stands fast” for us do so justifiably, with proper epistemicstanding, so that, for example, it can serve as a proper basis for knowledge, for theknowledge that, according to Wittgenstein himself, does gain proper support fromsuch fast-standing beliefs?

Also implausibly, Wittgenstein appears to reject Moore’s realism, the view that thereis an “external”world of facts independent of human minds, including facts accepted atthe core of common sense.

Finding no plausibility in either of those stances, I explore rather a response toMoorethat agrees with him both that we do know the core components of common sense, andthat these, along with much of what any of us knows, pertain to an objective realityindependent of the mind of any one thinker, and even of human minds altogether.

Given this twofold stance, we along with Moore still face the question of how wemanage to know the core components of common sense that we seem so obviously toknow, indeed with certainty, just as Moore asserts.

Once we avoid Wittgenstein’s linguistic deviance and his possible antirealist con-structivism, the best answer extractable, with some strain, from On Certainty is relativ-istic, either to one’s culture or to one’s humanity. But such relativism is hard to takeseriously. Consider again our questions:

How do we know the core components of common sense? Why are such beliefs so much as rationally,epistemically justified?

The answer “Just because they have gained consensual agreement in our culture”seems clearly wrong, as does even “Just because all humans agree eventually on thiscore of common sense, as they mature beyond infancy.” Cultural consensus is insuffi-cient for knowledge, since, for one thing, it is compatible with falsity. And even truthsconsensually agreed upon within a whole culture need not constitute knowledge, if theway these truths attract agreement is no better than the way falsehoods get consensuallyinstalled. That would make the culture’s access to such truths a matter of epistemic luckincompatible with true knowledge.

What is more, even if none of humanity had ever epistemically surpassed ouranimistic stages, consensually agreed upon animistic core beliefs would not therebyhave attained the status of knowledge. So, even what is consensually agreed upon byhumanity at large need not thereby rise to the status of shared knowledge.

Leaving relativism behind, shall we invoke evolutionary or other genetic theory?Can we thus support the view that human consensual agreement upgrades to know-ledge through our genesis, which guarantees reliable human faculties? Not clearlyenough, for more than one reason.

192 ERNEST SOSA

For one thing, it is not clear to what extent and with what scope evolution can beseen to ensure epistemic success. Even if it does yield considerable success, it wouldseem guaranteed only within limits set by what our species needs for its fitness andsurvival as a species.For another thing, take Swampman, whose core belief system is installed by the

incredible accident of molecules just coming together in a swamp through a bolt oflightning. The source of this creature and his belief system is of course unreliable.A bolt of lightning acting upon such molecules might as easily have put together alifeless smoking soup as a functioning human being, or so we can suppose. Even so, thisall seems irrelevant to Swampman’s ability to form beliefs and acquire knowledge soonafter he comes into being.What is more, even a species that comes into existence accidentally might eventually

acquire innate or culturally determined beliefs that derive from competence nonethe-less. Despite its accidental origins, such a species might still develop epistemic compe-tences in due course, and these might constitute proper sources thereafter, giving rise toretail knowledge and even, eventually, to species-wide innate beliefs.For more than one reason, then, appeal to the origins of humanity seems unpromis-

ing as the key to our question. It may of course provide materials for the specificaccount of how our species does in fact attain its relevant epistemic competences. Forthe reasons given, nevertheless, it does not seem right that the origin of our species isessentially implicated.Our lesson has two parts.Part 1. Moore is right to defend his common sense, including its core component

beliefs, but he is wrong to force our knowledge of them into the foundationaliststraightjacket characteristic of the epistemological tradition, with its immediate evi-dence epistemically supporting our body of beliefs.12

Part 2. Wittgenstein is right to reject the Moorean epistemology of indirect realism,and to offer an alternative picture of how our body of beliefs comes to be and hangstogether.What is wrong with foundationalism is the myth of the given, of presentational

mental states that supposedly yield justification just through their intrinsic character.There are two problems for this: first, the Speckled Hen problem; second, the apparentfact that we do not infer or base all the core components of our common sense pictureon such states. Our belief system falls into place as a whole in such a way that theepistemic standing of its core components does not derive wholly from some founda-tion of such self-presenting, given states.That is compatible with the fact that the child’s evolving system is visibly interlocking

when light dawns gradually on it. The system contains general as well as particularbeliefs, present observational beliefs included, as well as beliefs once observational andnow stored in memory. Vast systems of folk physics and folk psychology are alsoincluded. And there is of course much mutual support, including much that crossesthe generality lines. However, the epistemic standing of the whole cannot possibly


derivemerely from such mutual support. Nor is it plausible that the epistemic standing ofthe whole derives from its interlocking coherence along with the external, animalcompetence of the particular beliefs, the present observational ones and those stored inmemory from earlier observation. When one first believes that here is something red andround, that belief derives from the exercise of competence to discern colors and shapes,and this competence is tantamount to implicit beliefs (or commitments) of the form: if itlooks red (round) then it is red (round). And these must themselves have proper standing ifthey are to play their justifying role. And this standing cannot have derived merelythrough blatantly bootstrapping induction from corresponding particular beliefs.A human developmental competence is thus plausibly involved in the acquisition ofsuch perceptual abilities that come paired with corresponding implicit general beliefs (orcommitments).13

Wittgenstein seems alive to such insights. Unfortunately, he does not so much asaddress the sort of philosophical question that drives Moore to his straightjacket. Hedoes not take up the issue of just how it is that our core components of common senseamount to knowledge, if not in the way of Moore’s indirect realism.On Certainty doessuggest an answer, one not entangled misleadingly with the pragmatics of language useand with implausible antirealism. But the answer suggested is an equally implausiblerelativism, where knowledge and justification are supposed to derive just from what isconsensually and foundationally agreed upon by our culture, or by humanity at large.Given the potential for epistemically unfortunate groupthink and its ilk, it is implaus-ible to let sheer agreement constitute the fundamental basis for rationally justifiedbelief.

That yields the following desiderata:

1. Joining Moore in defending realist common sense.2. Joining Wittgenstein in rejecting Moore’s traditional epistemology of indirect,

foundationalist realism.3. Taking up the philosophical project of explaining how we can possibly know the

things that according to plain common sense we do know, including the corecomponents voiced by Moore.

4. Rejecting the sort of answer to that very general philosophical question suggestedby Wittgenstein’s response to Moore in On Certainty, once this is disentangledfrom the pragmatics of language use. That sort of answer amounts to an implaus-ible cultural or human relativism.

5. Rejecting also any appeal to the actual origins of humans, with our cultures andbelief systems, as essential to the current epistemic status of the beliefs in suchsystems. It seems conceivable that believers could come into being accidentally,eventually becoming rational believers and indeed sophisticated knowers none-theless.

Wittgenstein observes that the developing child “acquires” his belief system withoutthe explicit instruction that would make such acquisition a case of learning. Few if any

194 ERNEST SOSA

children are ever explicitly told that “there is a world of physical bodies in threedimensions, etc.”Nor is the system plausibly learned through inference from particularobservations. The child learns rather to interact effectively with such a world, and his“beliefs” are just manifest in his behavior the way deep prejudice is manifest in behaviortowards people perceived as members of the target group. In neither case need there beconscious formulation of belief contents. Such beliefs can remain implicit, and mighteven be denied if brought to consciousness.Such beliefs are acquired in interlocking systems. So, why not say that the basic

general beliefs are acquired with the support of specific evidence, such as supportivepositive instances, through inductive inference to the best explanation, just as Mooresuggests? Yes, possibly, but this is still at odds with the foundationalist model wherebythere is first a body of self-standing evidence (in epistemic or temporal order), whichcan then serve as a supportive source for general beliefs. No, the so-called evidenceitself requires general beliefs coordinately in place (at least in epistemic if not also intemporal order). So, we still face the question of how these general beliefs acquire theirepistemic standing, one that they cannot derive wholly from so-called evidence.14

Here’s a way to put our question in terms of the sort of competence that we haveseen already to be required: What sort of competence is manifest in the child’sdevelopment of his evolving system of common sense belief? One sort of competenceis the holist reasoning competence that keeps things coherent. But is only that sort ofcompetence required? Surely not; surely there can be ill-grounded systems of beliefthat manifest the rare and imaginatively coherent thinking of the paranoid. Shall wethen say that, in addition to his competence towards rational coherence, the evolvingthinker needs also perceptual and mnemonic competences, along with the socialcompetence that enables sensitive reception of good testimony? Yes, but now wemust ask: Is this a sequential process, temporally or at least logically, whereby thethinker first acquires lots of specific data, from which eventually he generalizes to hiscommon sense general principles? It is this that seems doubtful. It is doubtful that wecan have the required database absent principles already in place.That being so, the competence that leads to the general principles cannot be

constituted by pure perception, memory, reception of testimony, and introspection,unaided by general beliefs. No, general beliefs (or commitments) are needed from thestart. And in order for them to play their proper epistemic role, they must have beencompetently acquired. But this competence, involved in the initial acquisition of thosebeliefs, cannot be explained just in terms of inference from specific, particular dataalready in place.Some such general beliefs must hence have a source beyond inference from antece-

dent data. The operation of that source is hence not just a matter of rational basing. Noris it a matter of sensory experience directly supporting principle after principle seriatim.As Wittgenstein suggests, such a body of core beliefs is more plausibly acquired as asystem on which, as a whole, light gradually dawns. True, it is acquired together withmany specific details about the particulars of the young thinker’s life, and together with

INTUITIONS AND FOUNDATIONS 195

the many sense experiences that help constitute that thinker’s perception of hisenvironment, and with the effects of such experiences through perception or memory.

Such clusters of belief are plausibly required for understanding of the propositionsbelieved, just as Wittgenstein suggests. True, any one of them is dispensable withoutloss of understanding. But some critical mass must remain if the understanding is also toremain. Nevertheless, I see here no sufficient reason to deny that understanding-constitutive beliefs can amount to knowledge. Nor do I see why we should denythat the propositions understood can correspond to a world of facts metaphysicallyindependent of the thinker who understands, and even of human thinkers altogether.

Core beliefs constitutive of common sense are understanding-related in two ways.First, none requires a reason that surpasses the sheer grasp of its content. Second, each ispart of a cluster, understanding of whose members requires belief in a critical mass ofthem. Thus, understanding is in each case internally connected with a tendencytowards acceptance.

But how does that relate to the epistemic standing of any such cluster? Might therenot be a similar understanding-constitutive cluster associated with, for example, atheory of phlogiston? Yes, but no existential beliefs concerning phlogiston are requiredfor that cluster to play its proper understanding-constitutive role. All right, but whythen might we need existential beliefs concerning physical bodies in three dimensions,etc., for the common sense cluster to play its proper understanding-constitutive role?

There is an important difference here, and again Wittgenstein’s picture is relevant.I mean specifically his notion of how some basic beliefs are constituted, of action beingat the bottom of our belief system. If such basic beliefs of common sense must bemanifest in the subject’s actions, those beliefs will then have existential import, unlikethe purely theoretical beliefs constitutive of phlogiston theory. In acting as he doeswhen he reaches for fruit, or runs from a predator, the common-sense believer ismanifesting not just the conditional belief that fruit would satisfy certain conditions, orthat a predator would satisfy its own conditions. No, in each case his action manifeststhe further belief that something here and now existentially manifests the relevantconditions.

True, while asleep a common sense believer need have no there-and-then existen-tial beliefs. Even if now asleep, however, he must have lived through some waking life,during which he can hardly have avoided all action, and hence all correspondingexistential beliefs, ones that will now through memory support a body of stored beliefs,including the core (now existential) components of common sense.

Even given all of the above, we still face our unresolved issue. How do such generalcore components of common sense become knowledge? Suppose that, unlike Witt-genstein, we take up this question without commitment to relativism, while, unlikeMoore, we eschew traditional foundationalist indirect realism, but still hold to anobjective realism of common sense. How then do we so much as sketch an answer toour general philosophical question?

196 ERNEST SOSA

We must first recognize intuition as a source of knowledge. We must recognize asource that goes beyond perception, introspection, testimony, and inference, whetherdeductive or inductive, and indeed beyond rational basing generally. Some things weknow are not known in any of those ways. Simple arithmetic, geometry, and logicprovide examples aplenty. In such cases, we know something that we believe or areinclined to believe with no rational basis beyond our understanding of the question.As we have seen, however, irrational biases and superstitions can share that feature

with our simple beliefs of math or logic. So, not all intuitive beliefs are acceptable asipso facto rationally competent. Only by deriving from a competence does such a beliefor seeming constitute a rationally competent intuition. Simple beliefs of math or logicqualify. These are rationally competent even independently of any rational basis thatthey may have. Biases and superstitions imbibed arationally do not qualify. These arenot only arational but also irrational. An intuition is rationally competent only when itmanifests competence.The sort of indirect realism favored by Moore, a kind of traditional foundationalism

of the given, was said near the outset to be defensible as an account of a positiveepistemic status that the beliefs of knowers can share with the beliefs of victims inskeptical scenarios. But that status was said to fall short of the fuller epistemic compe-tence involved in our knowledge of contingent facts, particular and general. We arenow, I believe, in a position to pose a more serious problem for such internalism.Suppose Moore is right that we know with certainty the propositions constitutive of

our general picture of our contingent world. And suppose Wittgenstein is right that weneither acquire nor sustain that picture through ratiocinative learning. The epistemol-ogy of given evidence plus inductive inference does not explain our basic, generalcommonsense knowledge.We have drawn so extensively from Moore and Wittgenstein in hopes of help with

our problem of justified intuition: What distinguishes intuitions that are epistemicallyjustified from those that are not? The answer, we reasoned, may not be far to seek oncewe recall that even the presentational given can provide justification only through thesubject’s relevant competence. Perhaps we need only invoke a competence that is notreason-based, one that sub-personally enables reliable discernment of the true from thefalse in the relevant class of beliefs. This competence would yield seemings, attractionsto assent, even those that prove illusory as in the sorites paradox. Paradox-enmeshedseemings would still be competent so long as they derive from a normal humancompetence. In this regard they are analogous to the perceptual seemings of thechild who first encounters a Müller-Lyer setup. But what is it that distinguishesseemings that are competent from those that are not?Why not posit the epistemic efficacy of all seemings, adopting thereby a latitudinar-

ianism of seemings even beyond the intuitive? In my view this is plausible only throughconfusion of sensory experiences with perceptual seemings. Once having confusedthese, one might then extend to all seemings an epistemic standing that plausiblybelongs only to sensory experiences and not to perceptual seemings. Even with my


eyes closed, deprived of visual experience, it just might seem to me (intellectually) thatI face something red and round, or even that I see something red and round. Is such aseeming still ipso facto justified? Does it even prima facie justify the correspondingbelief? Not at all plausibly. But the same would then seemingly apply to seemingsgenerally. It is not their sheer existence that renders them justified, foundationally so. Itis rather something about them, even if it remains to be seen what this might be.

Take the case of a perceptual seeming. What renders such a seeming justified,foundationally so? Its reflecting a concurrent experience with the same content, itmight be thought, at least in part. That is part at most of the explanation, however; inaddition, as we have seen, the subject must not exceed his subitizing limits. Whatfigures more importantly at the basis of foundational justification is hence, I submit,competence, the sort of competence involved in the subject’s ability to subitize.

It might be argued to the contrary that what is important is the involuntary characterof seemings. Is it not rather this that puts them beyond relevant rational justificationand unjustification? Not so: plenty of beliefs are similarly involuntary without beinginsulated from rational assessment. Seemings, attractions to assent, are concept-involv-ing, just as are beliefs. One cannot be thus attracted to what one does not understand,but understanding requires concepts. Why should seemings be exempt from rationalassessment, if beliefs are never so exempt?

The bottom line is that internalists have no good recourse to mere seemings in order toexplain how it is that we are internally justified epistemically in the component beliefs ofour commonsense picture, beliefs acquired and sustained without ratiocination.

“Too bad,” some might say, some do say. Skepticism is then the upshot. What would berequired for knowledge of our commonsense picture? For the skeptic what would berequired is still that it be soundly inferred from the given, from the particular self-presentingexperiences undergone by the subject. To the extent thatWittgenstein and others make itplausible that this is not what happens in human life, to that extent must we then grant thathumans just do not know what they take themselves perhaps too readily to know.

But why? Why insist that the only way to know is through inductive inference fromthe presentational given? “For one thing,” it might be replied, “only thus could wehope to understand how we do know what we know. What other explanation couldthere be?”We find ourselves believing in a world of physical objects, some of them livehuman bodies, etc., and believing that we ourselves have been alive in such a world fora long time, and so on, for a vast set of propositions constitutive of one’s commonsensepicture. How can we possibly understand our knowledge of those things if not in theterms of “inference from the given”?

Well, consider this: Given our need to appeal to competence, to subitizing compe-tence, and given the place of competence in understanding basic knowledge of logic,arithmetic, and geometry, our understanding of how we know seems limited in anycase. How well do we really understand the variation among humans in their subitizingabilities, or the theoretical basis for such abilities? Besides, we also vary vastly in ourpowers of mathematical intuition. As far as I know, we have no good account of how it

198 ERNEST SOSA

is that Ramanujan was able to excel so far beyond the rest of humanity. None of this,however, holds us back from attributing to ourselves the knowledge that we enjoyeither foundationally through such intuition (within the bounds of one’s individualendowment) or inferentially based on such foundations.At several crucial junctures, then, the vaunted explanation of our knowledge, on the

basis of inference from the given, is limited by our need to attribute a competence ofwhich we have only slim understanding. Once again, it is the appeal to competencethat seems crucial, even while we still hope to gain fuller understanding of theworkings of such postulated competence.We are finally in a position to formulate our account as follows.

R IS has a rationally competent intuition that p if and only if:(a) just understanding (i.e., the proposition that p) attracts S to assent, and

does so regardless of whether S has any motivating reason for such attraction,beyond his understanding; and (b) the mechanism of understanding-basedattraction that accounts for S’s attraction to assent to , with no dependenceon any (further) rational basis, is an epistemic competence.

And we can now distinguish human rational intuitions from cultural rational intuitions:

HR IS has a human rationally competent intuition that p if and only if:(a) S has a rationally competent intuition that p; and (b) the mechanism of

understanding-based attraction that accounts for S’s attraction to assent to is one that operates in humans generally in their normal course ofdevelopment. Humans would normally be attracted to assent to bysheer understanding, even absent any differential causal mechanism (such asthose that cause biases). Normal human development would give S the con-cepts constitutive of along with an attraction to assent to .15

CR IS has a cultural rationally competent intuition that P if and only if:(a) S has a rationally competent intuition that p; and (b) the mechanism of

understanding-based attraction that accounts for S’s attraction to assent to is one that operates in members of S’s culture generally in their normalcourse of development. Members of S’s culture would normally be attracted toassent to with no rational basis being required, since attraction to assent to is instilled through enculturation in that culture.16

Note, finally, that epistemic competences can be of use in epistemology even in the absence of adetailed theory of their nature and operation. We can appeal to such competences inepistemology even with limited understanding of their modus operandi.17 This in factapplies not only to rational intuition but also to introspection and even to perceptionand memory. Up to a point, people could surely know how they knew things evenbefore we gained our fuller understanding of how perception actually works, in itsvarious modalities. Our knowledge through various perceptual sources is of course

enhanced with our improved understanding of their nature and operation, but thisboost of meta-sophistication is not essentially required. It is not required for animalknowledge through such competences. Nor is it required for somemeasure of reflectiveknowledge, even through limited understanding of the operative competences.

Notes1. And the kindred view that intellectual seemings, or attractions to assent, are ipso facto

“justified” is subject to a similar shortcoming.2. Presumably not even internalists would think that any belief whatsoever held in the absence

of reasons is ipso facto prima facie justified, which raises for them the question of what thefurther source of justification might be, beyond rational basing. This will be seen to pushadvocates of internal justification towards a more objective conception that includes rationalcompetence and not mere blamelessness.

3. A belief with propositional content may have as its conceptual content the thought [t]when [t] is a mode of presentation of to the subject at the time. Although a full treatmentwould need to go into this distinction and its implications, here for simplicity we work withpropositional contents only.

4. Whether the items are moving, for example, or very big, or qualitatively different from eachother, etc., will presumably matter.

5. This all suggests that the importance of experience in epistemology is vastly overrated. Majorcategories, and distinctions among them—such as that of a priori versus a posteriori know-ledge—should not turn on a matter of such limited importance by comparison with compe-tence. (Nor of course should major historical divisions—such as rationalism vs. empiricism—bedefined by reference to experience, unless it is clear beyond reasonable doubt that this is howthe parties themselves viewed the matter, as indeed it is at least in some cases.)More plausibly wecan highlight our competent (of course) knowledge of particular contingencies through somecombination of introspection, perception, testimony, or memory; and we can then distinguisha posteriori knowledge as knowledge of such contingencies or knowledge explained exclusivelythrough inference from such knowledge. A priori knowledge would then be explainable at leastin part through some source other than such inference from knowledge of particular contingen-cies. This would include abstract knowledge (not of particular contingencies) so explainable.

The notion of competent access to particular data (mainly through observation) wouldsuffice to pose the question of whether only through a posteriori, scientific, inductivetheorizing could we gain access to any abstract, general knowledge. Defenders of empiricalscience could then line up on one side, while on the other side would gather rationalistbelievers in direct access to abstract, general truths.

6. Objection: It is not self-contradictory to say “the one line seems like it’s longer, but I knowit’s not; I know this is an illusion and I have no inclination to believe it really is longer.”Reply: Indeed it need not be; but we must distinguish the intellectual seeming, where one isattracted to assent, from being inclined to assent, which is a resultant attraction that is positive, evenwhen it has not reached the intensity required for belief. In addition, ‘seems’ seems ambigu-ous, and might be used to pick out just the experience, and not at all an intellectual seeming,whether initial or resultant.

200 ERNEST SOSA

7. Reid’s commonsense philosophy is also relevant, or so I would argue (but not here).Relevant as well is the historical innatism debate, as is the contemporary nativism debate.It is widely (though not universally) believed that knowledge of linguistic universals isinnate. And a similar view seems plausible for information encapsulated in our fullyfunctioning perceptual faculties triggered through normal childhood development.

8. At the outset of his paper Moore writes:

(1) I begin, then, with my list of truisms, every one of which (in my own opinion) I know,with certainty, to be true. The propositions to be included in this list are the following:

There exists at present a living human body, which is my body. This body was born at acertain time in the past, and has existed continuously ever since . . . ; and, at every momentsince it was born, there have also existed many other things, having shape and size. . . . Amongthe things which have, in this sense, formed part of its environment (i.e. have been either incontact with it, or at some distance from it, however great) there have, at every moment sinceits birth, been . . . other living human bodies, each of which has, like it, (a) at some time beenborn, (b) continued to exist from some time after birth, . . . . Finally (to come to a different classof propositions), I am a human being, and I have, at different times since my body was born,had many different experiences, of each of many different kinds: e.g. I have often perceivedboth my own body and other things which formed part of its environment, including otherhuman bodies; I have not only perceived things of this kind, but have also observed facts aboutthem . . . ; I have been aware of other facts, which I was not at the time observing, such as, forinstance, the fact, of which I am now aware, that my body existed yesterday . . . ; I have hadexpectations with regard to the future, and many beliefs of other kinds, both true and false;I have thought of imaginary things and persons and incidents, in the reality of which I did notbelieve; I have had dreams; and I have had feelings of many different kinds. And, just as mybody has been the body of a human being, namely myself, who has, during his lifetime, hadmany experiences of each of these (and other) different kinds; so, in the case of very many ofthe other human bodies which have lived upon the earth, each has been the body of a differenthuman being, who has, during the lifetime of that body, had many different experiences ofeach of these (and other) different kinds.

And he adds the following, as a second important sort of proposition, also part of commonsense, (2) that he also knows with certainty: that many other people have known aboutthemselves and their bodies propositions corresponding to the ones that according to (1)above Moore knows with certainty about himself and his own body.

9. See especially the following entries in On Certainty: 94, 100, 101, 116, 136, 138, 141, 142,151, 192, 225, 232, 253, 273, 274, 275, 279, 287, 358, 359, 410, 411, 475, and 538.

10. Indeed, Moore makes some of the relevant points in a letter to Norman Malcolm just priorto the conversations between Malcolm and Wittgenstein that led to On Certainty. SeeT. Baldwin (ed.), G.E. Moore: Selected Writings (Routledge, 1993), pp. 213–16. The letteris dated 28 June 1949. Wittgenstein visited Malcolm from July to October of that year, andtheir conversations inspired Wittgenstein’s work until his death some two years later.

11. That is the sort of answer that according to Sellars is to be found in perennial philosophy’streatment of the manifest image.

12. How is this traditional epistemology related to a Platonic/Cartesian dualism, for which weare embodied souls? On that view we are pilots in a ship that must be steered from within


based just on the deliverances of certain screens or loudspeakers, where we have no reason totake these deliverances at face value, but must work out the extent to which they deserveour trust. We must hence rely on our reason to infer from our auditory and visual data to thereality outside the ship. Once we see ourselves as essentially embodied human beings, thismay free us more plausibly to include in our basic endowment the innate knowledgeconstitutive of our ability to act. Souls too, after all, need inbuilt knowledge constitutiveof their intuitive and deductive reason, of their ability to act intellectually. (Descartes does ofcourse explicitly deny that his soul’s embodiment fits the model of a pilot in his ship; but it isquite a mystery how he avoids this, and indeed how causal interaction can take placebetween two substances across his vast metaphysical divide.)

13. The general standing belief might plausibly have as its content that if things seem percep-tually to be thus and so, they tend to be or habitually are thus and so. This belief might be eitherinnate or installed through subpersonal mechanisms in the normal development of the child.Once installed, moreover, it would then be applied through a kind of logical reasoning tospecific situations again and again. If we view the matter thus, then the specific commitments(that here now, for example, if things look thus and so then they are thus and so) would bereason-based after all, with the general-tendency-belief furnishing the reason applied to thespecific instances.

14. This harks back some paragraphs to the point that foundationalism posits a pyramidalstructure wherein the standing of general upper-level contingent beliefs derives whollyfrom the evidence provided by the observational lower-level beliefs through inference,largely through inductive inference.

15. Human rational intuitions seem thus in line with the core components of Moore’s commonsense. They are also in line with Wilfrid Sellars’s manifest image, however, and withP. F. Strawson’s descriptive metaphysics, the metaphysics of the human form of life, sharedin common, and not just peculiar to any particular culture. They are beliefs fundamental to ahuman conceptual scheme that much of perennial philosophy aims to disclose and system-atize. Even if there is more than one optimal way of doing so, articulating even one would bean achievement to celebrate. Both Strawson and Sellars seem right about that, even if thatleaves open the ontological status of the result, whether as convenient fiction or as objectivefact.

16. If people can belong to more than one culture, then CRI needs to be tweaked. And, in anycase, not all rational intuitions are of either the human or the cultural variety. Consider,again, the superhuman intuitions of Ramanujan. It might be objected that the postulated“intuition” is motley, but is it more so than “perception”?

17. This is defended in my “Minimal Intuition,” in Michael R. DePaul and William Ramsey(eds) Rethinking Intuition: The Psychology of Intuition and Its Role in Philosophical Inquiry(Rowman and Littlefield, 1999).

202 ERNEST SOSA

III

Skepticism and the A Priori

9

Skepticism, Reason, and Reidianism

Joel Pust

1 The Cartesian Perspective and the TraditionalEpistemological Problems

The traditional problems of epistemology include the problems of justifying our perceptualbeliefs, our beliefs regarding other minds, our inductive beliefs, our testimonial beliefs,and our beliefs in unobservable entities. These beliefs of ours are thought to give rise tojustificatory problems sharing a common structure. In each case, the epistemologist (at leastone gripped by the problem in the traditional way) seeks a rationally cogent argument fromthe truth of propositions in one domain to the truth of propositions in another domain. Forexample, in the most prominent of these problems—the problem of the external world—what is sought is a reason for thinking that ordinary physical objects exist and have roughlythe properties we believe them to have. This reason must, it is usually assumed by thetradition, take the form of a reason to think that the occurrence and character of ourintrospectible experiences make suitably probable a significant percentage of our common-sense beliefs regarding ordinary physical objects and their properties. Something similar holdsof the other problems aswell, aswe are to begin from a set of justified propositions (regardingthe behavior of other bodies or past observable correlations or experimental results) and toprovide a reason for thinking that the truth of those propositions makes it sufficientlyprobable that some other proposition (regarding the mental states of other persons orsome future correlation or the existence of some unobservable particle) is true.In spite of the undeniable historical importance and the persisting influence of

this traditional picture of epistemology and its proper tasks, many epistemologistsnow reject such a picture. My focus in this chapter will be on a particular group ofphilosophers—consisting of Thomas Reid and a number of contemporary philosopherswho claim a kinship with Reid—who seek to undermine the notion that there is adistinctive justificatory problemwith respect to any of the beliefs previously mentioned.They do not claim that the tasks set out in these traditional problems can be accom-plished. Instead, they vigorously maintain that they cannot be. However, they reject asarbitrary, and hence unjustified, the deference to a priori intuition and introspection onwhich the traditional problems are based.

In the first part of this chapter, I’ll explain why Reid, and those “Neo-Reidian”philosophers on which I shall here focus, reject the standard formulation of theseproblems. In the second part of the chapter, I will attempt to defend the traditionalperspective on these problems against the Reidians, a perspective on which a priorijustification is absolutely fundamental and deserves its privileged position as a source ofevidence on which all others must be grounded. I’ll close with a brief account of someinconsistent concessions of Reid’s to the traditional perspective and an argument that itis actually the Reidians who are guilty of arbitrariness.

Many of the epistemological problems mentioned earlier arise only on the assump-tion that other problems are solved. The problem of other minds, for example,typically presupposes that we are justified in propositions regarding the existence andbehavior of bodies which are themselves the topic of concern in the traditionalproblem of the external world. Likewise, standard formulations of the problem ofinduction or testimony normally suppose us to be justified in believing in pastcorrelations and in the existence of other minds. It is natural to think that becausethese problems form an interlocking structure in which justification for believing oneclass of propositions rests on justification for believing another class, sufficient effort willreveal to us a set of propositions the justification of which does not depend on havingjustification to believe some other proposition. Let us call these propositions “thefoundationally justified propositions.”1

Many traditional epistemologists take the foundationally justified propositions to beof a very limited sort and would endorse the following principle:

The Narrow Foundations Principle (NFP): The only foundationally justified propositions arepropositions about one’s present state of consciousness and propositions which are the contentsof a priori intuitions.

In addition, many would impose a similar constraint on what is required of a rationallyacceptable support relation, endorsing something like the following principle:

The Inferential Justification Principle (IJP): Beliefs in propositions which are not foundationallyjustified are justified only if there are foundationally justified propositions which entail that suchnon-foundational propositions are sufficiently probable.

The conjunction of these two claims constitute what I shall call “The CartesianPerspective.” In so designating it, I do not mean to claim that Descartes agreed withboth of these claims. Furthermore, this perspective (or something relevantly similar) isshared by philosophers such as Locke, Berkeley, and Hume, who may not count, atleast on some influential taxonomies, as Cartesians. Still, in spite of these qualifications,it is clear that this perspective, and the attendant problems to which it gives rise, havebeen shared by a great number of past and present epistemologists.2

That these principles are definitive of the position under attack by the Reidians isshown by the following passages in which Reid characterizes his target. First, from theInquiry:

206 JOEL PUST

That our thoughts, our sensations, and everything of which we are conscious, hath a realexistence, is admitted to this system as a first principle; but everything else must be made evidentby the light of reason. Reason must rear the whole fabric of knowledge upon this single principleof consciousness. (1967, 206)

And from the Essays:

This, therefore, may be considered as the spirit of modern philosophy, to allow of no firstprinciples of contingent truths but this one, that the thoughts and operations of our own minds,of which we are conscious, are self-evidently real and true; but that everything else that iscontingent is to be proved by argument. (1967, 464)

This “method of philosophising,” Reid avers, is “common to Des Cartes, Male-branche, Arnauld, Locke, Norris, Collier, Berkeley, and Hume” (1967, 468).

2 Reidianism and the Cartesian PerspectiveWhile the Cartesian Perspective has no doubt had a huge influence on the subsequentdevelopment of epistemology, many contemporary epistemologists take issue with theperspective. A recognizable group of the contemporary critics of theCartesian Perspectiveexplicitly claim to be developing ideas which originated in Thomas Reid’s 18th-centuryresponses to the Cartesian Perspective. Among the most influential of these “Neo-Reidian” philosophers are Alvin Plantinga, William Alston, and Nicholas Wolterstorff.3

In this section, I characterize the views common to Reid and these Neo-Reidians.In what follows, I will be especially interested in the following claims of Reid and

the Neo-Reidians:

Conditional Skepticism (CS): Most of our ordinary beliefs fail to satisfy the demands of theCartesian Perspective.4

Anti-Cartesianism (AC): The demands of the Cartesian Perspective are unjustified because theydepend upon an arbitrary partiality toward a priori rational intuition and introspection.

Anti-Skepticism (AS): Most of our ordinary beliefs are justified.

Among the beliefs which the Reidians take to fail the demands of the CartesianPerspective are beliefs regarding the external world, beliefs in other minds, beliefsregarding the future, and, according to Reid himself, belief in a persisting self, memorialbeliefs, and beliefs based on testimony. My main focus in this chapter will be on AC.AC is obviously more fundamental than CS because if it is correct, then little of realepistemological significance hangs on whether or not CS is correct. In the final section,I’ll argue that there is a deep tension between these three claims of the Reidians.One might reject the Cartesian Perspective without necessarily indicating whether

one disagrees with NFP, IJP, or with both principles. It seems that the Reidians wouldprobably reject only NFP and would accept some construal of IJP. Of course, the moreone includes in the foundations, the less one needs to avail oneself of inferential routes

SKEPTICISM, REASON, AND REID IANISM 207

to justified belief and so the significance of IJP is correspondingly diminished. Further-more, if a variety of inferential principles are taken as foundationally justified, accept-able inferential routes become much more numerous. So, I will take the Reidians to berejecting NFP but retaining the view that our justified belief or knowledge has afoundational structure, even though their view of the basis of our knowledge and theconditions for acceptable epistemic ascent may be quite different.5,6

3 Neo-Reidianism and Reformed EpistemologyIt will not escape notice that the philosophical triumvirate—Alvin Plantinga, WilliamAlston, and Nicholas Wolterstorff—which I’m treating as representative of Neo-Reidianism consists of the three main proponents of that movement in analyticphilosophy of religion which goes by the name of “Reformed epistemology.”7

A core commitment of Reformed epistemology is the claim that belief in the traditionaltheistic God, and indeed belief in the distinctive doctrines of Christian theism, isepistemically appropriate even though such beliefs do not meet the epistemic standardsof the Cartesian Perspective. These Reformed epistemologists are explicit about theirReidian heritage, and to Reid’s claims that beliefs formed on the basis of testimony,sympathy, sense perception, memory, and induction are epistemically acceptable inspite of failing the demands of the Cartesian Perspective, they add the claim that belief inthe theistic (or Christian) God is, at least under certain conditions, epistemicallyacceptable even though it also fails to satisfy the demands of the Cartesian Perspective.

Wolterstorff makes clear the relationship between the two views in the followingpassage:

[T]he deepest significance of that recent development in epistemology of religion which hascome to be known as Reformed epistemology is its insistence that the failure of the religiousbelief of the ordinary person to measure up to the demands of The Doxastic Ideal [i.e. theCartesian Perspective] does not, so far forth, indicate any sort of deficiency or impropriety inthose beliefs or in the person holding them. We can now state why the “Reformed epistemolo-gist” makes that claim. It’s not an ad hoc thesis concerning religious beliefs; rather, it’s part of therecent “revolution” in epistemology. The Reformed epistemologist stands in the Reidiantradition. Religious beliefs, in good measure, are like inductive and testimonial beliefs, in thatthe latter also do not exemplify The Doxastic Ideal. But that fact . . . implies nothing at all aboutthe presence or absence of truth-relevant merits in such beliefs; it proves to be a fundamentallyirrelevant observation. (1999, 317, emphasis added)

So, the Reformed epistemologists seek to run a two-part argument regarding religiousbelief. The first part consists of a “partners in guilt” argument for the conclusion thatreligious belief is no worse off with respect to the demands of the Cartesian Perspectivethan are beliefs in other minds, beliefs about the future, beliefs about external objects,and many other of our beliefs. All such beliefs fail the strictures of the CartesianPerspective.

208 JOEL PUST

This parity claim, however, does not challenge the legitimacy of the demands of theCartesian Perspective. In the absence of such a challenge, the theist receives only thecold comfort of knowing that her religious beliefs are as epistemically illegitimate as herbeliefs regarding the external world. Of course, the passage fromWolterstorff just citedalso reveals a second, and more significant, part of the neo-Reidian position: the claimthat the demands of the Cartesian Perspective lack rational justification.Let us take “Reformed epistemology” to consist of the three claims identified as

commitments of Reidianism and also a specific instance of CS conjoined with a specificinstance of AS:

Parity of Religious Belief (PRB): Belief in the theistic (and Christian) God fails to satisfy thedemands of the Cartesian Perspective but is (often) justified.

The Reformed epistemologist is, then, best viewed simply as a Neo-Reidian who goesbeyond Reid in identifying an additional class of beliefs which do not meet thedemands of the Cartesian Perspective but are also, allegedly, none the worse epistemic-ally for this failing and are, in fact, justified.8

Notice that Reformed epistemologists do not merely claim that many people arejustified in traditional theistic belief. That claim can consistently be accepted by a propon-ent of the Cartesian Perspective. Indeed, it was accepted by many of the early modernphilosophers who Reid would describe as “Cartesians,” as many of them took themselvesto have good arguments for the existence of God. What distinguishes Reformed episte-mology is its claim that belief in God is often justified even though it in fact fails thedemands of the Cartesian Perspective. Therefore, a defense of AC is essential to the successofReformed epistemology and, correspondingly, any successful defense of the demands ofthe Cartesian Perspective constitutes an undermining of Reformed epistemology.As noted, in addition to claiming that the relevant beliefs are not shown unjustified or

irrational by their failure to live up to the requirements of theCartesian Perspective, thesereformed epistemologists often claim that the relevant theistic beliefs have a suitablepositive epistemic status. As indicated by PRB, I shall take the status of interest to be thatof being justified. However, it should be noted that Reformed epistemologists some-times propose that theistic beliefs (and others failing the requirements of the CartesianPerspective) have some other positive status. Plantinga, as his views have evolved, hasmoved from a focus on good arguments for believing (1967) and rationality of belief(1983), to whether such beliefs have that property which separates mere true belief fromknowledge (2000). Alston focuses on whether it is prima facie practically rational to formsuch beliefs on the basis of putative mystical experience of God (1993). Wolterstorffrefers to the vague catch-all property of having “truth-relevant merits” (2001, 188).

4 Conditional Skepticism and the Cartesian PerspectiveIt is abundantly clear from his writings that Reid believed the demands of the CartesianPerspective were unsatisfiable and that he took all of the epistemological problems


mentioned at the beginning of this chapter to be completely insoluble in theirtraditional form. His remarks in the Inquiry on Hume and Berkeley show that hebelieved that if one adheres to the Cartesian Perspective one will be driven unavoidablyto skepticism about the external world and other minds:

Bishop Berkeley hath proved, beyond the possibility of reply, that we cannot by reasoning infer theexistence of matter from our sensations; and the author of the ‘Treatise of Human Nature’ hathproved no less clearly, that we cannot by reasoning infer the existence of our own or other mindsfrom our sensations. (1967, 129, emphasis added)

While one might reasonably question Reid’s assessment of the strength of Berkeley’sand Hume’s arguments, Reid clearly holds that, as traditionally construed by theCartesian Perspective, there is no possible solution to the problem of the externalworld or the problem of other minds. Again and again throughout the Inquiry andEssays, Reid responds to epistemological problems with claims that if the problems areconstrued as the Cartesian Perspective would construe them, they are insoluble.

Indeed, in some respects he goes further than many of the proponents of theCartesian Perspective. For example, of memory, he notes,

The same difficulty with regard to memory naturally arises from the system of ideas; and the onlyreason why it was not observed by philosophers, is, because they give less attention to thememory than to the senses; for, since ideas are things present, how can we, from our having acertain idea presently in our mind, conclude that an event really happened ten or twenty yearsago, corresponding to it? There is the same need of arguments to prove, that the ideas of memoryare pictures of things that really did happen, as that the ideas of sense are pictures of externalobjects which now exist. (1967, 358)

and then adds that

In both cases, it will be impossible to find any argument that has real weight. So that this hypothesis leadsus to absolute scepticism, with regard to those things which we most distinctly remember, no lessthan with regard to the external objects of sense. (1967, 358, emphasis added)

This conditional skepticism is also shared by the contemporary philosophers I haveidentified as Neo-Reidians, though it appears that, unlike Reid, they take themselvesto have merely inductive evidence for the view. Throughout his writings of the lastthree or four decades, Plantinga has repeatedly claimed that a clear lesson of the historyof modern and contemporary philosophy is that most of our ordinary beliefs “cannotbe seen to be supported by, to be probable with respect to beliefs that meet” (2000, 98)the requirements of the Cartesian Perspective.9 In his treatment of traditional skepticalproblems inWarrant and Proper Function (1993b) he repeatedly concedes that, relative tothe classical formulation of the problems, there is no solution. This pattern of conces-sion with regard to the classical skeptical problems appears in his treatment of our beliefthat we are persisting things, of our belief (on the basis of memory) that certain thingshave happened, of our belief that other minds exist, of beliefs formed on the basis of

210 JOEL PUST

testimony of other persons, of perceptual belief in the external world, a priori beliefs,and inductive beliefs.Conditional skepticism of the sort I’ve outlined is also endorsed by Alston in his

work on these topics. In The Reliability of Sense Perception (1993), he argues at greatlength for the claim that one cannot show in a satisfactory manner, on the basis of thesorts of resources allotted by the Cartesian Perspective, that sense perceptual beliefs arereliable. Wolterstorff, however, embraces a more tentative conditional skepticismabout the material world as he claims that “no philosopher . . . has succeeded inshowing that the existence of material objects is probable on evidence consistingexclusively of items of direct awareness which we can see to be satisfactory as evidence”(1996, 178).

5 The Charge of Arbitrary PartialityIn the previous section, I demonstrated that Reid and the Neo-Reidians believe thattraditional epistemological problems cannot be solved in the way which the CartesianPerspective would require. As indicated earlier, however, neither Reid nor the Neo-Reidians are skeptics. Instead, they contest the demands being made of our beliefs inthe formulation of these traditional epistemological problems. In the canonical passagecited over and over by the Neo-Reidians, Reid writes

The sceptic asks me, Why do you believe the existence of the external object which youperceive? This belief, sir, is none of my manufacture; it came from the mint of Nature; it bearsher image and superscription; and, if it is not right, the fault is not mine; I ever took it upon trust,and without suspicion. Reason, says the sceptic, is the only judge of truth, and you ought tothrow off every opinion and every belief that is not grounded on reason. Why, sir, shouldI believe the faculty of reason more than that of perception? They came both out of the sameshop, and were made by the same artist; and if he puts one piece of false ware into my hands,what should hinder him from putting another? (1967, 183)

Reid’s complaint is composed of two parts: a question and a claim. The question askswhat reason one might have for believing that reason is more trustworthy thanperception. The claim is that no reason will be forthcoming as the ground for refusingto trust sense perception without independent confirmation except the possibility of itsunreliability, but then exactly the same ground suffices to undermine trust in reason.So, Reid concludes that the requirement that one show some candidate source ofcontingent beliefs is reliable by appeal to reason and consciousness is unjustified. It isthis alleged inability to justify, in an acceptable way, the claim that our other faculties orbeliefs must be held to the bar of reason and consciousness which Reid views as thefundamental flaw in the Cartesian Perspective. On this basis, Reid concludes that theproponent of the Cartesian Perspective (and though he refers to this person as “thesceptic,” there is no reason why they must be) is guilty of an arbitrary partiality inendorsing NFP.

SKEPTICISM, REASON, AND REIDIANISM 211

This charge of arbitrary partiality is made even more explicit at other locations inReid’s oeuvre. Of Hume he writes,

The author of the ‘Treatise of Human Nature’ appears to me to be but a half-skeptic. He hathnot followed his principles so far as they lead him; but, after having, with unparalleled intrepidityand success, combated vulgar prejudices . . . his courage fails him, . . . and [he] yields himself acaptive to the most common of all vulgar prejudices—I mean the belief of his own impressionsand ideas. (1967, 129)

and of Descartes he writes,

It might have been objected to this first principle of Des Cartes, How can you know that yourconsciousness cannot deceive you? You have supposed that all you see, and hear, and handle,may be an illusion. Why, therefore, should the power of consciousness have this prerogative, tobe believed implicitly, when all our other powers are supposed fallacious? (1967, 463)

and, finally, more generally and most explicitly,

Thus the faculties of consciousness, of memory, of external sense, and of reason, are equally thegifts of nature. No good reason can be assigned for receiving the testimony of one of them, which is not ofequal force with regard to the others. The greatest sceptics admit the testimony of consciousness, andallow what it testifies to be held as a first principle. If, therefore, they reject the immediatetestimony of sense or of memory, they are guilty of an inconsistency. (1967, 439, emphasis added)

These passages make clear that, as Reid saw the matter, the Cartesian Perspective isincapable of being justified because no good reason can be given for trusting some ofour belief-forming tendencies but not all of them. The Cartesian Perspective, recall,takes only consciousness and a priori intuition as the necessary basis for the justificationof all belief and then deems belief unacceptable if unsupportable by those two sources.Reid’s complaint is that this partiality towards consciousness and reason has nojustification, and so one must either be a complete antecedent skeptic with whomrational discussion is impossible and who must be left to their skepticism, or one mustaccept all our “natural” sources of belief as prima facie epistemically acceptable andinitially on a par.

Before proceeding to evaluate this suggestion, it is worth briefly locating it in theNeo-Reidians. Alston claims that the partial skeptic about, say, sense perception, isfaced with exactly the dilemma outlined earlier: Either she demands an independentjustification of all faculties and is driven into a complete and total skepticism or sheaccepts some faculties without independent verification and demands a justification ofothers. This second horn, Alston avers, involves an arbitrary selectivity since “there is norational basis for accepting some and not others without justification” (1985, 446).

Plantinga also repeatedly makes the charge of arbitrary partiality. As one example, inthe context of his discussion of Bonjour’s (1996) adherence to a version of theCartesian Perspective, he writes:

212 JOEL PUST

[T]he question for Bonjour is this: why are simple a priori beliefs exempt from this demand thatwe must have a good reason for thinking a belief true, if it is to constitute knowledge? He seemsto me to face a dilemma at this point. If he treats a priori belief differently from perceptual belief,belief about other minds, inductive beliefs, and the like, then he’s guilty of a sort of arbitrari-ness . . . On the other hand, if Bonjour insists that all beliefs, a priori beliefs as well as others, mustmeet this condition, then he insists on a condition for knowledge that cannot possibly be met, atleast by finite beings like us. (1996, 342)

Finally, Wolterstorff ’s (2001) exposition and defense of Reid’s epistemology is par-ticularly clear in presenting the Reidian point as a dilemma for the proponent of theCartesian Perspective. The Cartesian may try to ground all of our beliefs on introspec-tion and reason, but “what difference is there, between perception and memory on theone hand, and consciousness and reason on the other, that would authorize this radicaldifference in treatment?” (2001, 198). Alternatively, she may seek an independentground for every putative source of belief and thereby demand an impossibility.This argument suggests to the Reidians that one is justified in trusting all of one’s

belief-forming tendencies and divesting oneself of their deliverances only when one isrationally compelled to do so. As Alston puts it

Reid’s point could be put by saying that the only (noncircular) basis we have for trusting rationalintuition and introspection is that they are firmly established doxastic practices, so firmlyestablished that we “cannot help it,” and we have exactly the same basis for trusting senseperception, memory, nondeductive reasoning, and other sources of belief for which Descartesand Hume were demanding an external validation. (1993, 127, emphasis added)

Reason, inductive inference, perception, introspection, memory, and testimony are allof them equally appropriate sources of prima facie justified belief. As noted earlier, theNeo-Reidian movement in the hands of the Reformed Epistemologists maintains thatcertain sorts of propositions entailing the existence of the traditional (and Christian)God are themselves among the foundationally justified propositions.

Whatever else may be said about them, the Reidians have succeeded in raising anextremely important foundational question in epistemology, one which the proponentof the Cartesian Perspective must address rather than evade. That question is, given anyputative set of foundational resources, why restrict oneself to exactly that set of resourcesin an effort to determine which of our beliefs are justified? In the next section, I shall tryto argue that, contrary to the Reidians, the Cartesian Perspective has the resources toanswer this question.

6 Against the Charge of Arbitrary PartialityPrior to attempting a schematic defense of the Cartesian Perspective against the chargeof arbitrary partiality, we must be clear about what is required of such a defense.A successful defense of the Cartesian Perspective against the charge of arbitrarypartiality requires two things: First, an account of the properties of reason and intro-


spection which distinguish them from other bases of belief; Second, an account of whythose differences suffice to justify and render non-arbitrary the partiality captured inNFP. In sum, what is needed is an account of what epistemologically relevantdifferences distinguish reason and introspection from our other faculties.

Let us begin with a consideration of the possibility that comes most quickly to mind.Perhaps it will be said that reason and introspection are to be distinguished from senseperception because of their infallibility. That is, perhaps we may say that reason andintrospection are infallible, but that sense perception and other sources of belief areplainly fallible and lead us sometimes into error. The Neo-Reidians consider and rejectthis suggestion. Alston notes that some have claimed that “we are infallible with respectto our current states of consciousness, or at least that nothing could show that one hasmade a mistake about such matters” (1993, 128), and that “similar claims have beenmade for rational intuition” (1993, 128). His response is that it is “not at all clear” thateither of these faculties is infallible or incorrigible. With respect to rational intuition, henotes that philosophical disagreement seems to indicate that rational intuition is notinfallible. These considerations show, he avers, that it is not infallibility that distinguishesrational intuition from sense perception and non-deductive inference.

Plantinga and Wolterstorff are also united in claiming that a priori reason is notinfallible and so this cannot be the justifying ground of the narrow foundations principle.

Don’t the same questions arise about a priori beliefs and alleged knowledge? Can’t we raisethe same skeptical problems? I believe the corresponding conditional of modus ponens and that2 + 1 = 3; indeed, I believe each necessarily true. . . . They have about them, furthermore, thepeculiar feel that a priori beliefs have—that feel that somehow they just couldn’t possibly be false.But of course such a feel could be misleading. A false belief, obviously enough, could have thatsort of feel for me; I could be mad, or a victim of an Alpha Centaurian cognitive scientist, or abrain in a vat, or a victim of a Cartesian evil demon. . . . As a matter of fact, this isn’t merely anabstract possibility: some propositions that have that a priori feel about them are false, as is shownby certain versions of the Russell paradox. (Plantinga, 1996, 341)

It happens all the time that some proposition appears to a person as a proposition that isnecessarily true would appear, when it’s not; or that an argument appears to a person as anargument that is valid would appear, when it’s not. It’s only when we go beyond how theproposition or argument presents itself to us on that occasion, and explore its connections withother propositions and arguments, that we discover the truth of the matter. What’s especiallydisturbing is that sometimes the members of a set of propositions all retain the ‘glow’ of necessarytruth when we rightly come to realize that they can’t all be true, let alone necessarily true: witnessRussell’s Paradox. This already undercuts the skeptic’s injunction; if reason cannot be trusted,then the project of assembling evidence pro and con about the reliability of memory andperception can’t even get off the ground. (Wolterstorff, 2001, 200)

On one point, the Neo-Reidians are clearly correct. However things may stand withintrospection, it cannot be on account of their infallibility as a class that a priori rationalintuitions are epistemically superior to putative perceptions. We know from philo-sophical paradox that false propositions can be the contents of a priori intuitions. If there

214 JOEL PUST

is reason to favor reason over sense perception, it cannot be that the former is infalliblewhile the latter is fallible.However, the ground of serious skeptical worries is not mere fallibility. Instead, it is

the possibility of general unreliability or even total unreliability. While perceptualjudgment is fallible, it is also possible, as skeptical scenarios make clear, that each andevery one of our perceptual judgments is mistaken, consistent with our experiences.So, the skeptic asks, what reason have we to think that they are even reliable?The plausibility of the Reidian view requires that we have reason to think that it is

indeed possible for a priori intuition to be unreliable. In the passages quoted earlier,Plantinga andWolterstorff support this claim by noting (a) that a false proposition couldseem intuitive to one, and (b) that the Russell paradox shows that in fact some falsepropositions actually do seem very intuitive.10 That a priori intuition is fallible—that afalse proposition can present to intuition as true—does not, however, imply that it ispossible for a priori intuition to be utterly misleading or even unreliable. From the factthat it is possible that we gowrong about some a priori propositions, it doesn’t at all followthat we might go wrong about all of them or, indeed, about any particular individualone. Put another way, admitting the fallibility of a faculty or source of evidence does notnecessarily involve admitting that the faculty might go wrong on any and all particularjudgments. So, the considerations advanced by the Reidians do not suffice to generate askeptical problem for a priori intuition like that which arises for perception.Moreover, not only does the possibility of unreliability not follow from the fact of

actual error, there is, in the those judgments of reason which are strongest, decisivereason for thinking that reason cannot be entirely erroneous. When we consider suchpropositions as the simplest truths of arithmetic and logic, we find them maximallyintuitively obvious. They strike as so clear, that we cannot even imagine how we mightbe wrong about them (Nagel 1997). To suggest, as the Reidians do, that reason is apiece of false ware is to suggest that even those a priori intuitions in which thosemaximally evident propositions are grasped might be mistaken. However, one sees thetruth so clearly in such cases that no credibility attaches to the Reidian’s assertion thatone’s belief in such propositions is rendered doubtful by the possibility that one is thevictim of some evil demon or neurosurgeon. While such skeptical scenarios suffice togenerate serious skeptical worries regarding the external world, they do nothing at allto cast doubt upon the clearest truths of reason.The point I am endeavoring to make here is much like one touched upon by

Descartes at the beginning of the Third Meditation in discussing the hyperbolic doubtintroduced by God’s omnipotence. In spite of his earlier claim that God (or somepowerful demon) might cause him to be deceived about that which is most evident,Descartes writes:

Yet when I turn to the things themselves which I think I perceive very clearly, I am so convincedby them that I spontaneously declare: let whoever can do so deceive me, he will never bring itabout that I am nothing, so long as I continue to think that I am something; or make it true at


some future time that I have never existed, since it is now true that I exist; or bring it about thattwo and three added together are more or less than five, or anything of this kind in which I see amanifest contradiction. (1996, 25)

Descartes here admits that he finds himself unable to even entertain the possibility thatcertain propositions are false while considering attentively their content. Likewise, mysuggestion is that the clearest truths of reason immediately and evidently falsify anyskeptical hypothesis alleged to cast doubt upon them.

Following Lex Newman and Alan Nelson’s (1999) insightful interpretation of thispart of the Cartesian corpus, we may distinguish between direct and indirect doubt.A doubt that p is direct when it arises while one is carefully attending to p. As Newmanand Nelson point out, it may seem that propositions which cannot be directly doubtedsimply cannot be doubted since any turning of one’s attention away from the propos-ition at issue is, ipso facto, failing to make the proposition an object of one’s consider-ation and thus failing to doubt the truth of that proposition. This would, they note,seem to make skeptical worries about the clearest truths of reason simply impossible toformulate, just as I’m suggesting.

Newman and Nelson claim, however, that Descartes’ deepest doubt is a “meta-cognitive” doubt which “proceeds precisely by means of an attention shift away fromthe item in question and towards a (second-order) subsuming rubric referring tosimilarly grounded items” (1999, 375). So, they claim, one cannot doubt the clearesttruths of reason while one attends to them, but one can nonetheless use such anindirect means to undermine all such truths by attending to the general rubric whichsubsumes them. An indirect doubt, then, is a doubt which occurs only when one is notdirectly considering p. They conclude, “in thus rendering doubtful even”my cognitivebest, the meta-cognitive doubt “ipso facto renders doubtful—albeit indirectly—eachparticular item subsumed under the rubric by undermining its ground” (375).

It is worth noting that the tendency of Reid and the Neo-Reidians to refer to thefaculties of introspection, sympathy, perception, etc., has precisely the effect of allowingthe kind of attention shift required for a meta-cognitive doubt. Reid’s discussion in thepassage quoted earlier does exactly this in asking us to reflect on the possibility that our“reason” is a piece of “false ware.”11 Indeed, when it comes to the a priori, even thetypical rationalist’s focus on the type of propositional attitude common to all a prioriintuitions invites a similar attention shift. Having turned one’s attention to such ageneral subsuming rubric, rather than asking about the epistemic status of a particularattitude with a particular content, one asks about the status of the faculty producing theattitude or the status of the attitude type and thereby turns one’s direct attention awayfrom the proposition which is the content of the particular mental state in question.

Though Newman and Nelson have much of interest to say regarding the properinterpretation and evaluation of Descartes’ attempt to conquer such a meta-cognitivedoubt, we should ask if such indirect doubt has the suggested skeptical significance inthe first place. Is it, as Newman and Nelson claim, true that meta-cognitive doubt

216 JOEL PUST

really renders doubtful the items subsumed under a rubric such as “the clearestjudgments of reason”? If so, what kind of doubt is cast upon those items? Recall thatit is already conceded that I cannot doubt 2 + 2 = 4 while attending directly to theproposition. However, every time I get myself to doubt that “the clearest judgments ofreason are true,” have I really come to doubt that 2 + 2 = 4 is true? It seems to me thatI have not. In one sense, of course, I can doubt of the proposition that 2 + 2 = 4 that it istrue, by means of various referring descriptions such as “the proposition I thoughtabout yesterday morning” or, as Newman and Nelson prefer, “my epistemic best.”However, the question which we must ask is whether this indirect doubt is epistemic-ally significant and constitutes any reason at all to question the truth of the propositionand my epistemic grasp upon it.To aid in seeing that it is not and does not, perhaps an analogy will help. Suppose

I am told by someone quite trustworthy that they have found a book containing a longlist of propositions. It seems, of course, quite possible that all of the propositions writtenin the book are false. Suppose further that many of the propositions contained in thebook are propositions which I believe. It would not seem to follow that I have, indoubting that any proposition in the book is true, thereby cast epistemically significantdoubt upon the truth of any of my beliefs. This means of considering and referring tothe propositions in question fails to put me in a position relevant to their epistemicassessment. Just so, it seems to me, indirect doubt fails to engage a proposition in theright manner to generate skeptical problems.Even if, contrary to my suggestion, such indirect doubts do produce some sort of

genuine epistemic difficulty, the difficulty is a strangely insubstantial and evanescentone, as it is always entirely and wholly defeated by direct attention to the propositionsin question. After all, as Descartes forcefully noted, whatever doubts I may indirectlygenerate about the clearest propositions of reason, attention to such a particularproposition immediately dispels the doubt. I would add that it does so in a waywhich involves my seeing that such an indirect doubt was ill-founded, predicated ona failure to fully understand what it actually encompassed. Returning to the analogyabove, if one thought some genuine doubt was produced by my allowing that all thepropositions written in the book are false, such doubt is shown utterly mistaken by myfinding the clearest truths of reason in the book. Indeed, it seems that direct reflectionon that fact justifies me in thinking that I was simply mistaken in thinking that myepistemic best might be mistaken.In fact, it seems clear that the propositions which are the contents of the judgments

in question ought to take precedence here. This is to say that indirect reference to aproposition ought not be allowed to undermine whatever epistemic credentials accrueto a proposition thought in propia persona. When I am able to make clear sense of theskeptical possibility that some “faculty” of mine produces in me false beliefs of a certainnumber, it can only be because I can make sense of the idea that the content of theactual judgments it produces in me are false. This, however, requires attention to thecontent of the judgments and to the ways in which they may be false. Indirect means of


doubt are, I suggest, illegitimate, as they fail to genuinely engage the content of thejudgments with which one is concerned. For these reasons, then, I can’t see that thesort of move which Newman and Nelson attribute to Descartes is ultimately successfulin rendering coherent the possibility that my reason is entirely mistaken or constitutes,as Reid imagines, “a piece of false ware.” The Reidians are, I suggest, simply mistakenin holding that my reason might be entirely unreliable.

It may be replied that I have simply failed to appreciate the lesson taught by cases ofextremely intuitive but false propositions, such as the naıve comprehension axiom ofthe Russell Paradox. The neo-Reidians might be taken to be suggesting that the lessonof such cases is that even that which is most evident can be mistaken. However, thenaıve comprehension axiom is not as clearly true as the propositions used to justify itsrejection. That is to say, we are rational in thinking the comprehension axiom falseonly if there are other conflicting propositions (including non-contradiction) whichare more clearly true. Hence, we’ve reason to think that the naıve comprehensionaxiom is not among the most evident propositions and hence no reason to think thatthose most evident propositions of reason could be false.

If I am right, the fact that reason is infallible regarding certain basic truths can be usedto turn aside the claim that reason deserves no special regard. There is nothingepistemically arbitrary in treating as foundationally justified only those propositionsabout which one cannot be mistaken and which reveal to one that they cannot bemistaken. This is sufficient to show that there is nothing at all arbitrary in seeking tofind a reason (even a probabilistic one) for all of one’s beliefs grounded in propositionswith the requisite status. Hence, there is nothing epistemically arbitrary in a version ofthe Cartesian Perspective which limits itself to the most evident truths of reason andconsciousness. Indeed, I’ll show in the next section that Reid, in various asides,acknowledges such propositions have a special epistemic status.

To the forgoing, it may be rejoined that securing the most basic propositionspresented by reason (and introspection) from skeptical attack might be sufficient toreject a charge of arbitrary partiality, but it is not enough to show that the CartesianPerspective is correct. NJP, after all, makes general reference to the deliverances ofrational intuition and introspection rather than to some special subclass of such deliver-ances. Even if there is nothing arbitrary in a standard of perfection for full or completejustification, what of lesser degrees of justification? What makes a verdict of a prioriintuition which is less than certain, more justified than a belief that one has hands? Theanswer, I suggest, is that a priori intuition is, at least when limited to intuitions ofsufficient (though less than maximal) strength, necessarily reliable (Pust 2004). Such averdict is intuitively plausible. At the very least, there is no reason to regard rationalintuition as possibly unreliable, because (a) as noted earlier, possible general unreliabil-ity does not follow from the fact of known error, and (b) we lack the intuitionsrequisite for directly justifying such a claim of possible general unreliability. However,we can see, in light of skeptical scenarios, that all of our perceptual beliefs could bemistaken, consistent with our perceptual evidence.

218 JOEL PUST

The Reidians’ central charge of arbitrariness is predicated on the claim that cogentskeptical worries of the sort which afflict perceptual, inductive, testimonial, and otherordinary beliefs also afflict a priori intuition and consciousness. I have argued that this isnot so. Of course, such claims are themselves justified by a priori intuitions (or a lackthereof ) and so such a procedure might be alleged, in the present context, to bequestion-begging. It is not. The skeptic at issue is one who provides us with reasons orarguments for her views. She claims that we can see, by our own lights, that ourperceptual judgments might all be mistaken, consistent with the current and pastcourse of our experiences. So, if we lack the requisite intuitions, then it is not truethat by our own lights our a priori intuitions might be generally erroneous and theskeptic has no rational purchase upon us. The point is not that the skeptic has somespecial burden of proof, but rather that skeptical arguments which are to bring usrationally to skeptical conclusions must be cogent by our lights.

7 Reid’s Inconsistent Concessions to the CartesianPerspective

Curiously, Reid’s work contains a number of passages which reveal an acquiescenceto the Cartesian Perspective inconsistent with his charges against it.12 The followingpassages illustrate, in language reminiscent of the passage from Descartes’ ThirdMeditation cited earlier, Reid’s apparent concessions that there is something appropri-ate about the distinction which the Cartesian Perspective draws between our faculties:

When I compare the different kinds of evidence above-mentioned, I confess, after all, that theevidence of reasoning, and that of some necessary and self-evident truths, seems to be the leastmysterious and the most perfectly comprehended; and therefore, I do not think it strange thatphilosophers should have endeavored to reduce all kinds of evidence to these. . . . When I see aproposition to be self-evident and necessary, and that the subject is plainly included in thepredicate, there seems to be nothing more that I can desire in order to understand why I believeit . . . . The light of truth so fills my mind, in these cases, that I can neither conceive nor desireanything more satisfying. (1967, 330)

When I believe the truth of a mathematical axiom, or of a mathematical proposition, I see that itmust be so: every manwho has the same conception of it sees the same. There is a necessary and anevident connection between the subject and the predicate of the proposition; and I have all theevidence to support my belief which I can possibly conceive. When I believe that I washed myhands and face this morning, there appears no necessity in the truth of this proposition. It might be,or it might not be. A man may distinctly conceive it without believing it at all. (1967, 341)

and, most interestingly,

It is, no doubt, the perfection of a rational being to have no belief but what is grounded onintuitive evidence, or on just reasoning; but man, I apprehend, is not such a being; nor is it theintention of nature that he should be such a being, in every period of his existence. (1967, 332)


Among the Neo-Reidians, at least Wolterstorff appears to agree, holding that “wewould be more admirable than we are if our beliefs”met the strictures of the CartesianPerspective (2001, 194).

It is, however, inconsistent to claim that the demands of the Cartesian Perspectiveare epistemically arbitrary, which, as we have seen, the Reidians do, while at the sametime admitting that there is something epistemically superior about having beliefsgrounded in reason as the Cartesian Perspective demands—indeed, while concedingthat at least sometimes the evidence of reason (and consciousness) is the best one couldpossibly imagine and constitutes epistemic perfection.13 Surely, if one agrees that itwould be a very good thing, perhaps allowing us to attain “the perfection of a rationalbeing,” if our beliefs met a certain condition, then one should not hold that requiringthat condition of them is arbitrary.

Put another way, the following are inconsistent: (1) the philosopher, if she couldshow that our beliefs can be grounded in reason and consciousness, would show ourepistemic condition to be the best possible and, (2) the philosopher’s attempt, even ifsuccessful, would have no value, that, to use Wolterstorff ’s phrase, “nothing is gained”by success at such a project. Yet it seems that Reid, and among the Neo-Reidians atleast Wolterstorff, are committed to both.14 What could be arbitrary about seeking thekind of evidence which one “can best comprehend, and which gives perfect satisfactionto an inquisitive mind” (1967, 330)? The Cartesian Perspective seeks just such best kindof evidence that our beliefs are at least probable. Even if, as CS holds, such evidence isnot to be had, its pursuit is not arbitrary, as Reid’s rare, but presumably honest,concessions indicate.

8 The Arbitrary Partiality of ReidianismThe Reidians claim that the Cartesian Perspective is arbitrarily partial in its selection offoundational epistemic resources. I’ve argued that, at least with respect to a prioriintuition, this charge is mistaken because a priori intuition is immune to the standardsorts of skeptical challenges. As I have noted, the Reidians hold not only that ourordinary beliefs are none the worse for failing the demands of the Cartesian Perspec-tive, but also hold that such ordinary beliefs are very often justified. In this section, I’llargue that the Reidians cannot justifiably maintain AS consistent with their endorse-ment of CS. In other words, so long as the Reidians maintain that the CartesianPerspective leads to broad skepticism, their own anti-skepticism must be epistemicallyarbitrary in its features. This suggests that Cartesian Perspective is, in the end, unavoid-able for a properly justified epistemology.

To being with, it should be noted that the Reidians are, either explicitly byproclamation (Plantinga 1983, 75–8; Bergmann 2006, 210–11) or implicitly by theirinsistence that many of our actual beliefs are clearly justified, particularists. Accordingto particularism, epistemological theories are epistemologically justified to the extent towhich they properly categorize obvious cases of justified and unjustified belief.

220 JOEL PUST

Though this often goes unnoticed, particularism is only plausible if we are justified inbelieving that the relevant hypothetical or actual beliefs are justified or unjustified. So, acrucial issue for the particularist, and hence the Reidian particularist, is how it is that weare justified in believing that particular beliefs in particular circumstances are justified orunjustified.15

Once the issue is raised, it seems clear that the only plausible position is one holdingthat our beliefs about the justificational status of beliefs are justified by a priori intuitions.After all, what else could justify the claim that some actual or hypothetical belief isjustified? However, such a priori intuitions are one of the two sources of justificationprivileged by NFP. Hence it seems impossible that the Reidian can reflectively andconsistently believe that various of our ordinary beliefs are clearly justified whilesimultaneously maintaining that such beliefs lack support of the sort required byNFP. The Reidians appear to have mistakenly assumed that the only support whichcould be provided for a contingent claim, consistent with NFP, would be supportconsisting of an a priori cogent argument from the content of introspective judgmentsto the content of other contingent claims. However, if one is indeed justified by a priorirational intuition in believing that a particular contingent belief is justified in certainpossible circumstances, then one is justified by a priori rational intuition in believing thatsuch a belief is probably true in such circumstances. Moreover, if one is introspectivelyjustified in believing that one actually is in such circumstances, then one is justified,consistent with NFP, in believing that one’s own contingent belief is sufficientlyprobable.I am not maintaining that one is indeed a priori justified in believing the various

claims the Reidian particularist would require for the justification for her doctrine. Noram I maintaining that if one is so justified, that such justification is indefeasible. Onemight think that skeptical arguments of various sorts proceed from premises moresecure than such a priori intuitions regarding the justification of non-introspectivebeliefs with contingent content. After all, having seen that the content of the belief issome contingent proposition about the external world and that the belief could bemistakenly held on exactly the experiential grounds one has, how plausible is it thatone is justified in believing it true in the absence of an argument for the probable truthof its content from the resources of NFP? Rather, my claim is that such a prioriintuitions must prima facie justify claims about particular hypothetical beliefs if one isto be justified in appealing to such claims as the justificatory basis of one’s positiveepistemological views.So, the Reidians cannot consistently accept CS, AC, and AS unless they lack

intuitive justification for AS. Of course, if that is so, then they cannot regard thepositive epistemological position resulting from inductivist particularism as justified.Rather, the reflective Reidians must regard such a position as the product of their own(ultimately) arbitrary and unjustified beliefs regarding the extent of their justifiedbeliefs.16 This result makes clear that the various Reidian injunctions to ignoreskeptical concerns and to “simply go along with our natural reactions of trust” (Alston


1983, 119), to accept our beliefs as “the gift of Heaven” (Reid 1967, 330), to “joinwith the vulgar in taking for granted the fundamental reliability of [our] intellectualfaculties” (Wolterstorff 2001, 216), and to see that what we “properly take to berational . . . depends on what sort of metaphysical and religious stance [we] adopt”(Plantinga 2000, 190), are simply an expression of an arbitrary partiality toward someexisting beliefs (including, in the hands of the Reformed epistemologists, sectarianreligious beliefs). They are, in the end, simply an evasion of epistemology.

9 ConclusionThe general issue raised by the Reidians—whether a non-arbitrary account can begiven for thinking some putative evidential source(s) foundational while others requireindependent justification—is an absolutely crucial one for any epistemological projectwhich aspires to treat anything like the traditional problems. There is, however, arationale for rejecting the Reidian attack on the Cartesian Perspective’s privileging of apriori intuition and introspection. Finally, the fact that the positive Reidian accountmust be ultimately arbitrary suggests that the Cartesian Perspective must be correct ifthe epistemological enterprise is to have the universal character to which it aspires.

Notes1. My focus throughout is on propositional rather than doxastic justification—on whether we

have justification for our beliefs rather than on whether our beliefs are properly based on sucha justification.

2. Among contemporary epistemologists, Richard Fumerton (1995; 2001) and Laurence Bon-jour (1998; 2001; 2002) stand out as two of the clearest proponents of the CartesianPerspective.

3. Bergmann (2006) is another, more recent, neo-Reidian.4. Note that the two principles which constitute the Cartesian Perspective do not themselves state

any skeptical thesis. A defense of CS requires argument involving either a demonstration thatthese principles entail skepticism or, more plausibly, an inductive case on the basis of thefailures of various attempts to respond to the skeptical problems from within the strictures ofthe Cartesian Perspective.

5. This is certainly Plantinga’s view (1983; 1993a; 1993b; 2000). In spite of his rejection of“foundationalism” in his (1976), Wolterstorff might be better seen as rejecting only a certainsort of foundationalism with a restrictive conception of the foundationally justified propos-itions. Alston (1993) prefers, as befits his focus on practical rationality, to speak of our belief-forming “practices” which are demarcated in roughly the traditional ways.

6. It should be noted that the Neo-Reidian’s occasionally write as though we need not befoundationally justified in believing an inferential principle which might correctly describe ourinferences, preferring instead to maintain that an inferential disposition which accords with asuitable inferential principle produces justified beliefs from other foundationally justifiedbeliefs.

222 JOEL PUST

7. While some of the elements of Reformed epistemology were present in Plantinga’s“partners in guilt” argument in God and Other Minds (1967) and in Wolterstorff ’s Reasonwithin the Limits of Religion (1976), it is most clearly identified with the essays in Plantinga’sand Wolterstorff ’s edited book Faith and Rationality (1983).

8. See Alston’s admission that in focusing on what he calls the “Christian mystical practice”he “departs from Reid, who restricted himself to universal practices” (1991, 169). SeePlantinga’s claim that he aims to go beyond Reid in restoring the “sensus divinitatus” tothe basic sources of belief (1993a, 86). See also Wolterstorff ’s (1999; 2001).

9. See also (1983, 59), (1993a, 85), (1993b, 97), and (2000, 221).10. Exactly the same move is made in Bergmann’s work (2006, 222) when he suggests that all of

a person’s introspective beliefs and a priori beliefs might be mistaken on the basis of the factthat an introspective or a priori belief can be or is mistaken.

11. Notice that Reid’s problematic concessions to the Cartesian Perspective (see Section 7)result precisely when he directly considers the clearest truths of reason and introspectionrather than the relevant “faculties.”

12. Plantinga is the most trenchant critic of the demands of the Cartesian Perspective. Thoughhe was once willing (1993a, 102–3; 1993b, 107) to grant some epistemic superiority ofreason and consciousness, he has since indicated no such conciliatory stance. Indeed, inresponse to Bonjour’s criticism from the Cartesian Perspective, Plantinga writes, “Bonjourcomplains that on my accounts of induction and our knowledge of other minds, we just findourselves believing, under certain circumstances, that the sun will rise tomorrow or that Sallyis angry, without any real insight into how it is that the sun’s having risen lo! these many daysmakes it likely that it will rise tomorrow. But are we any better off in the a priori case? Whenwe contemplate the corresponding conditional of modus ponens, we just find ourselves withthis powerful inclination to believe that this proposition is true, and indeed couldn’t befalse . . . . We really don’t have any reasons or grounds for this belief; we simply, so to say,start with it” (1996, 342, emphasis added).

13. Notice that Reid similarly grants an epistemic superiority to consciousness, the other facultyprivileged by the Cartesian Perspective:

It is impossible that there can be any fallacy in sensation: for we are conscious of all oursensations, and they can neither be any other in their nature, nor greater or less in theirdegree than we feel them. It is impossible that a man should be in pain, when he does notfeel pain; and when he feels pain, it is impossible that his pain should not be real, and in itsdegree what it is felt to be; and the same thing may be said of every sensation whatsoever.(1967, 335)

For from this source of consciousness is derived all that we know, of the structure and of thepowers of our own minds; from which we may conclude, that there is no branch ofknowledge that stands upon a firmer foundation; for surely no kind of evidence can gobeyond that of consciousness. (1967, 443)

14. Again, Plantinga demurs, apparently holding that the only epistemic value which wouldaccrue to us in virtue of finding a rationally compelling argument from sensory states to theexternal world is that of the true belief that there is such an argument.

15. Bergmann’s defense of a Reidian position is, to his credit, sensitive to existence of thequestion of how the data beliefs for the particularist procedure are justified (2006, 209–11).


16. This result seems to me another expression of the worry undergirding “the Great PumpkinObjection” to Plantinga’s defense of Reformed epistemology (1983, 65). That objectionconsists of the suggestion that believers in the annual return of the Great Pumpkin mightequally use Plantinga’s defense of the rationality of theistic belief to justify their own bizarrebeliefs. (For a very early discussion of a similar concern, see Plantinga 1967, 268–70). Inresponse, Plantinga claimed that the Christian will “of course” conclude that he has founda-tional justification for believing as he does while the proponent of the Great Pumpkin does not.This response neglects to address what justification, if any, the Christian would have for such aconclusion and so invites the objection that the proponent of some alternative view mightsimply begin with the assumption that her beliefs are paradigms of justified beliefs and thenutilize the particularist methodology to “justify” an epistemological principle which deemsthem justified.

ReferencesAlston, W. (1983). “Christian Experience and Christian Belief.” In Alvin Plantinga and Nicholas

Wolterstorff (eds), Faith and Rationality. Notre Dame, IN: University of Notre Dame Press.Alston, W. (1985). “Thomas Reid on Epistemic Principles.” History of Philosophy Quarterly 2:

435–52.Alston, W. (1991). Perceiving God. Ithaca, NY: Cornell University Press.Alston, W. (1993). The Reliability of Sense Perception. Ithaca, NY: Cornell University Press.Bergmann, M. (2006). Justification without Awareness. New York: Oxford University Press.Bonjour, L. (1996). “Plantinga on Knowledge and Proper Function.” In J. Kvanvig (ed.),Warrant in Contemporary Epistemology. Lanham, MD: Rowman and Littlefield.

Bonjour, L. (1998). In Defense of Pure Reason. New York: Cambridge University Press.Bonjour, L. (2001). “Toward a Defense of Empirical Foundationalism.” In Michael DePaul

(ed.), Resurrecting Old-Fashioned Foundationalism. Lanham, MD: Rowman and Littlefield.Bonjour, L. (2002). Epistemology. Lanham, MD: Rowman & Littlefield.Descartes, R. (1996). Meditations on First Philosophy. John Cottingham (ed.). Cambridge: Cam-

bridge University Press.Fumerton, R. (1995). Metaepistemology and Skepticism. Lanham, MD: Rowman & Littlefield.Fumerton, R. (2001). “Classical Foundationalism.” In Michael DePaul (ed.), Resurrecting Old-Fashioned Foundationalism. Lanham, MD: Rowman and Littlefield.

Nagel, T. (1997). The Last Word. New York: Oxford University Press.Newman, L. and A. Nelson. (1999). “Circumventing Cartesian Circles.” Noûs 33: 370–404.Plantinga, A. (1967). God and Other Minds. Ithaca, NY: Cornell University Press.Plantinga, A. (1983). “Reason and Belief in God.” In A. Plantinga and N. Wolterstorff (eds),

Faith and Rationality. Notre Dame, IN: University of Notre Dame Press.Plantinga, A. (1993a). Warrant: The Current Debate. New York: Oxford University Press.Plantinga, A. (1993b). Warrant and Proper Function. New York: Oxford University Press.Plantinga, A. (1996). “Respondeo.” In J. Kvanvig (ed.), Warrant in Contemporary Epistemology.

Lanham, MD: Rowman and Littlefield.Plantinga, A. (2000). Warranted Christian Belief. New York: Oxford University Press.Pust, J. (2004). “On Explaining Knowledge of Necessity.” Dialectica 58: 71–87.Reid, T. (1967). Philosophical Works, 8th edition, 2 vols. Hildeshiem: Georg Olms.

224 JOEL PUST

Wolterstorff, N. (1976). Reason Within the Bounds of Religion. Grand Rapids, MI: Eerdmans.Wolterstorff, N. (1996). John Locke and the Ethics of Belief. Cambridge: Cambridge University Press.Wolterstorff, N. (1999). “Epistemology of Religion.” In John Greco and Ernest Sosa (eds), TheBlackwell Guide to Epistemology. Malden, MA: Blackwell.

Wolterstorff, N. (2001). Thomas Reid and the Story of Epistemology. New York: CambridgeUniversity Press.


10

A Priori Bootstrapping

Ralph Wedgwood

In this chapter, I shall explore the problems that are raised by a certain traditionalsceptical paradox. My conclusion, at the end of this chapter, will be that the mostchallenging problem raised by this paradox does not primarily concern the justificationof beliefs; it concerns the justification of belief-forming processes. I shall argue for thisconclusion by showing that if we can solve the sceptical problem for belief-formingprocesses, then it will be a relatively straightforward matter to solve the problem thatconcerns the justification of beliefs.

In the first section of the chapter, I shall set out the problem that this scepticalparadox raises for the justification of your beliefs. In the second section, I shall presentsome reasons for thinking that any adequate solution to this problem must imply thatyou have a priori justification for believing that you are not in a sceptical scenario (thatis, a situation in which your sensory experiences are in some undetectable wayunreliable).

In the third section, I shall try to make it plausible that you do indeed have sucha priori justification, by arguing that there is at least one possible process of non-empirical reasoning—what I call the a priori bootstrapping reasoning—that can leadyou to a justified belief in the proposition that you are not in such a sceptical scenario.As I shall explain in the fourth section, however, there is absolutely no prospect thatthis argument will be able to solve the problem that the sceptical paradox raises for thejustification of belief-forming processes: that deeper problem will have to be solved insome other way. Finally, I shall close by commenting on the significance of myarguments for the idea of a priori justification, and for the attempts that other philoso-phers have made at solving the problems that are raised by the sceptical paradox.

1 The Sceptical ParadoxThe sceptical paradox that I have in mind has been perhaps most clearly stated in recentyears by James Pryor (2000) and BrianWeatherson (2005).1 In this section, I shall give asomewhat rough and slipshod statement of this argument; for our present purposes, thisrough formulation of the argument will suffice.

Consider the hypothesis that, in some way that you cannot possibly detect, yoursensory experiences are radically unreliable guides to the truth about your environ-ment. (We could fill in the details of this hypothesis in various different ways. Forexample, perhaps you are dreaming an extraordinarily lucid dream; or perhaps you arethe victim of a Cartesian evil demon; or perhaps you are in the “Matrix”; or. . . . Forour purposes, it does not matter exactly what the details of this hypothesis are, so longas the hypothesis implies that in some undetectable way, your sensory experiences areradically unreliable guides to the truth about your environment.) Let us call thishypothesis the sceptical hypothesis, or s for short.Now, are you justified in believing ‘¬s’, the negation of the sceptical hypothesis?

Clearly, ‘¬s’ is a contingent proposition about the situation that you—a particularindividual—are in, at a particular point in time. So, surely, the only kind of justificationthat you could have for believing ‘¬s’ is empirical. Ultimately, if you are justified inbelieving ‘¬s’ at all, you are justified in believing it at least partly on the basis of yoursensory experiences.There are, however, two reasons for thinking that your sensory experiences could

not possibly justify you in believing ‘¬s’, the negation of the sceptical hypothesis. First,it may seem to be circular or question-begging to rely on any of your sensory experiencesin believing that your sensory experiences are not undetectably unreliable. Forexample, suppose that you were serving on a jury, and during the jury’s deliberations,a question was raised about the trustworthiness of a certain witness. Surely it would beabsurdly circular or question-begging to brush this question aside by pointing out thatthe witness himself asserted in his testimony that he was entirely trustworthy. And howcan it be any more rational to rely on your sensory experiences in believing that yoursensory experiences are reliable than it would be to rely on the testimony of a witness inbelieving that the witness is reliable?Secondly, it may also seem that your sensory experiences would not disconfirm the

sceptical hypothesis anyway (that is, they would not confirm the negation of ‘¬s’). Thisis because the sensory experiences that you are currently having are of precisely thekind that would be predicted by the sceptical hypothesis. For both of these two reasons,then, it seems that your sensory experiences cannot possibly justify you in believing‘¬s’. If it is also true that you cannot be justified in believing ‘¬s’ at all unless yoursensory experiences can justify you in believing it, it follows that you cannot bejustified in believing ‘¬s’ at all.Suppose that this conclusion is true. That is, suppose that you are not in any way

justified in rejecting the hypothesis that your sensory experiences are undetectablyunreliable. In that case, it seems that you cannot really be justified in believing anythingon the basis of your sensory experiences.This last point can be bolstered by considering some analogies. For instance, let us

return to the example of the jury’s deliberations about a certain witness’s testimony. Ifyou are not in any way justified in regarding the witness as trustworthy, surely youcannot be justified in believing something solely because the witness has asserted it.

A PRIORI BOOTSTRAPPING 227

Similarly, it seems that if you are not in any way justified in believing that a certainmeasuring instrument is reliable, you are surely not justified in believing anythingsolely on the basis of trusting that measuring instrument.

If you could be justified in believing a proposition p on the basis of your sensoryexperiences even if you were not in any way justified in believing your sensoryexperiences to be reliable, then there would be a radical asymmetry between (i) thejustificatory role of sensory experiences and (ii) the justificatory role of testimony andmeasuring instruments. But unless this radical asymmetry can be explained, it seemsimplausible. So it seems that if you are not justified in believing ‘¬s’, you are notjustified in believing anything on the basis of your sensory experiences at all.

As I have remarked, I shall not go to the trouble of giving a more rigorous andprecise statement of this argument here. The rough statement of the argument givenhere is enough for my purposes. This is because in this chapter, I shall argue against oneof the argument’s clearest premises—specifically, the premise that the only kind ofjustification that you could have for believing ‘¬s’ is empirical. That is, I shall argue thatyou have a kind of a priori justification for rejecting the sceptical hypothesis.

2 Our Justification for Rejecting the ScepticalHypothesis Must be A Priori

In fact, there is an argument, due to Roger White (2006), for the conclusion that ourjustification for believing the negations of at least some sceptical hypotheses must be apriori.

The argument rests on two fundamental epistemological assumptions. First, thedegree to which one is justified in believing the various propositions that one is capableof believing can be modelled by means of a probability function: that is, the degree towhich one is justified in believing any proposition p can be represented by theprobability of p according to this probability function. Secondly, the way in whichthe degrees to which one is justified in believing these propositions evolves over time isby responding to new evidence in accordance with classical Bayesian conditionalization.We can sum up these two assumptions by saying that they amount to a “probabilistic”and “evidentialist” view of justification.

Admittedly, this sort of probabilistic evidentialist view is highly controversial.2

Indeed, I do not accept every detail of this view myself.3 So I am not claiming herethat White’s argument is obviously sound. I shall just rehearse this argument herebecause it is one possible reason for thinking that our justification for believing theseanti-sceptical propositions is a priori. It seems plausible that evenwithin other epistemo-logical frameworks, a broadly analogous argument will be available for the conclusionthat our justification for rejecting at least some sceptical hypotheses must be a priori.

White’s argument starts out with the premise that there is at least one ordinaryproposition—let us call this ordinary proposition op—such that one is justified in

228 RALPH WEDGWOOD

believing op, even though one’s total evidence e does not entail the truth of op. (Clearly,if this premise is false, then a very radical form of scepticism follows immediately.) Wecan now construct a rather special sceptical hypothesis se: specifically, se is an extremelyspecific sceptical hypothesis—a hypothesis that actually entails that one has this evidencee, but is incompatible with the ordinary proposition op. (For example, suppose that op isthe proposition ‘I have hands’. Then se might be the proposition ‘I am a handlessthinker who is having the evidence e fed to me by a deceiving demon’. Or more simplystill, se might just be ‘e & ¬op’.)White’s crucial point is that this evidence e cannot lower the probability of this

sceptical hypothesis se. Indeed, presumably if e is genuine evidence at all, it cannot havebeen utterly certain, before one acquired the evidence e, that e would be true. Thus,the prior probability of ‘¬e’ must have been greater than 0; and if the prior probabilityof the sceptical hypothesis se was also greater than 0, then e must actually raise theprobability of se. We can illustrate this point by means of the following figure:

e

¬e

se

¬se

op

se = Sceptical hypothesise = Evidenceop = Ordinary proposition

se entails eop entails ¬se

In this figure, the space of possibilities is divided into three cells: the top cell, whereboth e (the evidence) and se (the sceptical hypothesis) are true; the middle cell, whereboth e (the evidence) and ‘¬se’ (the negation of the sceptical hypothesis) are true; andthe bottom cell, where neither the evidence nor the sceptical hypothesis is true. (Sincese entails e, there is no cell where se is true but e is not.) When one learns that e is true,the region of the space of possibilities where ‘¬e’ is true can be ruled out (this is why it isrepresented as shaded in the figure). But then se will occupy a larger proportion of theremaining possibility-space than it occupied of the earlier possibility-space: formerly seoccupied one cell out of three; now it occupies one cell out of two. If the relevant


probability distributions evolve in response to evidence according to Bayesian con-ditionalization, this means that the probability of se has gone up in response to one’slearning e, and the probability of its negation ‘¬se’ has gone down. Moreover, the priorprobability of ‘¬se’, before one acquired evidence e, must have been at least as high asthe new probability of op after receiving evidence e.Moreover, this argument does not rest on any special assumptions about the relevant

probability function. So the same point can be made about every probability function:none of these probability functions can allow that the probability of ‘¬se’ was raised bythe evidence e; and on every probability function, the prior unconditional probabilityof ‘¬se’ must have been at least as high as the conditional probability of op given theevidence e.

We may assume that this evidence e is one’s total evidence—the result of the accumu-lation of many pieces of evidence over time. So every earlier piece of evidence that onehas ever received consisted of some proposition e– that is entailed by e; and since se entailse, it must also entail e– as well. Thus, just as the probability of ‘¬se’ cannot have beenraised by e, given that se entails e, so too, for exactly the same reason, the probability of‘¬se’ also cannot have been raised by any of those past pieces of evidence e– either. Atevery stage as the probability of the relevant propositions evolved in response toevidence, the negation of the sceptical hypothesis ‘¬se’ must already have had aprobability that is at least as high as the conditional probability of op given e.

Assuming a probabilistic and evidentialist conception of justification, it follows thatone must always already have had a high degree of justification for believing ‘¬se’—indeed, this degree of justification must have been at least as high as the degree ofjustification that the ordinary proposition op has in the presence of e—in advance ofreceiving every piece of evidence that one has ever had. In this sense, one must have apriori justification for ‘¬se’. Moreover, if one has a priori justification for ‘¬se’ (thenegation of this very specific sceptical hypothesis), then as I shall now argue, it seemslikely that one also has a priori justification for the negation of other sceptical hypothesesas well.

As we have seen, e cannot raise the probability of op any higher than the priorprobability of ‘¬se’; and within the framework of a probabilistic and evidentialist viewof justification, this implies that one’s possession of a priori justification for ‘¬se’ is anecessary condition of one’s being justified in believing op when one’s total evidence ise. In general, the same point will also apply to any other body of evidence (like e) andany other ordinary proposition (like op). So we can draw the following generalconclusion: for any body of evidence and any ordinary proposition that is not logicallyentailed by that body of evidence, there is some corresponding sceptical hypothesis(like se) such that having a priori justification for rejecting this sceptical hypothesis is anecessary condition for one’s being justified in believing that ordinary proposition inthe presence of that evidence. In that sense, having a priori justification for rejectingsceptical hypotheses is a necessary condition for having any empirical justification forordinary propositions whatsoever.

230 RALPH WEDGWOOD

What of the more generic sceptical hypotheses (like s)? It seems prima facie plausiblethat the epistemology of the more generic sceptical hypotheses (like s) is broadly similarto the epistemology of these very specific sceptical hypotheses (like se). So it seemsplausible that our justification for ‘¬s’ must also be a priori.This general conclusion does not entail that every thinker who is empirically

justified in believing an ordinary proposition must actually believe ‘¬s’, or indeed thatshe must actually believe any anti-sceptical proposition (that is, any proposition that isincompatible with sceptical hypotheses like s) at all. It only entails that the thinker hassuch a priori justification for such anti-sceptical propositions available to her. In thejargon, it only entails that the thinker is propositionally justified in believing ‘¬s’, not thatshe is doxastically justified in so doing (in other words, it does not entail that shejustifiedly believes ‘¬s’).In general, however, there seems to be a fundamental connection between propos-

itional and doxastic justification. If one has propositional justification for believing aproposition p, then it seems that there must be a possible course of reasoning that onecould engage in, which would lead to one’s having a doxastically justified belief in p.Moreover, if—in addition—the propositional justification that one has for believing pis a priori, then there must be a possible course of reasoning that could lead to one’shaving a doxastically justified belief in p, which does not rely on any of one’s experi-ences or empirically justified beliefs. Instead, this possible course of reasoning mustsomehow be non-empirical: it must rely on one’s rational capacities alone. This raises thequestion: What is the possible course of non-empirical reasoning that could lead one tohave a doxastically justified belief in the negation of the sceptical hypothesis ‘¬s’? Thisis one of the questions that I shall answer in the next section.Even though I shall be arguing that there is indeed a non-empirical course of

reasoning that can lead us to believing ‘¬s’, this does not entail that there may notalso be empirical courses of reasoning that can lead us to the same conclusion. It ispossible for there to be both a non-empirical route and an empirical route that bothcount as rational ways of coming to believe the very same proposition.4 Still, as wehave seen, the existence of a priori propositional justification for the negation of thesceptical hypothesis is a necessary condition of any ordinary empirical justificationwhatsoever—whereas presumably if there is an empirical route that can lead to adoxastically justified belief in the negation of the sceptical hypothesis, this route willitself have to rest in some way on empirical justification for ordinary propositions. Sothe non-empirical route to acquiring doxastically justified beliefs in anti-scepticalpropositions is in a way more fundamental than any empirical route.5

Finally, even though a priori justification for anti-sceptical propositions is a necessarycondition for empirical justification for ordinary propositions, this does not entail thatone’s empirical justification for those ordinary propositions itself rests on or is in partconstituted by one’s a priori justification for any anti-sceptical propositions. Even if it isimpossible to have empirical justification for ordinary propositions without also havinga priori justification for these anti-sceptical propositions, it does not follow that this a


priori justification in any way explains or underlies this empirical justification.6 On thecontrary, it may be that the availability of a priori justification for anti-scepticalpropositions is simply a necessary consequence of the fact that sensory experiences justifyordinary propositions, and not any part of what constitutes our ordinary empiricaljustification. It is precisely this possibility that I shall exploit in my solution to thesceptical paradox.

In fact, the probabilistic argument given in this section already shows how theexistence of a priori propositional justification for ‘¬se’ is a necessary consequence ofthe logical relations between one’s evidence and the sceptical hypothesis se, along withthe fact that one is justified in believing the ordinary proposition op in the presence ofevidence e. In this way, this probabilistic argument can be taken to show how theexistence of this a priori propositional justification is explained by the probabilistic andevidentialist structure of justification, together with the fact that op is justified in thepresence of e. But this still leaves at least two questions. First, this probabilistic argumentis restricted to the extremely specific sceptical hypotheses, like se; we need to knowexactly how we are to extend the argument to the more generic sceptical hypotheses,like s. Secondly, we also need to know what kind of non-empirical reasoning couldpossibly lead us to a doxastically justified belief in ‘¬s’. I shall answer both of these twoquestions in the next section.

3 The A Priori Bootstrapping ReasoningThe goal of this section is to make it plausible that there is a kind of non-empiricalreasoning that could lead any thinker who is capable of going through this reasoning toa doxastically justified belief in ‘¬s’, the negation of the sceptical hypothesis. I shall infact describe two such processes of non-empirical reasoning here: I shall call the first thea priori bootstrapping reasoning, and the second the meta-justificatory reasoning.

First, we need to remind ourselves what exactly this proposition ‘¬s’ is. The scepticalhypothesis s, as I characterized it in Section 1, is the hypothesis that your sensoryexperiences are, in some undetectable way, unreliable guides to the truth. For short, letus say that it is the hypothesis that your experiences are undetectably unreliable. So thenegation of this hypothesis is the proposition that your experiences are either reliable orelse detectably unreliable. In other words, the negation of the sceptical hypothesis is theproposition that you can express by the following material conditional: ‘I cannot detectthat my sensory experiences are unreliable ! my sensory experiences are reliable’.It seems clear that you cannot have a priori justification for believing this material

conditional simply by having a priori justification for believing the negation of itsantecedent (that is, for believing that you can detect that your sensory experiences areunreliable), nor by having a priori justification for believing the consequent (that is, forbelieving that your experiences are reliable). Both the negation of the antecedent andthe consequent of this conditional are propositions for which the only availablejustification would have to be empirical, not a priori. Nonetheless, I shall argue in this

232 RALPH WEDGWOOD

section that you have a priori justification for this material conditional as a whole, byshowing how there are two possible processes of non-empirical reasoning that can leadyou to a doxastically justified belief in this conditional.Before describing these kinds of reasoning, I should note that my claims that this

reasoning is non-empirical, and that it can lead to a doxastically justified belief in thismaterial conditional, will be both based on a number of assumptions. These assump-tions seem plausible to me, but they are certainly not beyond question. Unfortunately,I shall not have time to offer a full defence of these assumptions here; I shall rely onthese assumptions purely for the sake of argument, in order to explore what theirconsequences will be.First, my claims will rely on the following assumption about what it is for a process of

reasoning to count as non-empirical. Suppose that there is a set of concepts and basicreasoning capacities such that it is necessary that any thinker who possesses thoseconcepts and has those capacities is in a position to engage in a certain process ofreasoning7 that would rationally lead the thinker to have a certain belief. Then we cansay that this process of reasoning is non-empirical, since it can rationally lead anythinker who possesses these concepts and reasoning capacities, regardless of whatexperiences or empirical background beliefs they may have, to have that belief. Inthis case, all thinkers of this sort have a priori justification for that belief. In this section,I shall sketch a process of reasoning that can lead us to believe ‘¬s’ (the negation of thesceptical hypothesis), which counts as non-empirical in this way.Secondly, my claim that the reasoning in question can lead to a doxastically justified

belief rests on a number of assumptions about which belief-forming processes arerational.8 Thus, one of these assumptions is that a certain belief-forming process thatwe might call “taking experience at face value” is necessarily rational. Very roughly,the process of taking experience at face value is the process that one engages in byresponding to the fact of one’s having a conscious experience that has the proposition pas part of its representational content by coming to believe p. For example, if mypresent experience has the proposition ‘I am holding my hands in front of my face’ aspart of its content, then I could engage in this process if I responded to this experienceby forming the belief that I am indeed holding my hands in front of my face.Now, this specification of the process is in one way obviously too rough. I could

have an experience of this kind even if it also appeared to me that a demon was talkingto me out of the palm of my hand, mocking me with the taunt that all my experiencesare complete illusions. In this case, it would clearly not be rational for me to form thebelief that I really am holding my hands in front of my face. If this process of “takingsensory experience at face value” is to be genuinely rational, it is not sufficient forengaging in this process that one just responds to any experience that has p as part of itscontent by forming the corresponding belief. It must also be the case that one’sexperiences, background beliefs, and other mental states do not contain any specialdefeating reasons of this kind. So a more precise name for this process would be the


“process of taking one’s experiences at face value, when such special defeating reasons areabsent”.9

If the general process of taking one’s experiences at face value is indeed necessarilyrational, then it is rational to form beliefs in this way whenever one’s experiences andbackground beliefs contain no special defeaters of the relevant sort. In other words, toadapt some terminology from Crispin Wright (1989, 251), the rationality of formingbeliefs by means of the process is “positive–presumptive”: the rationality of formingbeliefs in this way is the default position, which can be overturned only by the presenceof special defeaters.

This is admittedly still only a very rough description of the process of takingexperience at face value. In general, it will clearly be a difficult and controversialmatter to give a fully precise account of this process.10 I believe, however, that thisrough description will be sufficient for present purposes. At this point, it is moreimportant to see what the process of taking experience at face value does not involve.Specifically, as I understand it, engaging in this process does not involve relying on anyantecedent belief in the reliability of one’s sensory experiences (or indeed on any beliefabout one’s sensory experiences at all). Engaging in this process involves coming tobelieve p in direct response to one’s having an experience that has p as part of itsrepresentational content (together with the absence of defeating reasons of the relevantsort); it is not required that one should in addition have any belief about one’s experi-ences.

Moreover, I shall suggest here that the rationality of this process of taking one’ssensory experiences at face value (in the absence of defeaters of the relevant kind) is notitself explained, even in part, by one’s possession of justification for any higher-orderbeliefs about the reliability of one’s experiences.11 Nonetheless, as we shall see, even ifthe rationality of this process is not explained or constituted by one’s possession ofjustification for such higher-order beliefs, the rationality of this process may still entailthe existence of such justification: in this way, the existence of justification for suchhigher-order beliefs may be explained by the rationality of this process, instead of beingwhat explains it or constitutes it.

According to this assumption, then, it is always rational for you to engage in theprocess of taking your experiences at face value. At least so long as no special defeatersare present, this belief-forming process is a way for you to come to have doxasticallyjustified beliefs in the ordinary propositions that form part of the content of yoursensory experiences. It presumably follows from this that the fact that you have such anexperience and no special defeaters are present provides you with propositional justifica-tion for these ordinary propositions.

If the fact that you have such an experience and no special defeaters are presentprovides you with propositional justification in this way, then it seems that a certaincorresponding inference will also be rational. This inference starts from the premise thatyou have an experience as of p’s being the case and no special defeaters are present,and then infers from this premise—by means of a distinctive sort of non-deductive

234 RALPH WEDGWOOD

inference—that the proposition p is true. More precisely, this form of inferenceinvolves inferring a proposition p from the first-person present-tensed premise thatone might express by saying ‘I have an experience as of p’s being the case, and nospecial defeaters are present’. We could call this form of inference the rule of “external-world introduction”.It is hard to see how this form of inference could be any less rational than the non-

inferential process that I have called “taking experience at face value”; indeed, in asense this type of inference could be called the “inferential analogue” of that non-inferential process.12 I shall assume then the rule of “external-world introduction” isalso rational. Since the non-inferential process of taking sensory experience at facevalue is—as I have argued—necessarily rational for all thinkers who are capable ofengaging in this process (regardless of the specific sensory experiences they havehappened to have), I shall assume that this rule of inference is similarly necessarilyrational for all thinkers who are capable of both taking experience at face value and ofreasoning in accordance with this rule.In addition to this assumption about the rule of external-world introduction, I shall

also rely on a further assumption about a certain process for forming conditional beliefs.Suppose that one supposes one proposition p to be true, and then rationally infers asecond proposition q from this first proposition p. Now to “infer” q from p need notinvolve forming an unconditional belief in q itself: if one is merely supposing that p istrue, then one might not actually have an unconditional belief either in p or in q.Rather, to infer p from q is to conditionally accept q, conditionally on the supposition ofp; at least this account of inference is plausible if such “conditional acceptance” isunderstood along the lines that have been recommended by Dorothy Edgington(1995), not as the acceptance of an intrinsically conditional proposition, but as anintrinsically conditional attitude towards a pair of propositions <p, q>, involving theconditional acceptance of the second proposition q, under the supposition of the firstproposition p. According to the further assumption that I am relying on here, when-ever one rationally infers a proposition q from a proposition p in this way, it will also berational to form a belief in the corresponding material conditional ‘p ! q’.13 (Some-times, it may also be rational to respond to this inference by forming a belief in thecorresponding subjunctive or counterfactual conditional as well. But for our purposes, it isenough that this process can lead to a doxastically justified belief in this materialconditional.)Finally, I shall also rely on the assumption that there are two further belief-forming

processes that also count as necessarily rational. The first is the process of forming beliefsin the obvious logical consequences of propositions that one already rationally believes; andthe second is the process of forming beliefs by means of “inference to the bestexplanation”.I am now in a position to describe the a priori bootstrapping reasoning itself.14 The

reasoning begins with the following two crucial steps. In the first step, the thinker firstsupposes that she has an experience as of p’s being the case (that is, an experience that has

p as part of its representational content), and that her experiences, background beliefs,and other mental states contain no special defeaters; and then, from the supposition thatshe has an experience as of p’s being the case and no special defeaters are present, thethinker infers p itself, by means of the rule of external-world introduction. As I havesuggested, for the thinker to “infer” p from this supposition in this way is for thethinker to conditionally accept p, conditionally on the supposition that she has an experi-ence of this sort, and no special defeaters are present. (Because the inference is based onthe reasoner’s merely supposing that she has an experience as of p’s being the case, shedoes not need actually to have any such experience, or even to believe that she has anysuch experience, in order to draw this inference.) In the second step, the thinkerresponds to her having drawn this inference by coming to believe the correspondingmaterial conditional ‘[I have an experience as of p’s being the case & no special defeatersare present] ! p’.

Now, the reasoning consisting in these two steps seems clearly non-empirical—thatis, it is available to all reasoners who have the relevant concepts and reasoningcapacities, regardless of the specific experiences and empirical background beliefs thatthey have. Moreover, given the assumptions that I am relying on here, it is also anentirely rational process, which can lead to a doxastically justified belief in this materialconditional. So this supports the conclusion that all such thinkers have a priorijustification for believing this material conditional.

This material conditional, however, is already incompatible with the highly specificsceptical hypotheses (like the hypothesis that I labelled “se”) that I considered inSection 2. As I described them, each of these highly specific sceptical hypotheses hasthe following two features: (i) if one’s total evidence is e, then this specific scepticalhypothesis entails e; and (ii) this hypothesis is inconsistent with the ordinary propos-itions that one would normally believe on the basis of that evidence. Presumably, it ispart of one’s total evidence that one has an experience that has p as part of itsrepresentational content (and that no defeaters of the relevant kind are present); andwe may presumably also assume that p is an “ordinary proposition” of the relevant sort.So, this material conditional—‘[I have an experience as of p’s being the case & nospecial defeaters are present] ! p’—is incompatible with this highly specific scepticalhypothesis se. Thus, there is a process of non-empirical reasoning that can lead allthinkers who possess the relevant concepts and reasoning capacities to have a doxas-tically justified belief in the negation of this highly specific sceptical hypothesis.Moreover, as I shall now argue, there is a way of extending this non-empirical

reasoning so that it leads to the rejection of the more general sceptical hypotheses thatI considered in Section 1. The crucial point is that the thinker can perform the samemanoeuvre for a whole sequence of propositions p1, . . . , pn, as well as just for p. Thiswill lead her to form beliefs in a whole sequence of material conditionals, each ofwhich has the form ‘[I have an experience as of pi’s being the case & no special defeatersare present] ! pi’. Since she is capable of coming to believe the obvious logicalconsequences of propositions that she believes, she can then also form a belief in the

236 RALPH WEDGWOOD

conjunction of these conditionals: ‘[[I have an experience as of its being the case that p1 &no defeaters are present]! p1] & . . . [I have an experience as of its being the case that pn& no defeaters are present] ! pn]]’.Now this long conjunction seems to demand explanation. It seems plausible that the

best explanation of this long conjunction is that if no defeaters are present, then thethinker’s experiences are generally reliable. So it seems possible, by means of aninference to the best explanation, for the thinker to form a belief in the propositionthat if no defeaters are present, then her experiences are generally reliable.Presumably, if the thinker’s experiences contained a defeater of the relevant kind,

then those experiences would effectively reveal their own unreliability to the thinker,in which case it would be possible for the thinker to detect the unreliability of herexperiences. Thus, the proposition that if no defeaters are present, one’s experiencesare generally reliable effectively entails the further proposition that if one cannot detectthat one’s experiences are unreliable, they are reliable; but that further proposition isprecisely ‘¬s’, the negation of the sceptical hypothesis. Thus, this process ofreasoning—which I am here calling the a priori bootstrapping reasoning—can leadus to believing the negation of the sceptical hypothesis.For reasons that we have already explored, this whole process of reasoning seems to

be both non-empirical and rational. It is available to all thinkers who have the relevantconcepts and reasoning capacities, regardless of the experiences and empirical beliefsthat they may have; and it consists entirely of rational steps of reasoning. So theavailability of this a priori bootstrapping reasoning to every thinker who possesses therelevant concepts and capacities may provide the answer to the two questions that weraised at the end of Section 2.Moreover, there is also a second process of non-empirical reasoning that seems

capable of leading any thinker who possesses the relevant concepts and reasoningcapacities to rationally rejecting the sceptical hypotheses. In the rest of this section,I shall explore this second process of reasoning, which I shall call the “meta-justificatoryreasoning”.15

Imagine a Platonic soul waiting to be embodied—as it were, waiting to beam downfrom the intelligible world into the sensible world. The only information that thePlatonic soul has is purely a priori information. Suppose that this Platonic soul knowsall the principles of rational belief. (In this way, the claim that this “meta-justificatoryreasoning” is a non-empirical process of reasoning that can result in our rationallyrejecting the sceptical hypothesis relies on the assumption that all the relevant principlesof rational belief are a priori.) Since the Platonic soul knows all these principles ofrational belief, she can predict that as soon as she beams down into the sensible worldand starts having experiences, it will be rational for her to take her experiences at facevalue, and also to form introspective beliefs about the contents of her experience. Soshe knows that, if no special defeaters are present, it will be rational for her to believewhat Roger White (2006, 546) calls a “Track Record Proposition”—that is, someproposition of the form:


I have an experience as of its being the case that p1, and p1; . . . and I have an experience as of itsbeing the case that pn, and pn.

If it were rational to believe any such “Track Record Proposition”, it would surely alsobe rational to believe the following proposition, which White (2006, 546) calls “NoErrors”:

I have a great many experiences, the contents of which are all true.

If it were rational to believe “No Errors”, it would surely also be rational to believe thefollowing (“Reliability”) which seems the best and simplest explanation of “NoErrors”:

My experiences are generally reliable guides to the truth.

So, the Platonic soul already knows—even before she has beamed down to the sensibleworld—that if no defeaters are present among the experiences that she has after arrivingin the sensible world, it will be rational for her to believe that her experiences aregenerally reliable.

NowWhite (2006, 538) has defended a principle that he calls the “meta-justificationprinciple”. Roughly, this is the principle that if one knows that at a certain future time tit will be rational for one to believe a proposition p, and one also knows that one willnot lose any information between now and that future time t, then it is already rationalfor one to believe p. To be more precise, this principle has to be restricted to cases inwhich one has no special reason to think that one will encounter misleading evidencebetween now and t. It will admittedly be a challenging task to give a perfectly preciseformulation of this meta-justification principle. But it seems plausible that something atleast roughly like this principle is correct.16

If this meta-justification principle is correct, it seems plausible that another moregeneral principle is also correct. According to this more general principle, if one knowsthat if a certain condition C holds (and one will also not lose any information betweennow and t), it will be rational for one to believe p at t, then it is already rational for oneto believe that if condition C holds, p is true. (As with the simpler meta-justificationprinciple, to make this precise, we must restrict it to cases where one has no specialreason to think that if C holds, one will encounter any misleading evidence betweennow and t.)

Suppose that this generalized version of the meta-justification principle is correct.Then, given that the Platonic soul knows that if her experiences contain no defeaters, itwill be rational for her to believe that her experiences are generally reliable, it seemsthat it must already be rational for the Platonic soul to believe that if her experiencescontain no defeaters, her experiences are generally reliable. (The Platonic soul certainlyhas no special reason, before beaming down into the sensible world, to expect that ifher experiences contain no defeaters, her evidence will have been misleading in therelevant way.) Thus, it seems that it must already be rational for the Platonic soul to

238 RALPH WEDGWOOD

believe ‘¬s’, the negation of the sceptical hypothesis. Since the Platonic soul can gothrough this reasoning even before she beams down into the sensible world, it seemsthat her justification for believing ‘¬s’ must be a priori.It seems, then, that there are two processes of non-empirical reasoning—the a priori

bootstrapping reasoning and the meta-justificatory reasoning—that can lead anyrational thinker who has the relevant concepts and reasoning capacities to a doxasticallyjustified rejection of the sceptical hypothesis. This helps to make it plausible that we doindeed have a priori justification for rejecting such sceptical hypotheses.

4 The Significance of A Priori Bootstrappinga. In arguing that we have a priori justification for rejecting the sceptical hypothesis,I relied on the assumption that the process of taking experience at face value isnecessarily rational. I have not in any way explained why this assumption is true. Thispoint suggests that there is also a second problem that is raised by the sceptical paradox,different from the problem that I outlined in Section 1—specifically, a scepticalproblem about the rationality of belief-forming processes (such as the process of takingexperience at face value), rather than about the rationality or justification of beliefs.Indeed, it seems that the problem of explaining the rationality of belief-forming processes isin a way more fundamental than the problem of explaining why we are justified inholding particular beliefs.17

Unfortunately, the arguments given here cannot solve the more fundamental problemof how to explain the rationality of the process of taking experience at face value.These arguments can only solve the less fundamental problem of explaining how weare justified in believing ‘¬s’, the negation of the sceptical hypothesis. However, thesearguments clearly rely on the assumption that the process of taking experience at facevalue is rational. Since they rely on this assumption, they obviously cannot themselvesexplain why this assumption is true.18

As I have argued earlier, both of the two pieces of reasoning that I discussed in theprevious section—the a priori bootstrapping reasoning and the meta-justificatoryreasoning—are non-empirical. It follows that neither piece of reasoning actuallyinvolves engaging in the process of taking experience at face value. Nonetheless, thereis a sense in which both pieces of reasoning involve relying on that process. The a prioribootstrapping reasoning involves drawing inferences by means of the rule of “external-world introduction”, and I have suggested that this rule of inference is simply the“inferential analogue” of the process of taking experience at face value. The meta-justificatory reasoning involves drawing an inference from the explicit belief that theprocess of taking experience at face value is rational.In that sense, the availability of these rational pieces of non-empirical reasoning

cannot offer any “process-independent” justification of the process of taking experi-ence at face value. To use the terms that were made famous by Michael Dummett


(1975), these pieces of reasoning are “rule circular” even though they are not “premisecircular”. Indeed, it seems plausible that if the process of taking one’s sensory experi-ences at face value is rational, it is simply a primitively rational process—a process that isrational, but not because of the availability of any process-independent way of comingto form a rational belief in the process’s reliability.

I have called one of the two pieces of non-empirical reasoning that can lead anyrational thinker to believe the negation of the sceptical hypothesis the “a prioribootstrapping reasoning” because of its similarity to the “bootstrapping reasoning”that has been criticized by Jonathan Vogel (2000) and Stewart Cohen (2002). How-ever, to distinguish this reasoning from the reasoning that Vogel and Cohen havecriticized, I have called it the a priori bootstrapping reasoning; the reasoning that Vogeland Cohen have criticized could be called the empirical bootstrapping reasoning.In the empirical bootstrapping reasoning, the thinker reasons as follows (looking at

her hand): “I have a hand; it is part of the content of my current experience that I havea hand; so in this respect my experience has got things right!” That is, the thinker firstrelies on her actual current sensory experience to form a belief about the world, thenforms an introspective belief about the content of her current experience, and finallyconcludes from these two beliefs that her experience has got things right on thisoccasion.

Both Vogel and Cohen think that the empirical bootstrapping reasoning is obviouslyworthless. In fact, it is not clear that this reasoning really is entirely worthless. It couldhave happened, after all, that defeaters of some sort would have arisen. (Your currentexperience could have been an experience as of a demon speaking to you out of thepalm of your hand, taunting you with the claim that your experiences are all hallucin-ations.) If in fact no such defeaters are present, then the empirical bootstrappingargument allows you to go one step beyond the conditional conclusion of the a prioribootstrapping reasoning—the conditional belief that if no defeaters are present, yourexperiences are reliable—to the consequent of that conditional, the conclusion thatyour experiences are indeed reliable.

Otherwise, however, there is nothing else that the empirical bootstrappingreasoning can tell you that you were not already in a position to know on the basisof the a priori bootstrapping reasoning. Thus, even if Vogel and Cohen were wrong toclaim that the empirical bootstrapping reasoning is worthless, they were right to pointout that the reasoning does little to increase your knowledge; there is little that theempirical bootstrapping reasoning adds to what was already available thanks to the apriori bootstrapping reasoning.

Some philosophers may still feel that the a priori bootstrapping reasoning is some-how lacking in force. Various attempts can be made to explain this feeling away. Forexample, James Pryor (2004) has argued that the bootstrapping reasoning—like otherpieces of “rule circular” reasoning such as Moore’s (1939) “proof of an externalworld”—will be incapable of persuading interlocutors who harbour serious doubtsabout whether or not their experiences are undetectably unreliable. Pryor suggests that

240 RALPH WEDGWOOD

some philosophers may have mistakenly taken the fact that the reasoning is dialecticallyineffective in this way to show that the reasoning cannot produce rational belief inindividuals who start out without having any attitude (whether belief or doubt)about the reliability of their experiences.There may also be another reason why some philosophers have regarded the

bootstrapping reasoning as worthless. These philosophers may have implicitly supposedthat the purpose of the reasoning was not merely to arrive at an a priori justified belief inthe proposition that one’s experiences are not undetectably unreliable, but to givesome sort of justification of the process of taking one’s experiences at face value. In fact,however, as I have argued, there is no way in which the bootstrapping reasoning cando anything of the kind. But that does not show the bootstrapping reasoning cannotachieve a different and more modest goal: even if the availability of such reasoningcannot explain the rationality of the belief-forming process of taking experience at facevalue, it could still help to show that we are a priori justified in holding the belief thatour experiences are not undetectably unreliable.b. The solution to the sceptical paradox advocated in this chapter occupies an

intermediate position between two positions that I shall call (i) the Moorean view and(ii) the propositional foundationalist view. The general idea of this intermediate positionhas been clearly demarcated by Nicholas Silins (2008); this solution advocated here hasbeen designed as a way of occupying this intermediate position.

Mooreans—that is, philosophers who are inspired by G. E. Moore (1939)—say that anexperience as of p’s being the case would still justify you in believing p even if you hadno antecedent justification for believing the negation of the sceptical hypothesis at all.19

Propositional foundationalists like Crispin Wright (2002) say that the justification that youhave for believing pwhen you have an experience as of p’s being the case is explained, atleast in part, by your having some antecedent justification or “warrant” for acceptingthe negation of sceptical hypothesis, ‘¬s’—which functions on this view as a founda-tional “hinge proposition” underlying all empirical justification whatsoever.By contrast, the position advocated here rejects both Mooreanism and propositional

foundationalism. It rejects Mooreanism because according to this position, it is anecessary consequence of the fact that your experience justifies ordinary empirical beliefsthat you have antecedent (indeed, a priori) justification for believing the negation ofthe sceptical hypothesis. So it is not possible for experiences to justify such beliefswithout such anti-sceptical justification being in place. This is why the positionadvocated here is incompatible with Mooreanism.However, my position also rejects Wright’s propositional foundationalism, since

according to my position, the warrant that you have for rejecting the scepticalhypothesis does not in any way explain the fact that your experience justifies ordinaryempirical beliefs. On the contrary, the warrant that you have for rejecting the scepticalhypothesis is itself explained by the fact that your experience justifies such ordinary


beliefs. Thus, the position advocated here is incompatible with Wright’s propositionalfoundationalism.

Although my position is incompatible with both of these positions, it can alsocapture many of the intuitions behind both. It agrees with Wright that one musthave some a priori warrant for rejecting the sceptical hypothesis, and that this a prioriwarrant is a necessary condition of its being rational to take one’s experiences at facevalue. On the other hand, it also agrees with the Moorean that the rationality of takingone’s experiences at face value is a basic feature of rationality, which does not need tobe explained by any antecedent justification or warrant for rejecting the scepticalhypothesis. The existence of this antecedent a priori justification for rejecting thesceptical hypothesis is a necessary consequence, not an explanatory precondition, ofthe more basic truth that it is rational to take one’s experiences at face value.

c. The arguments advanced here also shed light on the nature of a priori justification.In particular, it supports two crucial points about the a priori.

First, the a priori is a less mysterious phenomenon than one might think. Indeed, theempirical is related to the a priori roughly as a natural-deduction derivation frompremises is related to a natural-deduction proof in which all assumptions are “dis-charged”. This is why so many a priori justified beliefs are in conditional propositions.One might infer that at least one of Joseph’s parents has a sibling from one’s empiricalbelief that Joseph has an uncle—in which case this inference would yield a rationalempirical belief in the conclusion that one of Joseph’s parents has a sibling. On theother hand, one could also perform this inference in a purely suppositional fashion: andin the way that was described in Section 3, one could then come to believe theconditional proposition that if Joseph has an uncle, then at least one of his parentshas a sibling; and in that case one’s justification for this conditional belief would be apriori, because it no longer depends for its justification on any empirical belief (such asthe empirical belief that Joseph has an uncle).

Secondly, we can have a priori justification for believing contingent propositions.20

The negation of the sceptical hypothesis—the proposition that your experiences arenot undetectably unreliable—is obviously contingent: there are surely possible worldsin which the sceptical hypothesis is true, and in which your experiences are in someundetectable way unreliable (for example, because you are being deceived by a demonor the like). So it seems that Kant was wrong to claim that all propositions for which wehave a priori justification are necessarily true.

Indeed, even if you really were in a sceptical scenario, you would still have a priorijustification for believing that you were not in such a sceptical scenario. In this case,then, you would have a priori justification for believing a proposition that is in fact false!False propositions, of course, cannot be known. So this also reveals that it is a mistake tothink that the a priori is primarily a matter of a certain kind of knowledge. The a priori isa certain kind of justification, not a kind of knowledge. In general, justified beliefs can befalse; and the same is true of a priori justified beliefs.21

242 RALPH WEDGWOOD

Notes1. For other statements of closely related sceptical arguments, seeWright (1985) andWilliamson

(2005).2. For example, many philosophers object to probabilism, on the grounds that if degrees of

justification can always be modelled by a probability function, then one must always havethe maximum degree of justification for believing every logical truth, and for any twopropositions p and q, if p is logically equivalent to q, one must always have exactly thesame degree of justification for believing p as for believing q.

3. Although I accept probabilism, I do not accept this sort of evidentialism: this is becauseevidentialism implies that (i) whenever a belief is justified at all, it is justified solely by one’sevidence, and (ii) whenever a proposition p counts as part of one’s evidence, one mustalways have the maximum degree of justification for believing p—and I doubt that this is thecorrect account of what makes our beliefs justified.

4. For some compelling arguments for this point, see Silins (2005).5. For this reason, I believe that the approach that was pioneered by Moore (1939)—of giving

an empirical justification for rejecting these sceptical hypotheses—fails to get to the heart ofthe matter.

6. The importance of this distinction is rightly stressed by Silins (2008).7. It is important that this process must be a process of reasoning or inference. We must exclude

the process of forming beliefs by introspection, since otherwise we would wrongly concludethat all thinkers who are capable of both introspection and mathematical reasoning havea priori justification for believing the proposition that they could express by saying ‘I believethat 313 is prime’.

8. Strictly speaking, to achieve the most explanatory level of generality, we should includebelief-revising processes as well as belief-forming processes. (Moreover, we should understandbelief “revision” broadly so that it includes merely reaffirming a belief as well as abandoning oradjusting one’s level of confidence in a belief.) But to simplify the discussion I shall just talk aboutbelief-forming processes here.

9. In commenting on this essay, Stephen Schiffer argued that my reference to “the absence ofspecial defeaters . . . stands in for something extremely complicated”. But on the contrary, itseems that we do have a general notion of a defeater: this is just the general notion of a set ofmental states that one is in at the relevant time such that although normally a certain processof reasoning would count as rational, it does not count as a rational process of reasoning forone to engage in at that time because of the presence of these mental states. It is this generalnotion that I am using here.

10. For a more detailed account of this belief-forming process, see Wedgwood (2011).11. In this way, this assumption about the rationality of taking one’s sensory experiences at face

value is similar in sprit—though not identical—to the position that James Pryor (2000) hascalled “dogmatism”.

12. Although it seems intuitively plausible that there is a rational rule of inference that corres-ponds to the non-inferential process in this way, it is a good question why this is so. (I havebeen helped to see that this is a good question, and that it is not obvious that it can beanswered merely with the resources of this essay, by Magdalena Balcerak Jackson.) It may bethat the “meta-justificatory” reasoning explored at the end of this section can itself help to


explain why this rule of inference is rational; but unfortunately I cannot attempt to answerthis question in detail here.

13. Compare the comments about the epistemology of subjunctive conditionals that are madeby Timothy Williamson (2007, chap. 5). It is important that this belief-forming processinvolves forming a belief in the material conditional ‘p! q’ on the basis of rationally inferringq from p. Not all rational thought-processes that could lead from one’s believing p tobelieving q count as an “inference”: otherwise it would always be rational for one to forma belief in the material conditional ‘p ! I believe p’!

14. I first discussed this a priori bootstrapping reasoning at conferences in the summer of 2008. In2010 I discovered that Stewart Cohen (2010, 150–5) had independently discussed afundamentally similar process of reasoning (which he calls “the a priori suppositionalreasoning”). My argument differs from Cohen’s in two main ways. First, the ultimateconclusion of Cohen’s a priori suppositional reasoning is the unconditional proposition‘My colour vision is reliable’, whereas the conclusion of my a priori bootstrapping reasoningis the conditional ‘If my experiences do not contain any defeaters, my experiences arereliable’. Secondly, Cohen seems to think that the availability of the a priori suppositionalreasoning is what explains why we have a priori propositional justification for the conclusionof this reasoning. My own view is that our possession of this a priori propositional justifica-tion is explained by the probabilistic considerations that I rehearsed in Section 2; theavailability of the a priori bootstrapping reasoning only explains how we could achieve ana priori doxastically justified belief in that conclusion.

15. I first discussed this meta-justificatory reasoning on the epistemology weblog Certain Doubtson 12 March 2006 <http://certaindoubts.com/?p=548>. This meta-justificatory reasoningrests on different assumptions from the reasoning based on the rule of “external worldintroduction” that I discussed earlier. Both processes of reasoning seem rational to me;however, the first kind of reasoning seems to me dialectically more effective than the second,while I suspect that the second kind of reasoning is explanatorily more fundamental than thefirst. Unfortunately, I cannot explore the relationship between these two kinds of reasoningin depth here.

16. As White (2006, 539) puts it, even if we have not succeeded in ruling out all “trickyexceptions” to this principle, “it is clear enough the case that concerns us does not involveany tricky business like this.” The same qualifications must be understood to apply to the“more general principle” that I articulate in the following paragraph.

17. For the general idea that belief-forming processes are of fundamental importance, and notreducible to beliefs, compare Lewis Carroll (1895).

18. I have developed a proposal about why this assumption is true elsewhere (Wedgwood2011).

19. James Pryor (2000, 519) has been interpreted by many readers as accepting this Mooreanview, when he endorses the claim that “when it perceptually seems to you as if p is the case,you have a kind of justification for believing p that does not presuppose or rest on yourjustification for anything else, which could be cited in an argument—even an ampliativeargument—for p.” If saying that your justification for p “presupposes” your justification foranother proposition q simply means that it is impossible to have this justification for pwithout having antecedent justification for q, then my position is incompatible with Pryor’sclaim, since according to my position, it is necessary that if your sensory experience justifies

244 RALPH WEDGWOOD

http://certaindoubts.com/?p=548

you in believing an ordinary proposition op, then you must have antecedent justification for‘¬s’, the negation of the sceptical hypothesis.

20. In this way, I strongly agree with John Hawthorne (2002) that some propositions are deeplycontingent a priori, and with Brian Weatherson (2005) that if scepticism and externalism arefalse (as I believe they are), then anti-sceptical propositions belong to this category.

21. An ancestor of this chapter was presented at two conferences on scepticism in 2008—one atthe University of Edinburgh (where my commentator was Alan Millar), and the other atNew York University’s Villa La Pietra in Florence (where my commentators were StephenSchiffer and Ram Neta). Something much closer to the present version was presented in2009 at the University of Southern California and at the University of Bristol, and in 2010 ata conference on justification and scepticism at the University of Bologna. I am grateful to allthose audiences, and also to Jeremy Goodman, Scott Sturgeon, and Timothy Williamson,for helpful comments. Some of the work on this chapter was carried out in 2009–10, duringmy tenure of a Research Fellowship from the Leverhulme Trust, to whom I should alsorecord my gratitude.

ReferencesCarroll, Lewis (1895). “What the Tortoise Said to Achilles,” Mind 4: 278–80.Cohen, Stewart (2002). “Basic Knowledge and the Problem of Easy Knowledge,” Philosophy and

Phenomenological Research 65: 309–29.Cohen, Stewart (2010). “Bootstrapping, Defeasible Reasoning, and A Priori Justification,”Philosophical Perspectives 24: 141–58.

Dummett, Michael (1975). “The Justification of Deduction,” Proceedings of the British Academy59: 201–31.

Edgington, Dorothy (1995). “On Conditionals,” Mind 104: 235–329.Hawthorne, John (2002). “Deeply Contingent A Priori Knowledge,” Philosophy and Phenomeno-

logical Research 65 (2), 247–69.Moore, G. E. (1939). “Proof of an External World,” Proceedings of the British Academy 25:

273–300.Pryor, James (2000). “The Skeptic and the Dogmatist,” Noûs 34: 517–49.Pryor, James (2004). “What’s Wrong with Moore’s Argument?” Philosophical Issues 14: 349–78.Silins, Nicholas (2005). “Transmission Failure Failure,” Philosophical Studies 126: 71–102.Silins, Nicholas (2008). “Basic Justification and the Moorean Response to Skepticism,” OxfordStudies in Epistemology, vol. 2: 108–40.

Vogel, Jonathan (2000). “Reliabilism Leveled,” Journal of Philosophy 97: 602–23.Weatherson, Brian (2005). “Scepticism, Rationalism, and Externalism,” in Oxford Studies inEpistemology, Vol. 1, ed. Tamar Szabo Gendler (Oxford: Clarendon Press), 311–31.

Wedgwood, Ralph (2011). “Primitively Rational Belief-Forming Processes,” in AndrewReisner and Asbjørn Steglich-Petersen (eds), Reasons for Belief (Cambridge: CambridgeUniversity Press), 180–200.

White, Roger (2006). “Problems for Dogmatism,” Philosophical Studies 131: 525–57.Williamson, Timothy (2005). “Knowledge and Scepticism,” in Frank Jackson and MichaelSmith (eds), The Oxford Handbook of Contemporary Philosophy (Oxford: Clarendon Press).

Williamson, Timothy (2007). The Philosophy of Philosophy (Oxford: Blackwell).


Wright, Crispin (1985). “Facts and Certainty,” Proceedings of the British Academy 71: 429–72.Wright, Crispin (1989). “Wittgenstein’s Rule-Following Considerations and the Central Project

of Theoretical Linguistics,” in Alexander George (ed), Reflections on Chomsky (Oxford: Black-well), 233–64.

Wright, Crispin (2002). “(Anti-) Sceptics Simple and Subtle: G. E. Moore and JohnMcDowell,”Philosophy and Phenomenological Research 65: 330–48.

246 RALPH WEDGWOOD

IV

Challenges to the A Priori

11

Articulating the A Priori–APosteriori Distinction1

Albert Casullo

The distinction between a priori knowledge and a posteriori knowledge has comeunder attack in the recent literature. Here are some examples:

It seems to me that discussions of the past decades have made clear how intricate and complex theclassical notion of the a priori is, and neither the Strong conception nor the Weak conception (noranything else) can provide a coherent explication. (Kitcher 2000, 85)

My own externalist commitments . . . lead me to think that the a priori–a posteriori distinction isnot a particularly natural one, and hence that its importance to epistemology has been grosslyoverestimated. (Hawthorne 2007, 201)

In short, what we are seeing is that there is a deep instability in the classic collection of platitudesabout a priori knowledge. (Jenkins 2008, 255)

The distinction [between a priori and a posteriori knowledge] is handy enough for a rough initialdescription of epistemic phenomena; it is out of place in a deeper theoretical analysis, because itobscures more significant epistemic patterns. (Williamson 2007, 169)

The target of the attacks is a particular concept—the concept of a priori knowledge—or, alternatively, a particular distinction—the distinction between a priori and a poster-iori knowledge. The attacks are related, but different: two are directed at the coherenceof their target; two at its significance.

Evaluating the attacks requires answering two questions. First, have they hit theirtarget? Second, are they compelling? My goal is to argue that the attacks fail becausethey miss their target. Since the attacks are directed at a particular concept or distinction,they must accurately locate the target concept or distinction. Accurately locating thetarget concept or distinction requires correctly articulating that concept or distinction.The attacks miss their target because they fail to correctly articulate the target concept ordistinction.

1A successful attack on the concept of a priori knowledge requires a correct articulationof the target concept. Correctly articulating the concept of a priori knowledge ischallenging since the target concept is obscured by several factors. First, there aretwo different approaches to articulating it. A reductive approach articulates the conceptof a priori knowledge in terms of the concept of a priori justification. According to thisapproach, S knows a priori that p just in case S’s belief that p is justified a priori and theother conditions on knowledge are satisfied. Its primary target is the concept of a priorijustification. A nonreductive approach provides an articulation of the concept of apriori knowledge that does not include any conditions involving the concept of the apriori. Its primary target is the concept of a priori knowledge.

Second, there are two approaches to providing an analysis of the target concept:theory-neutral and theory-dependent. The goal of a theory-neutral articulation of theconcept of a priori knowledge (or justification) is to provide an analysis of the conceptof a priori knowledge (or justification) that does not presuppose any particular analysisor account of the more general concept of knowledge (or justification). It aims atneutrality among the competing conceptions of knowledge (or justification). The goalof a theory-dependent analysis of the concept of a priori knowledge (or justification) isto provide an analysis of the concept of knowledge (or justification) within the moregeneral framework of a particular theory of knowledge (or justification), which I callthe background theory of knowledge (or justification).

Third, a priori knowledge (or justification) is a species of knowledge (or justification).Consequently, any item of a priori knowledge (or justified belief) must satisfy both thegeneral conditions on knowledge (or justification) and the conditions that differentiate apriori knowledge (or justification) from a posteriori knowledge (or justification). Thegoal of an analysis of the concept of a priori knowledge (or justification) is to identify theconditions that differentiate a priori knowledge (or justification) from a posterioriknowledge (or justification), rather than those that are common to both. The formerconditions are constitutive of a priori knowledge (or justification); the latter are consti-tutive of the background theory of knowledge (or justification).These obscuring factors introduce a number of common errors in accurately locating

and articulating the concept of a priori knowledge. Three are worth noting from thestart. First, most theorists who offer articulations of the concept of a priori knowledgeclaim to be articulating the so-called traditional Kantian concept of a priori knowledge.This concept arose in a period dominated by Cartesian assumptions about the nature ofknowledge and justification. Hence, when offering an articulation of this concept, it iscritical to distinguish between the features constitutive of a priori knowledge (orjustification) as opposed to the features constitutive of the background theory ofknowledge (or justification). Failure to distinguish between these two features canlead to mistaken articulations of the traditional concept of a priori knowledge.

250 ALBERT CASULLO

Second, most contemporary theorists reject the traditional Cartesian accounts ofknowledge and justification, and offer articulations of the traditional concept of a prioriknowledge within the context of their preferred theory of knowledge. When articu-lating the traditional concept of a priori knowledge within a nontraditional theory ofknowledge, one must take care to ensure that the articulation coheres with therequirements of the new background theory. Failure to do so can result in mistakenconclusions about the implications of the new background theory for the a priori.Third, if one offers an articulation of the traditional concept of a priori knowledge

within a nontraditional background theory and arrives at the conclusion that theresulting account of a priori knowledge is problematic, one must take care to deter-mine whether the source of the problem is the requirements of the background theoryof knowledge or the requirements of the a priori. Failure to do so can lead to themistaken conclusion that the a priori is problematic when the problem is rooted in thebackground theory of knowledge.The question before us is whether the concept of a priori knowledge is coherent and

significant. The four authors cited earlier express reservations about its coherence orsignificance. I begin with a selective review of some of Philip Kitcher’s seminal workon the a priori.2 My purpose is twofold. The first is to introduce two different strategiesfor challenging the coherence or significance of the concept of a priori knowledge.The second is to provide clear examples of arguments against the a priori that miss theirtarget by committing one of the errors that I identified earlier. In the subsequentsections, I will turn to the arguments of John Hawthorne, C. S. Jenkins, and TimothyWilliamson and identify variations of these errors.

2Philip Kitcher offers two different strategies for arguing that the concept of a prioriknowledge is either incoherent or insignificant. His original strategy is to argue thatthere is no a priori knowledge. If he is correct, it follows that the a priori–a posterioridistinction does not mark a significant division in epistemic reality; it does not mark anydivision. It does not follow, however, that the concept is incoherent. In fact, Kitcheremploys an analysis of the concept of a priori knowledge as an essential premise in hisargument in support of the conclusion that there is no a priori knowledge. In his laterwork, Kitcher adopts a different approach. Rather than offering an argument againstthe existence of a priori knowledge that involves an analysis of the concept, hechallenges the concept itself. He maintains that the traditional concept is too complexto be coherently articulated.Kitcher (1983) offers the following influential argument against the traditional view

that some knowledge is a priori:

(K1) The concept of a priori knowledge entails that a priori warrant is indefeasibleby experience.

ARTICULATING THE A PRIORI–A POSTERIORI DIST INCTION 251

(K2) The warrant conferred by alleged a priori sources of knowledge is defeasibleby experience.

(K3) Therefore, no knowledge is a priori.

(K1) is supported by the following analysis of the concept of a priori knowledge:

X knows a priori that p iff X knows that p and X’s knowledge that p was produced by a processwhich is an a priori warrant for it.Æ is an a priori warrant for X’s belief that p if and only if Æ is a process such that, given any life esufficient for X [to acquire the concepts in] p(a) some process of the same type could produce in X a belief that p;(b) if a process of the same type were to produce in X a belief that p, then it would warrant X in

believing that p;(c) if a process of the same type were to produce in X a belief that p, then p. (Kitcher 1983, 24)

The analysis is reductive since it analyzes a priori knowledge in terms of a prioriwarrant. It is also theory-dependent since Kitcher (2000, 66) maintains that his“general understanding of warrants is a version of reliabilism.”Conditions (b) and (c) of the analysis share a common feature: both impose higher

standards on a priori warrant than those required by Kitcher’s background theory ofwarrant. Condition (c) precludes the possibility of a priori warranted false beliefs.Condition (b) requires that S’s a priori warranted belief that p be indefeasible byexperience in any world in which S has sufficient experience to acquire the conceptsin p. Reliabilism, however, does not preclude either the possibility of warranted falsebeliefs or the possibility of empirically defeasible warranted beliefs. Consequently, inthe absence of some compelling supporting argument, the higher standards are ad hocand should be rejected.3

Kitcher’s argument provides a clear example of an error of the second type. He offersan articulation of the traditional concept of a priori justification within a reliabilisttheory of warrant, but the articulation introduces conditions on a priori warrant thatare not supported by his background theory of warrant. As a consequence, he arrives atthe mistaken conclusion that the theory precludes a priori warrant.

Kitcher (2000) adopts a different strategy. He acknowledges that his articulation ofthe concept of a priori warrant faces difficulty, but now argues that the alternativefavored by his opponents, the Weak conception or (WC),

(WC) S’s belief that p is justified a priori just in case S’s belief that p is justified by anonexperiential process,

is also open to objection. The upshot is that there is no coherent articulation of thetraditional concept of a priori knowledge.

We now turn to two of the arguments that Kitcher offers against (WC).4 The firstcontends that (WC) fails to capture a feature of the traditional concept of a prioriknowledge: “the tradition ascribes to a priori knowledge the functional significance ofbeing in a position to prescribe to future experience; knowledge that prescribes to

252 ALBERT CASULLO

future experience is irrefutable by future experience” (Kitcher 2000, 77). Let us grantthat the tradition ascribes to a priori knowledge the functional significance of prescrib-ing to future experience. This observation, taken by itself, does not provide a basis forrejecting (WC) since Kitcher has not shown that this feature is constitutive of thetraditional concept of a priori knowledge. Moreover, there is good reason to deny thatit is. It is generally acknowledged that the traditional concept of knowledge is Cartesianfoundationalism, according to which S knows that p only if S’s justification for thebelief that p is infallible, incorrigible, and indubitable.5 Hence, incorrigibility, orimmunity to revision, is a feature constitutive of the traditional concept of knowledge.It is not a feature that differentiates a priori knowledge from empirical knowledge.This argument provides a clear example of an error of the first type. Kitcher mistakes

a feature of the traditional theory of knowledge for a feature of the traditional conceptof a priori knowledge and, as a consequence, draws the mistaken conclusion that (WC)does not provide an adequate articulation of the traditional concept of the a priori.The second argument contends that (WC) is too weak. Kitcher invites us to consider

the following thought experiment. Suppose that a cubical die, composed of somehomogeneous material and whose faces are numbered 1 through 6, is rolled once.What is the chance that the uppermost face will be the one numbered 6? One mightreason as follows. Since the material is homogeneous, the situation is symmetrical withrespect to the six faces. One of the numbered faces will be uppermost. Therefore, theprobability that it will be the one numbered 6 is 1/6. Kitcher (2000, 78) maintains thatthis thought experiment involves a nonexperiential process that meets reliabilist stand-ards. Therefore, according to (WC), the conclusion in question is known a priori. But,Kitcher (2000, 79) contends, “this will set the Weak conception at variance with theclassical view of the bounds of apriority.”

Kitcher’s contention is mistaken. He fails to distinguish between the requirements of(WC) and the requirements of the background theory in which (WC) is embedded. If(WC) is embedded within the traditional Cartesian theory of knowledge then itdelivers results consistent with the classical view of the bounds of apriority. In orderfor the conclusion in question to be known, on the traditional theory, it must beinfallible, incorrigible, and indubitable. Since it does not meet those conditions, thatconclusion is not known, a priori or otherwise. Kitcher generates the appearance ofvariance between (WC) and the classical theory by embedding (WC) within a reliabilisttheory of knowledge. Since reliabilism imposes lower standards for knowledge than thetraditional theory, the conclusion of the argument is (let us grant) known a priori. Thevariance is due entirely to the difference in the standards for knowledge imposed by thetwo different background theories in which (WC) is embedded. It is not due to (WC).This argument provides a clear example of the third type of error. Kitcher mistakes a

consequence that is due to the nontraditional background theory in which he embeds(WC) for a consequence of (WC) itself and, as a result, draws the mistaken conclusionthat (WC) does not provide an adequate articulation of the traditional concept of apriori knowledge.


3John Hawthorne contends that the a priori–a posteriori distinction is not a natural one.This conclusion, he maintains, derives from his commitment to epistemological andsemantic externalism. My main concern is the case based on his commitment toepistemological externalism which, I believe, carries the primary burden of his argu-ment. Hawthorne offers three leading arguments based on epistemological consider-ations. I will argue that all three fall short of their goal. The first fails because it does notdistinguish between reductive and nonreductive analyses of the concept of a prioriknowledge. The second fails because it turns on an incorrect account of the relation-ship between experiences that warrant belief and experiences that defeat warrant. Thethird raises a problem that arises from embedding the traditional concept of a prioriknowledge within his externalist general theory of knowledge, but the problem is dueentirely to the background theory of knowledge.

According to the safety account of knowledge preferred by Hawthorne:

(S) S knows p iff there is no close world where S makes a mistake that is relevantlysimilar to his actual belief that p. (2007, 202)

The central idea of the account is that of a relevantly similar mistake. What constitutessuch a mistake? Details are scarce, but Hawthorne tells us two things. First, the fact thatS makes a mistake about p in some close possible world does not show that S does notactually know that p “since he may use a relevantly different method at that closeworld” (2007, 202). Second, the fact that S does not make a mistake about p in anyclose possible world does not show that S actually knows p. For example, suppose thatS uses an unreliable method to form the belief that p, where p is some necessary truth.Although there is no world in which S believes p falsely, “there are relevantly similarcases in which [S is] in error” (2007, 202). So we have two leading ideas. Mistakeregarding p, where a different belief forming method is used, is not relevantly similar;and mistake regarding propositions other than p, where the same belief formingmethod is used, is relevant. Hence, on Hawthorne’s account, method of belief forma-tion plays a central role in marking the difference between knowledge and true belief.

Hawthorne (2007, 203 n. 5) takes as his starting point Kitcher’s account of a prioriknowledge and maintains that “it is quite clear that Kitcher’s basic idea is that a processwarrants a belief a priori iff, no matter how the environment is, that process is a warrantprovider.” Building on this idea, Hawthorne (2007, 202) maintains: “It is oftenthought that in a case of a priori knowledge, the status of a belief as knowledge doesnot constitutively depend on the external environment (this being one natural take onthe idea that a priori knowledge is independent of perceptual experience).” Ratherthan endorsing Kitcher’s articulation, Hawthorne offers the following alternative:

(HA1) x knows p a priori after duration d of x’s existence iff any possible intrinsicduplicate y of x knows p after duration d of y’s existence. (2007, 202)

254 ALBERT CASULLO

Hawthorne quickly disposes of (HA1) by inviting us to consider a variant of the well-known fake barn case. Suppose that Henry is driving in barn gas country, an areariddled with barn gas that induces hallucinations of barns. Suppose that the barns are allreal in barn gas country and that Henry happens to look at a barn from one of the fewlocations that is not riddled by the gas. As in the original fake barn case, Henry does notknow that there is a barn in front of him. Hawthorne now introduces a parallel casewith respect to a priori beliefs:

Suppose there exists a priori gas that induces the phenomenology of blatant obviousness for falsepropositions. Consider a person who believes a proposition not for any empirical reason butbecause the phenomenology of obviousness causes him to do so. Suppose the claim in question isthat all bachelors are men. Consider a duplicate of that person who is embedded in an environ-ment riddled with a priori gas. As a matter of luck he does not stumble into the gas. He in factforms the belief that all bachelors are men. But he could very easily have stumbled into the gasand believed—due to felt obviousness—that all bachelors are women. Insofar as one judges thatthe person does not know in fake barn cases, it is natural enough to judge that the person doesnot know in a priori gas cases. But this means that if we cling to the environment dependenceidea, very few of our beliefs will count as a priori. (2007, 205)

Suppose that we grant that the argument is compelling. What moral should we drawfrom this?The source of the problem is clear. Hawthorne takes as his starting point Kitcher’s

analysis of the concept of a priori knowledge. Kitcher’s analysis, however, is reductive.He analyzes that concept in terms of the concept of a priori warrant. Hence, hearticulates the idea that a priori knowledge is independent of perceptual experience as

(K) The status of a belief as warranted does not constitutively depend on theexternal environment.

Hawthorne’s analysis, however, is nonreductive. He articulates the idea that a prioriknowledge is independent of perceptual experience as

(H) The status of the belief as knowledge does not constitutively depend onthe external environment.

If the idea that a priori knowledge is independent of perceptual experience is articu-lated as a condition on a priori warrant, the problem that Hawthorne raises vanishes.There is no immediate objection to maintaining that Henry’s belief is warranted in fakebarn country or barn gas country. Similarly, a priori gas cases pose no immediateobjection to (K). In short, the failure to distinguish between reductive and nonreduc-tive analyses of the concept of a priori knowledge undermines Hawthorne’s firstargument.Hawthorne considers the option of providing a reductive analysis of a priori

knowledge in terms of a priori justification in a later discussion. There he considers aparticular version of the view that intellectual intuition is a source of a priori justification.


That view, according to Hawthorne (2007, 214), “is most naturally motivated by theidea that a priori knowledge decomposes into an ‘internalist’ component that isaccessible to the subject . . . and an ‘externalist component’ that includes various reli-ability conditions.” He rejects the view by raising a number of considerations againstthe accessibility features of its internalist component.

Hawthorne’s case against reductive analyses is not compelling since it is based on theassumption that an account of a priori warrant must take on the objectionableaccessibility features that are the target of his attack. Kitcher, however, provides anaccount of a priori warrant that is avowedly externalist. His preferred view of warrantsis a version of Goldman’s process reliabilism that does not involve the internalistcomponent that Hawthorne rejects. So Hawthorne has not provided any compellingreason to conclude that epistemological externalism presents any special barrier toarticulating a reductive analysis of the concept of a priori knowledge.

Hawthorne acknowledges that environment independence is not the traditionalway of marking the view that a priori knowledge is independent of experience. Heintroduces two alternative ways of developing the view. The first appeals to grounds of abelief:

(HA2) A case of knowing is a priori iff experience does not form part of thegrounds of the belief. (Hawthorne 2007, 208)

Hawthorne rejects (HA2) because he finds the notion of grounds unclear. The secondappeals to methods of belief formation, where a method of belief formation is a process thatdelivers or sustains a belief:

(HA3) Experience–Independence: A case of knowing is a priori if it is sustained bya method that is not experience involving. (Hawthorne 2007, 208)

Hawthorne maintains that, on initial examination, (HA3) gives favorable results sincemathematical knowledge based solely on reasoning, say working through a mathemat-ical proof that p, comes out a priori. Closer examination, however, reveals that there isa problem lurking:

Even though I have carefully worked through a mathematical proof that p, I will not know p ifI get empirical evidence that I am mad, or that human or mechanized experts have agreed thatnot-p, or that there is a priori gas in the area, or that I have made lots of mistakes using a verysimilar proof technique in the past, or that lots of smart people are inclined to laugh when theyhear my proof . . . . Call knowledge-destroying experiences Bad experiences. Call the remainderGood experiences . . . . That my proof counts as knowledge appears to depend crucially on itsbeing accompanied by Good experiences. But if the process of arriving at putatively a prioriknowledge is individuated so as to include Good experiences, then it will count as a posteriori bythe experience dependent criterion. (Hawthorne 2007, 209–10)

Hence, (HA3) appears to have the consequence that little, if any, knowledge is a priori.The structure of Hawthorne’s argument is immediately puzzling. It looks like this:

256 ALBERT CASULLO

(H1) Experience–Dependence: A case of knowing is a posteriori if it sustained by amethod that is experience-involving.

(H2) Suppose that S has carefully worked through a mathematical proof that p.(H3) S’s proof does not count as knowledge if it is accompanied by Bad experiences.(H4) Therefore, S’s proof counts as knowledge only if it is accompanied by Good

experiences.(H5) If the process of working through a proof is individuated so as to include

Good experiences, then it will count as experience-involving and, hence, aposteriori by the experience-dependent criterion.

Consider a parallel argument restricted to the empirical domain. Suppose thatS carefully forms the true belief that there is a barn over there on the basis of herperceptual experience. S will not know that there is a barn over there if she getstestimonial evidence that there is barn gas in the area, or that she frequently mistakesother buildings for barns, or that no one else in the vicinity sees a barn. Call know-ledge-destroying testimony Bad testimony. Call the remainder Good testimony. Theparallel argument looks like this:

(T1) Testimony-Dependence: A case of knowing is testimonial if it is sustained by a methodthat is testimony-involving.

(T2) Suppose that S has carefully formed the true belief that there is a barn over there on thebasis of her perceptual experience.

(T3) S’s true belief does not count as knowledge if it is accompanied by Bad testimony.(T4) Therefore, S’s true belief counts as knowledge only if it is accompanied by Good

testimony.(T5) If the process of forming a belief based on one’s perceptual experience is individuated so

as to include Good testimony, then it will count as testimonial by the testimony-dependent criterion.

Although there are many different ways of individuating belief forming processes, it isevident that very few epistemologists, if any, will take seriously a way of individuatingthem on which paradigm cases of perceptual belief turn out, on closer examination, tobe testimonial merely in virtue of the fact that testimonial evidence can destroyperceptual knowledge.In order to sort out matters, we need to distinguish between experiences that

warrant the belief that p as opposed to experiences that defeat warrant for the beliefthat p. In Hawthorne’s terminology, the experiences that defeat warrant are Badexperiences; the remaining experiences are Good experiences. A similar distinctioncan be made in the testimony example. Testimony that defeats warrant is Badtestimony; testimony that does not is Good testimony. Returning now to the testi-mony version of Hawthorne’s argument, we can see clearly the source of the problem.The step from (T3) to (T4) is invalid. From the fact that Bad testimony can defeat S’sperceptual warrant for the belief that p, it does not follow that perception can warrantS’s belief that p only if it is accompanied by Good testimony. Perceptual experience


can warrant S’s belief in the absence of Good testimony. All that is necessary is that S’sbelief not be accompanied by Bad testimony. By parity of reasoning, the step from (H3)to (H4) in Hawthorne’s original argument is also invalid. From the fact that Badexperience can defeat the warrant conferred on S’s belief that p by having carefullyworked through a proof that p, it does not follow that carefully working through aproof that p can warrant S’s belief that p only if accompanied by Good experiences. S’sproof can warrant S’s belief that p in the absence of Good experiences. All that isnecessary is that S’s proof not be accompanied by Bad experiences.6

Hawthorne contends that his externalist theory of knowledge challenges the signifi-cance of the a priori–a posteriori distinction. His second argument, pared down to itsessential elements, rests on twopremises.7 First, experiences of typeE candefeat thewarrant(destroy the knowledge) conferred on S’s belief that p by virtue of being sustained bymethod M. Second, if experiences of type E can defeat the warrant (destroy the know-ledge) conferred on S’s belief that p by virtue of being sustained by method M, then S’sbelief that p is warranted (known) by virtue of being produced by method M only if it isaccompaniedby experiences of typeE.Hawthorne’s externalist theory of knowledge is notessential to the formulation of the argument and does not support its key second premise.Hence, the argument neither derives from nor is supported by his externalist theory ofknowledge. It rests on an incorrect account of the relationship between experiencesthat confer warrant on a belief and experiences that defeat warrant. From the fact thatexperiences of type E can defeat the warrant (destroy the knowledge) conferred on S’sbelief that p by virtue of being sustained by method M, it does not follow that S’sbelief that p is warranted (known) by virtue of being produced by method M onlyif it is accompanied by experiences of type E. Although (bad) testimonial experi-ences can defeat the warrant conferred on S’s belief that p by virtue of beingsustained by visual perception, it does not follow that S’s belief that p is warrantedby visual perception only if it is accompanied by (good) testimonial experiences.

Hawthorne’s final argument turns on the individuation of belief forming methods.Suppose that a student learns some laws of nature from a teacher, remembers them at alater time, and applies them to derive further nomic beliefs and conditional predictions.We are to suppose that the student’s beliefs and predictions are highly reliable; most aresufficiently safe to count as knowledge. Hawthorne maintains that our natural reactionis that the student’s knowledge is not a priori:

the process that led to the fixation of belief included experiential exposure to the teacher.The knowledge is a posteriori knowledge, achieved via testimony, not a priori knowledge.(2007, 211)

There is, however, a twist. Suppose that the student derives a conditional predictionfrom some laws stored in memory. We can distinguish at least two different belief-forming processes:

258 ALBERT CASULLO

There is a process that begins with the teacher telling him the laws and ends with applying somelaws to derive a conditional prediction. But there is a shorter process that begins with retrievingthe laws from the relevant internal information bank and ends with producing the conditionalprediction. One of the processes is experience-dependent. One is not. Which shall we use to testwhether the belief is a priori? Let us call the process beginning with the retrieval Short and theprocess beginning with the interaction with the teacher Long. Is there any deep mistake in takingShort to be the relevant safe method? (Hawthorne 2007, 211)

The student’s conditional prediction can be viewed as the product of either of twodifferent belief forming processes. Both are safe and yield the result that the studentknows the conditional prediction. On one, according to Hawthorne, the resultingknowledge is a priori, but on the other, it is a posteriori. Yet, there is no “deepmistake” in choosing one over the other.Hawthorne’s final argument can be summarized as follows:

(H*1) Whether the knowledge in question is a priori turns on the choice of belief formingprocess.

(H*2) There is no deep mistake in choosing one over the other.(H*3) Therefore, the a priori–a posteriori distinction is not deep.

The argument is puzzling since the primary question that it raises concerns theepistemology of preservative memory rather than the epistemology of the a priori.The initial premise of the argument is supported by two claims. If Long is the relevantbelief forming process, then the student’s knowledge is a posteriori since the student’soriginal knowledge is warranted by testimony. Second, if Short is the relevant beliefforming process, the student’s knowledge is a priori since it is not warranted bytestimony and the student has no other relevant empirical evidence that the law inquestion is true. Hence, the primary question raised by the argument is

(Q) If S’s original knowledge that p is warranted by testimony and that knowledgeis preserved by memory, is S’s later knowledge that p warranted by testimony?

On the face of it, (Q) is interesting and significant.8 Moreover, it is a question aboutpreservative memory and not the a priori. But, on Hawthorne’s account, (Q) is notvery deep since the answer to it turns on the choice of belief forming process.There are other examples of epistemological questions that appear, on the face of it, to

be interesting and significant but turn on the choice of belief forming process. For example,Kitcher rejects reliabilism in favor of a socio-historical conception of knowledge:

Onmy socio-historical conception of knowledge, the knowledgewe have today isn’t simply amatter ofwhatwehave experienced or thought during the course of our lives, but is dependent on the historicaltradition in which we stand and on the social institutions that it has bequeathed to us. (2000, 80)

More generally, he rejects “synchronic” conceptions of knowledge, which entail thatS’s warrant for the belief that p depends only on S’s cognitive states and processes(2000, 81–2). Hence, if we return to the example of our student who learns scientific


laws from a teacher at an early age and later retrieves them from preservative memory,Kitcher would maintain that the relevant belief forming process does not stop with thestudent’s teacher but extends back to his teacher’s teachers and beyond. The questionwhether a socio-historical or synchronic conception of knowledge is correct seemsinteresting and significant. But, by Hawthorne’s lights, it turns out not to be very deep,since the answer to it turns on the choice of belief forming process. So the problemHawthorne raises, if it is genuine, extends far beyond the a priori. If he is right, manyepistemological questions are superficial.

Hawthorne, however, cannot maintain, by the lights of his own theory, thatquestions that turn on the choice of belief forming process are superficial. Consider avariant of his case of the student who learns laws of nature from a teacher. The variant isidentical to Hawthorne’s original case with one exception. As in the original case, thestudent learns many laws of nature from a competent teacher. But the student alsolearns a single law of nature from an incompetent teacher, whose scientific pronounce-ments are usually false. On this particular occasion, by sheer luck, her pronouncementis true. Let us now suppose the student retrieves from memory the law he learned fromthe incompetent teacher and bases his conditional prediction on that law. Call theprocess that results in the conditional prediction Long*. In this case, the student doesnot know the conditional prediction since it is based on a law that he does not know.

Hawthorne maintains that, in the original Short and Long cases, the choice betweenthe belief forming processes is insignificant because, from the perspective of a safetytheory, both yield the result that the student knows the conditional prediction. ButShort is also a terminal segment of the process Long*, and the choice between Shortand Long* does make a difference in terms of whether the subject knows. If there is nodeep mistake in choosing Short over Long*, then the distinction between knowledgeand true belief is not very deep. But, assuming that the distinction between knowledgeand true belief is significant, then Hawthorne must provide some account of how toindividuate belief forming processes for purposes of determining whether a true beliefconstitutes knowledge. Once such an account is provided, it can be employed toindividuate belief forming processes for purposes of determining whether they areexperience-dependent.

The case of the incompetent teacher shows that Long* is the relevant belief formingprocess for determining whether the student knows the conditional prediction. It alsoshows that the student’s original warrant for the belief that p is relevant to whether alater true belief based on the student’s memory that p constitutes knowledge; that is, itshows that the correct answer to (Q) is affirmative. Therefore, in Hawthorne’s originalcase, Long is the relevant belief forming process. It is the process that should be used totest whether the student’s knowledge is a priori. Since the student’s knowledge in Longis warranted by the teacher’s testimony, it is a posteriori.9

We can now draw a more general conclusion regarding Hawthorne’s final argu-ment. Hawthorne endorses a general theory of knowledge in which method ofbelief formation plays a central role in determining whether a true belief constitutes

260 ALBERT CASULLO

knowledge. He then maintains that if one embeds the Experience-Independentaccount of a priori knowledge in that general theory of knowledge, a problem arises.Whether items of knowledge turn out to be a priori depends on how belief formingprocesses are individuated. This problem, however, is not due to the embeddedconcept of a priori knowledge. It is due to the fact that the general theory ofknowledge in which it is embedded assigns a central role to methods of beliefformation without providing an account of how they are to be individuated.

4Carrie Jenkins’s (2008) goal is to offer an epistemology of arithmetic that reconcilesapriorism, realism, and empiricism. Empiricism maintains that “all our knowledge ofthe world as it is independently of us must either be, or ultimately rest upon,knowledge obtained through the senses” (Jenkins 2008, 2). A priori knowledge is“knowledge secured without epistemic reliance on any empirical evidence” (Jenkins2008, 4). Jenkins (2008, 4) contends that it is often further assumed that “a prioriknowledge is knowledge which does not epistemically depend on the senses at all” andmaintains, to the contrary, that “there is a significant difference between epistemicindependence of empirical evidence and epistemic independence of the sensesaltogether.” This difference is the focus of her investigation. The leading idea of heraccount of arithmetical knowledge is that “experience grounds our concepts (which isnot the same as supplying evidence for any proposition), and then mere conceptualexamination enables us to learn arithmetical truths” ( Jenkins 2008, 4). The account,according to Jenkins (2008, 4–5), “makes it reasonable to describe our means ofacquiring such knowledge as both a priori (in the sense of independent of empiricalevidence) and empirical.”Jenkins’s account of arithmetical knowledge is rich and nuanced. Here I offer a

general outline that highlights only the features relevant to my discussion. Jenkinsassumes that there is a sense of ‘knowledge’ and a sense of ‘justification’ on whichexternalism is correct. Her goal is to defend the view that we have a priori knowledgeof arithmetic in the externalist sense. According to Jenkins (2008, 126), arithmeticaltruths are conceptual truths—that is, “we can know about arithmetic by examining ourconcepts.” But, in order to know such truths by examining our concepts, the conceptsin question must be grounded. A concept is grounded just in case “it is relevantlyaccurate and there is nothing lucky or accidental about its being so” (Jenkins 2008,128). Moreover, a concept must be justified in order to be grounded. A concept isjustified just in case “it is rationally respectable for us to rely on it as a relevantly accurateguide to the world” ( Jenkins 2008, 129). Hence, according to Jenkins,

Concept accuracy, justification, and grounding are important because, while we have no reasonto suppose that examining just any old concepts will help us learn about the independent world,examining accurate concepts can help us acquire true beliefs about the world, examining justified


concepts can help us acquire justified beliefs about the independent world, and examininggrounded concepts can help us acquire knowledge of it. (2008, 131)

Finally, empiricism mandates that the only data relevant to concept justification andgrounding are “data obtained through the senses” ( Jenkins 2008, 137).

My focus is on the alleged reconciliation between apriorism and empiricism. Thereconciliation rests on the following contentions:

( J1) S knows a priori that p iff S knows that p and S’s knowledge is epistemically independ-ent of empirical evidence.

( J2) S’s (basic) arithmetical knowledge that p depends epistemically on the concepts consti-tutive of S’s belief that p being grounded by the senses.

( J3) The sensory input that grounds the concepts constitutive of S’s belief that p does notconstitute evidence for S’s belief that p.

( J4) S knows empirically that p iff S knows that p and S’s knowledge depends epistemicallyon the senses in some way.

Since S’s basic arithmetical knowledge does not depend epistemically on empiricalevidence, it is a priori; but since it does depend epistemically on the senses, it is alsoempirical.

A question immediately arises regarding the reconciliation. Since Jenkins allows thatknowledge can depend epistemically on the senses without depending on empiricalevidence, her characterization of the a priori seems too narrow. Why not characterize apriori knowledge more broadly as knowledge that is epistemically independent of thesenses? Jenkins is sensitive to the issue and offers two arguments in defense of hercharacterization. The first contends that it is in line with the tradition. The secondmaintains that the traditional account of the a priori is unstable. Consequently, nocharacterization can fully salvage it.

The defense based on tradition appeals to Kant, Chisholm, and Moser. With respectto Kant, Jenkins remarks:

I hope that qualms about my decision to retain the term ‘a priori’ may be dispelled when werecall that modern usage of the term ‘a priori’ was largely determined by Kant, and that Kantallowed that some a priori knowledge—the ‘impure’ sort—depends upon experience in so far asthe concepts involved are ‘derived from’ experience . . . I have proposed that the only way inwhich arithmetical knowledge depends on sensory input is in so far as the concepts involved mustbe appropriately related to that input in order for us to count as knowing arithmetical propos-itions. (2008, 252)

Kant’s distinction between pure and impure a priori knowledge offers little precedentor support for Jenkins’s characterization of a priori knowledge. Kant maintains thatimpure a priori knowledge involves concepts derived from experience, but pure apriori knowledge does not. The point of Kant’s distinction is negative. He maintainsthat the fact that experience may be necessary to acquire the concepts constitutive ofsome proposition that p does not preclude a priori knowledge that p. Moreover, Kant

262 ALBERT CASULLO

does not maintain that the experience necessary to acquire concepts plays any epistemicrole. Jenkins, however, proposes a dependence on experience that is both positive andepistemic. On her account, S has a priori arithmetical knowledge that p only if theconcepts constitutive of p are derived from experience in an appropriate manner. Heraccount, unlike Kant’s, rules out the possibility of pure a priori knowledge based on anexamination of concepts.Jenkins next appeals to Chisholm, who states:

Speaking very roughly, we might say that one mark of an a priori proposition is this: once youunderstand it, you see that it is true. We might call this the traditional conception of the a priori.(1977, 40; quoted in Jenkins 2008, 252)

Here she maintains:

If this characterization is even ‘very roughly’ correct, then it looks as though arithmeticalknowledge as I envisage it has as good a claim as any to count as a priori. With groundedarithmetical concepts in place, we are in a position to see that 7 + 5 = 12 is true. (2008, 252–3)

But it is clear that Chisholm’s remarks, taken at face value, do not support Jenkins’scontention. On Chisholm’s account, it is understanding alone that is sufficient for apriori knowledge. Putting this point in Jenkins’s terminology: with arithmeticalconcepts in place, we are in a position to see that 7 + 5 = 12 is true. Chisholm’saccount does not require that those concepts be grounded.Jenkins (2008, 253), however, does not think that we should read Chisholm too

literally here: “Chisholm’s characterization is not (in my opinion) best interpreted asimplying that concepts don’t need to be grounded in order to be a source ofknowledge.” This suggestion does not square well with Chisholm’s official character-ization of immediate a priori knowledge:

D 3.1 h is an axiom = Df h is necessarily such that (i) it is true and (ii) for everyS, if S accepts h, the h is certain for S.

D 3.2 h is axiomatic for S = Df (i) h is an axiom and (ii) S accepts h. (1977, 42)

For Chisholm (1977, 41), accepting some proposition requires that “you grasp what it isfor that proposition to be true,” but there is no requirement that, in order to grasp whatit is for some proposition to be true, its constituent concepts must be grounded. Jenkinsalso overlooks the fact that Chisholm explicitly addresses the role of experience inconcept acquisition, and argues that the manner in which one acquires concepts isirrelevant to the epistemic status of beliefs formed on the basis of examining thoseconcepts.10

Jenkins’s final appeal is to Paul Moser (1987, 1), who maintains that the distinctionbetween a priori and a posteriori knowledge “may be plausibly regarded as connotingtwo kinds of epistemic justification.” He (1987, 1) goes on to articulate this idea asfollows: “an instance of knowledge is a priori if and only if its justification conditionis a priori in the sense that it does not depend on evidence from sensory experience.”


Moser’s characterization, with its specific reference to independence of “evidence”from sensory experience, is similar to that of Jenkins. Hence, it appears to offer somesupport for her contention that her characterization accords with the tradition.

But there is reason to be cautious here. Moser maintains that justification is a necessarycondition for knowledge and that the a priori–a posteriori distinction is fundamentally adistinction between two types of justification. So Moser is offering a reductive analysis ofthe concept of a priori knowledge in terms of the concept of a priori justification. I haveargued that the traditional concept of a priori justification is best articulated as

(APJ) S’s belief that p is justified a priori just in case S’s belief that p is nonexper-ientially justified.

Moser, however, appears to disagree. He maintains:

(APJ*) S’s belief that p is justified a priori just in case the justification of S’s beliefthat p does not depend on evidence from sensory experience.

What is the source of this difference?Moser (1989, 42) favors a particular conception of epistemic justification: “the

notion of justification as an adequate indication, relative to one’s total evidence, that aproposition is true.” We can characterize this conception, which I call the AdequateEvidence conception, as follows:

(AE) S’s belief that p is justified just in case S’s belief that p is adequately supportedby S’s total evidence.

The conjunction of (APJ) and (AE) yields (APJ*).Two points are important for our purposes. First, (APJ) is a theory-neutral articula-

tion of the traditional concept of a priori justification; it does not presuppose anyparticular theory of epistemic justification. (APJ*) is a theory-dependent articulation ofthat concept; it presupposes (AE). Second, (APJ) and (APJ*) are equivalent if, but onlyif, justification is a function of evidence alone. Since Moser takes justification to be afunction of evidence alone, he would take (APJ) and (APJ*) to be equivalent.

Given that Moser’s articulation of the traditional concept of a priori justification istheory-dependent, Jenkins should endorse it only if she endorses the theory ofjustification that it presupposes. She should, however, have serious reservations aboutendorsing (AE) for two reasons. First, Jenkins’s account of arithmetical knowledgepresupposes an externalist theory of justification, but Moser (1989, 71–7) contends that(AE) is incompatible with externalist theories of justification. Second, and moreimportant, (AE) is at odds with Jenkins’s account of the relationship between justifiedconcepts and justified belief. According to Jenkins (2008, 129), a correctly conductedexamination of concepts yields justified belief only if the concepts involved in thatexamination are justified. But, as she stresses, the experiences that justify a concept donot constitute evidence for beliefs based on an examination of those concepts. So either

264 ALBERT CASULLO

justified belief based on an examination of concepts does not require that theexamined concepts be justified or justified belief is not a function of evidencealone. In other words, if Jenkins’s account of arithmetical knowledge is correct,then (AE) is false. Since Moser’s theory-dependent analysis of the concept of a priorijustification presupposes (AE), it is not a precedent that Jenkins can coherentlyembrace.Jenkins recognizes that, despite the alleged historical precedents, many will balk at

the very idea of knowledge that is both a priori and empirical. Her second argumentaddresses this concern directly. Here Jenkins (2008, 255) maintains that there is a deepinstability in the classic collection of platitudes about a priori knowledge, since itincludes all of the following:

(A) All a posteriori knowledge is knowledge that depends on empirical evidence.(B) Only knowledge that is independent of experience is a priori.(C) All knowledge is either a priori or a posteriori and none is both.

Jenkins contends that her account of arithmetical knowledge shows that some memberof the collection must be given up. She acknowledges that her proposal to reject (B)may sound radical, but counters that the remaining options are also radical.But this is a mistake. The instability in the triad (A)–(C) is due to the fact that

Jenkins introduces into her theory of arithmetical knowledge a conception of a prioriknowledge that presupposes (AE), but (AE) does not cohere with her general theoryof justification. Since she maintains that justified belief based on an examination ofconcepts requires justified concepts, she denies that justified belief is a functionof evidence alone. Hence, Jenkins’s argument rests on an error of the second type:she endorses an articulation of the traditional concept of a priori knowledge that doesnot cohere with her background theory of justification.

(APJ), however, coheres with her theory of justification since it is theory-neutral.Moreover, if one replaces (APJ*) with (APJ) in her theory of arithmetical knowledgeand amends the triad to reflect that change, the resulting triad is stable:

(A*) All a posteriori knowledge is knowledgewhose justificationdependson experience;(B*) Only knowledge whose justification is independent of experience is a priori;(C) All knowledge is either a priori or a posteriori and none is both.

Nothing must be given up to accommodate Jenkins’s theory of arithmetical know-ledge. Such knowledge is a posteriori since concept justification is a necessary condi-tion of such knowledge and concept justification depends on experience.

5Timothy Williamson also has misgivings about the a priori–a posteriori distinction. Hismisgivings arise within a broader investigation into the methodology and subjectmatter of philosophy. In particular, he argues against the view that philosophy involves


a distinctive subject matter such as conceptual, linguistic, or analytic truths. Moreover,Williamson denies that philosophical investigation involves a distinctive cognitivefaculty. The metaphysical modalities represent one significant domain of philosophicalinvestigation. Here he maintains that knowledge of the metaphysical modalities isreducible to knowledge of counterfactual conditionals, and offers an account of suchknowledge in terms of the exercise of the imagination. Williamson’s (2007, 165)misgivings about the a priori–a posteriori distinction are based in the observation that“in our imagination-based knowledge of counterfactuals, sense experience can play arole that is neither strictly evidential nor purely enabling.”

Since Williamson’s misgivings are based on his account of knowledge of themetaphysical modalities, one strategy for rejecting them is to reject his account ofknowledge of the metaphysical modalities. There are two basic ways to reject thataccount. The first is to deny that knowledge of the metaphysical modalities is reducibleto knowledge of counterfactual conditionals. The second is to deny his imagination-based account of knowledge of counterfactuals. My goal is not to assess the cogency ofhis account of knowledge of the metaphysical modalities.11 My goal is to assess theimplications of his account, if cogent, for the a priori–a posteriori distinction. Myprimary contention is that he overestimates the implications of his account and that thereason he does so is that he invokes a conception of a priori knowledge that does notcohere well with his background theory of knowledge.

The structure of Williamson’s argument is straightforward. He begins (2007, 165) bydistinguishing two roles that experience can play in the acquisition of knowledge,enabling and evidential, and maintains that a priori knowledge “is supposed to beincompatible with an evidential role for experience . . . [but] supposed to be compatiblewith an enabling role for experience.” According to the tradition, most, if not all,propositions known a priori are necessary truths. But Williamson maintains thatknowledge of necessary truths is reducible to knowledge of counterfactuals and, onhis account of knowledge of counterfactuals, experience can play a role that is neitherpurely enabling nor strictly evidential. Suppose that S knows that p and experienceplays a role that is neither purely enabling nor strictly evidential. Williamson contendsthat S’s knowledge that p is not happily classified either as a priori or a posteriori. Heconcludes (2007, 169) that the a priori–a posteriori distinction “is handy enough for arough initial description of epistemic phenomena; it is out of place in a deepertheoretical analysis, because it obscures more significant epistemic patterns.”Williamson’s argument can be summarized as follows:

(W1) Knowledge of necessary truths is reducible to knowledge of counterfactuals.(W2) Experience can play a role in knowledge of counterfactuals that is neither

purely enabling nor strictly evidential.(W3) In such cases, the resulting knowledge is not happily classified either as a

priori or a posteriori.

266 ALBERT CASULLO

(W4) Therefore, the distinction between a priori and a posteriori knowledge is notuseful for deep theoretical analysis.

Significant questions can be raised about the premises of the argument as well as itsvalidity. As I indicated earlier, one can question both Williamson’s account of know-ledge of necessary truths and his account of knowledge of counterfactuals. I propose togrant both. One can also question Williamson’s understanding of the distinctionbetween purely enabling and strictly evidential roles, and whether the cases in whichhe alleges that experience plays neither role are convincing. Once again, I propose togrant his understanding of the distinction and his verdict on the cases. Still one mightmaintain that even if there are such cases, it does not follow that the distinctionbetween a priori and a posteriori knowledge is not theoretically significant since suchcases are few or inconsequential. Alternatively, one might maintain that such cases areborderline and that the existence of borderline cases is not sufficient to challenge thecogency or importance of a distinction. Once again, I will not pursue those responses.I will grant the centrality and importance of his cases. My focus will be on (W3) since itis the premise that bears directly on the a priori–a posteriori distinction.In order to assess (W3), we must address two questions. What is the role of experi-

ence in those cases in which it is alleged that it is neither purely enabling nor strictlyevidential? Why are such cases not happily classified as either a priori or a posteriori? Inorder to answer the first question, let us focus on Williamson’s central example.Consider a person who learns the words “inch” and “centimeter” independently ofone another by learning to make reliable naked eye judgments of distances. Supposethat such a person visually judges, for example, that two marks are at most twoinches apart. Since the judgment is sufficiently reliable, the person knows a posteriorithat the two marks are at most two inches apart. Williamson (2007, 166) contends thatthe person can employ their capacity to judge distances visually offline to make thefollowing counterfactual judgment:

(25) If two marks had been nine inches apart, they would have been at leastnineteen centimeters apart.

Here the person visually imagines two marks nine inches apart and employs theircapacity to judge distances in centimeters visually offline to judge that they are at leastnineteen centimeters apart. Since the person’s judgment is sufficiently reliable, thatperson knows (25).The role of experience in this case is not evidential, according to Williamson,

because the judgment is not based on memories of having visually encountered similardistances in the past and it is not deduced from general principles inductively orabductively inferred from past experiences. Moreover, it does not play a purelyenabling role since the experiences necessary to reliably evaluate (25) go beyond theexperiences necessary to acquire the concepts involved in (25). Williamson articulatesits role as follows:


I know (25) only if my offline application of the concepts of an inch and a centimeter wassufficiently skilful. Whether I am justified in believing (25) likewise depends on how skilfulI am in making such judgments. My possession of the appropriate skills depends constitutively,not just causally, on past experience for the calibration of my judgments of length in those units.(2007, 166)

There are three central claims about the role of experience in this passage:

(C1) One knows (justifiably believes) (25) only if one’s offline application of therelevant concepts is sufficiently skillful.

(C2) One’s possession of the appropriate skills depends constitutively, not justcausally, on past experience.

(C3) The experiences necessary to skillfully apply the relevant concepts go beyondthe experiences necessary to acquire those concepts.

(C1)–(C3) introduce an epistemic role for experience that is not evidential: experienceis necessary for the skillful application of concepts.

We are now faced with the following question:

(Q) If S knows (justifiably believes) that p and S’s knowledge (justified belief) that pdepends (epistemically) on experience for the skillful application of concepts butnot for evidence, does S know (justifiably believe) that p a priori or a posteriori?

Premise (W3) contends that neither response to (Q) is satisfactory. Each response faces asignificant threat. If one maintains that knowledge of (25) is a priori because experiencedoes not play an evidential role, there is the threat that too much will count as a priori. Ifone maintains that knowledge of (25) is a posteriori because experience plays more than apurely enabling role, there is the threat that knowledge of many philosophically signifi-cant modal truths will turn out to be a posteriori. Therefore, Williamson’s case againstthe significance of the a priori–a posteriori distinction turns on whether both threats aregenuine. My contention is that, although the first is genuine, the second is not.With respect to the first response, Williamson (2007, 167) maintains that “Long

forgotten experience can mold my judgment in many ways without playing a directlyevidential role” and, as a consequence, our knowledge of (25) may be quite similar toour knowledge of (26):

(26) If two marks had been nine inches apart, they would have been further apartthan the front and back legs of an ant.

One can know (26), according to Williamson (2007, 167), without having any evidencebased on sense experience: “The ability to imagine accurately what an ant would looklike next to two marks nine inches apart suffices.” Knowledge of (26), acquired in thismanner, is clearly a posteriori, but the first response would classify it as a priori.Even if one has reservations about Williamson’s example, my earlier discussion of

Hawthorne underscores the threat faced by the first option. Williamson’s example

268 ALBERT CASULLO

appeals to the role of forgotten experience in molding judgment. Hawthorne’sdiscussion highlights the role of forgotten evidence. He introduces the example of astudent who learns a scientific law on the basis of the testimony of a teacher, and laterrecalls that law but not the testimonial evidence on which his belief was originallybased. Here we argued that whether the student knows the law at the later timedepends epistemically on his original evidence even if it is forgotten. If the studentoriginally learned that law from a competent teacher, he knows the law at the latertime. But if the student originally learned that law from an incompetent teacher, hedoes not know the law at the later time. We also argued that, in the case of the studentwho knows the law, if the student’s original (but now forgotten) evidence is based onexperience, then the student’s later knowledge is properly classified as a posteriori. Thefirst response, however, would classify it as a priori. Therefore, the threat facing the firstresponse is genuine.With respect to the second response, Williamson invites us to consider three

propositions:

(27) It is necessary that whoever knows something believes it.(28) If Mary knew that it was raining, she would believe that it was raining.(29) Whoever knew something believed it.

Here he maintains that

The experiences through which we learned to distinguish in practice between belief and non-belief and between knowledge and ignorance play no strictly evidential role in our knowledge of(27)–(29). Nevertheless, their role may be more than purely enabling . . . . Why should not subtledifferences between two courses of experience, each of which sufficed for coming to understand“know” and “believe,”make for differences in how test cases are processed, just large enough totip honest judgments in opposite directions? (2007, 168)

If this account of the role of experience in our knowledge of (27)–(29) is correct,should we draw the conclusion that the knowledge in question is a posteriori?Williamson contends:

Not if that suggests that (27)–(29) are inductive or abductive conclusions from perceptual data. Insuch cases, the question “A priori or a posteriori?” is too crude to be of much epistemological use.(2007, 169)

Therefore, knowledge of (25) is not happily classified as a posteriori because, if it is soclassified, then knowledge of (27)–(29) would also be properly classified as a posterioribut, contends Williamson, such knowledge is not properly classified as a posteriori.Williamson’s argument against the significance of the a priori–a posteriori distinction

turns on two basic ideas. First, a background theory of knowledge (and justification)that introduces an epistemic role for experience that is nonevidential. One knows (orjustifiably believes) (25)–(29) only if one can skillfully apply the relevant concepts. Butwhether one can skillfully apply the concepts depends constitutively on one’s past


experiences. Second, a conception of a priori knowledge on which such knowledge isincompatible with reliance on experiential evidence. The background theory opens upthe possibility of knowledge (and justified belief) that depends epistemically on experi-ence but not on experiential evidence. And we are faced with the question: Is suchknowledge a priori or a posteriori?

This question is left open by Williamson’s articulation of the concept of a prioriknowledge because it consists solely of a necessary condition in terms of the role ofexperiential evidence, but his background theory of knowledge denies that justificationis exclusively a function of one’s evidence. Hence, his articulation of the concept of apriori knowledge does not cohere well with his background theory of knowledge. Thetraditional conception of a priori knowledge, as I have articulated it, coheres betterwith Williamson’s background theory of knowledge since that conception does notpresuppose that justification is exclusively a function of one’s evidence. It maintainsthat a priori knowledge is incompatible with a justificatory role for experience, but itdoes not restrict the justificatory role of experience to experiential evidence. If wereplace Williamson’s articulation of the concept of a priori knowledge with thetraditional conception, it follows straightforwardly, that if S knows (justifiably believes)that p only if S can skillfully apply the concepts in p and S’s skillful application of thoseconcepts depends constitutively on S’s past experience, then S knows (justifiablybelieves) a posteriori that p.

Williamson (2007, 169) maintains that classifying knowledge of (27)–(29) as aposteriori is unacceptable “if that suggests that (27)–(29) are inductive or abductiveconclusions from perceptual data.” The traditional conception of a posteriori justifica-tion neither entails nor suggests that if S knows a posteriori that p, then S knows that pon the basis of an inductive or abductive inference from perceptual data. The trad-itional conception of a posteriori justification, like the traditional conception of a priorijustification, is theory-neutral. It does not entail or suggest that one’s justification isexclusively a function of one’s evidence. The suggestion that if S knows a posteriorithat p, then S knows that p on the basis of an inductive or abductive inference fromperceptual data derives from Williamson’s articulation of the a priori–a posterioridistinction solely in terms of the role of experiential evidence. Therefore, his argumentagainst the second option fails. The threat facing that option is merely apparent.

Williamson’s argument fails because (W3) is false. (W3) is false because knowledgeof (25) is happily classified as a posteriori. The alleged threat to classifying knowledge of(25) as a posteriori is merely apparent. Williamson maintained that knowledge of (25) isnot happily classified as a posteriori because, if it is so classified, then knowledge of(27)–(29) would also be properly classified as a posteriori. But such knowledge is notproperly classified as a posteriori. I maintained, in response, that the argument that heoffers to show that knowledge of (27)–(29) is not properly classified as a posterioridepends on a conception of a priori knowledge that does not cohere well withhis background theory of knowledge. Once that conception is replaced by thetraditional conception of a priori knowledge, the threat that he envisages evaporates.

270 ALBERT CASULLO

There is, however, a residual concern with this response. One might argue that even ifWilliamson’s supporting argument fails, he is correct to insist that knowledge of (27)–(29)is not properly classified as a posteriori. Therefore, if the traditional conception of a prioriknowledge classifies such knowledge as a posteriori, it should also be rejected.This concern is misplaced. The traditional conception of a priori knowledge

classifies knowledge of (27)–(29) as a posteriori only if Williamson is right about therole of experience in our knowledge of (27)–(29). I conceded his account of the role ofexperience in such knowledge in order to evaluate its implications. But anyone whothinks that knowledge of (27)–(29) is properly classified as a priori will reject his accountof the role of experience in such knowledge. Most likely, such a person will reject hisclaim that the experiences through which we learned to distinguish between belief andnonbelief and between knowledge and ignorance are more than purely enabling, andinsist that the conditions for concept possession that he endorses are too lax.The traditional conception of a priori knowledge is not vulnerable to the argument

that Williamson offers in support of (W3). Therefore, his argument fails to show thatthe traditional a priori–a posteriori distinction is not useful for deep theoretical analysis.Nevertheless, even if his supporting argument fails, the charge may be accurate. So thefinal issue we must address is whether the traditional version of the distinction is opento that charge. Williamson alleges that the distinction is not useful because it obscuresmore significant epistemic patterns. His conception of a priori knowledge in terms ofindependence from experiential evidence is open to that charge since it obscures thefact that a belief can depend epistemically on experience in a nonevidential way—thatis, it can depend on experience for the skillful application of concepts. But thetraditional conception of a priori knowledge does not suffer from this shortcoming;it articulates that concept in terms of nonexperiential justification. Since Williamsonmaintains that conceptual skill is a necessary condition of justified belief, the traditionalconception does not obscure this form of epistemic dependence on experience. So weare left with the question: What significant epistemic pattern does it obscure?The a priori–a posteriori distinction has come under attack in the recent literature.

I have surveyed the challenges of the leading critics and provided answers to them.I conclude that these critics have not provided any compelling reason to deny eitherthe cogency or the significance of the distinction.12

Notes1. This is a shorter version of a paper with the same title that originally appeared in Casullo

(2012a). The primary difference between the two versions is articulated in note 12.2. For a fuller discussion, see Casullo (1988, 2003, 2009).3. Kitcher argues that (c) is supported by (b) and that (b) is supported by the intuitive idea that a

priori knowledge is independent of experience:

But if alternative experiences could undermine one’s knowledge then there are features ofone’s current experience which are relevant to the knowledge, namely those features whose


absence would change the current experience into the subversive experience. The idea of thesupport lent by kindly experience is the obverse of the idea of the defeat brought byuncooperative experience. (1983, 89)

His account of the relationship between supporting and defeating evidence, however, is notsupported by his background theory of warrant. It is uncontroversial that if S’s belief that p issupported (i.e., warranted) by experience then S’s belief that p is not warranted (and, hence,not known) a priori. But suppose that S’s belief that p is warranted nonexperientially and thatS’s nonexperiential warrant for the belief that p is defeasible by experience. It does notfollow that S’s belief that p is supported (i.e., warranted) by experience. Kitcher nowconcedes this point and agrees that his original defense of (b) fails. This argument resurfacesin Hawthorne (2007), which is discussed in Section 3.

4. For a discussion of his remaining arguments, see Casullo (2009).5. William Alston offers the following characterization of the traditional concept of know-

ledge:Descartes, along with many other seventeenth- and eighteenth-century philosophers, took itthat any knowledge worthy of the name would be based on cognitions the truth of which isguaranteed (infallible), thatweremaximally stable, immune to ever being shown to bemistaken(incorrigible), and concerning which no reasonable doubt could be raised (indubitable).(1992, 146)

6. Hawthorne acknowledges that a natural reaction to his concern is to distinguish between casesin which the presence of an experience is epistemologically relevant as opposed to cases inwhich the absence of an experience is epistemologically relevant. He replies, however, that

There are a variety of tricky questions in the vicinity here. Can omissions as well as positiveevents count as part of a process? If so, should the presence of an experiential omission in aprocess count as experience involving the relevant sense? (2007, 210)

Suppose that we grant that the answer to the first question is affirmative. An affirmativeanswer to the second has the consequence that all knowledge is testimonial (since, presum-ably, all knowledge can be destroyed by appropriate testimonial evidence). That result byitself should indicate that, whatever the correct metaphysics of omissions or absences, theanswer to the epistemological question is clear.

7. Hawthorne’s argument is essentially the same as the argument that Kitcher offers in supportof condition (b) in his analysis of the concept of a priori warrant. See note 3.

8. Tyler Burge (1993), for example, argues for an affirmative answer.9. Hawthorne seems to think that if a supporter of the distinction between a priori and a

posteriori knowledge opts for Long over Short her position is compromised, since know-ledge of mathematical principles acquired via earlier training and preserved by memory is aposteriori. But why is this a problem? If testimony is an experiential source of knowledge,then knowledge based on testimony is properly classified as a posteriori. The suggestionseems to be that if this is the case, then much of our mathematical knowledge may turn outto be a posteriori. But how does this support the claim that the a priori–a posterioridistinction is not deep?

10. Chisholm (1977, 39) maintains that it is a distinguishing characteristic of intuitive induc-tion—the process of examining concepts that he alleges is the source of immediate a priori

272 ALBERT CASULLO

knowledge—that the manner in which one acquires the requisite concepts is irrelevant tothe epistemic status of beliefs acquired by this process.

11. See Casullo (2012b) for such an assessment.12. I do not maintain that the distinction is invulnerable to attack. In the longer version of the

paper (in Casullo 2012a), I articulate a challenge which leads to the rejection of thetraditional view that all knowledge (or justified belief) is either a priori or a posteriori. InCasullo 2003, I offer a more radical challenge to the distinction that turns on the questionwhether the distinction between experiential and nonexperiential sources of justification canbe coherently articulated.

ReferencesAlston, William. 1992. “Foundationalism.” In A Companion to Epistemology, ed. J. Dancy andE. Sosa. Oxford: Blackwell.

Burge, Tyler. 1993. “Content Preservation.” Philosophical Review 102: 457–88.Casullo, Albert. 1988. “Revisability, Reliabilism, and A Priori Knowledge.” Philosophy andPhenomenological Research 49: 187–213.

Casullo, Albert. 2003. A Priori Justification. New York: Oxford University Press.Casullo, Albert. 2009. “Analyzing A Priori Knowledge.” Philosophical Studies 142: 77–90.Casullo, Albert. 2012a. Essays on A Priori Knowledge and Justification. New York: OxfordUniversity Press.

Casullo, Albert. 2012b. “Counterfactuals and Modal Knowledge.” In Casullo 2012a.Chisholm, R. M. 1977. Theory of Knowledge, 2nd ed. Englewood Cliffs: Prentice-Hall, Inc.Hawthorne, J. 2007. “A Priority and Externalism.” In Internalism and Externalism in Semantics andEpistemology, ed. Sanford Goldberg. Oxford: Oxford University Press.

Jenkins, C. S. 2008. Grounding Concepts. Oxford: Oxford University Press.Kant, Immanuel. 1965. Critique of Pure Reason. Trans. Norman Kemp Smith. New York: St.

Martin’s Press.Kitcher, Philip. 1983. The Nature of Mathematical Knowledge. New York: Oxford University Press.Kitcher, Philip. 2000. “A Priori Knowledge Revisited.” In New Essays on the A Priori, ed.P. Boghossian and C. Peacocke. Oxford: Oxford University Press.

Moser, Paul. 1987. “Introduction.” In A Priori Knowledge, ed. P. Moser. Oxford: OxfordUniversity Press.

Moser, Paul. 1989. Knowledge and Evidence. Cambridge: Cambridge University Press.Williamson, Timothy. 2007. The Philosophy of Philosophy. Oxford: Blackwell.


12

Naturalistic Challenges to theA Priori*

C. S. I. Jenkins

1 IntroductionIn Jenkins 2008a, I described a form of a priorism which was designed to be consistentwith—indeed, was motivated by—a certain kind of naturalism. The kind of naturalismI had in mind is the view that ‘the world is the way that best science says it is (wherebest science need not be identical to any actual science, whether present or future), andthere is nothing more to be truly said about the world’. (This statement comes fromJenkins 2007, p. 259, not from Jenkins 2008a, where I was less explicit than I shouldhave been about what I meant by ‘naturalism’.)

However, there are many different varieties of ‘naturalism’ around, and variousdifferent reasons why certain of them might be considered inimical to a priorism. Thepurpose of this chapter is to examine a few of these naturalistic challenges to the apriori, and explore the prospects for the kind of a priorism described in Jenkins 2008ain the light of them. My motivation for doing this is not based solely on my interest inthe prospects for the (2008a) view qua view developed by me, however; I think the apriorism of Jenkins (2008a) may well be a priorism’s best shot (at least, for thenaturalistically inclined). If that’s right, what I’m investigating in this chapter is whetherthe a priori’s best shot is good enough—good enough, that is, to withstand a representa-tive range of challenges from ‘naturalism’.

I shall be aiming to show that the challenges considered here can be resisted withoutsimply rejecting the naturalistic underpinnings that give rise to them. In each case,I shall try to be as concessive as possible to the underlying naturalistic ideas in thevicinity of the challenges, and argue that these naturalistic starting points do not give usgood enough grounds either to reject or to downplay the a priori.

I shall begin, in }2, with a brief classificatory sketch of some of the philosophicaltheses (and other things) that can attract the label ‘naturalism’. This is not an attempt at

a comprehensive survey; rather, the section has two aims. First, I want to make clearhow radically unwise it would be to imagine there is just one thing that can sensibly bedescribed as the naturalistic challenge to the a priori. There are in fact several clusters ofarguments and objections, with various aims, motivated in various ways by variousforms of ‘naturalism’. Second, I want to set up a structured background against whichI can locate the naturalistic views of the authors discussed in the later sections, so thatthese naturalisms can be clearly identified and their differences understood.In the remainder of the chapter, }}3–6, I shall discuss a selection of prominent

naturalistic challenges to the a priori arising from the work of Quine, Maddy, andPapineau. Some other interesting naturalistic challenges to the a priori (or, relatedly, to‘intuition’) can be found in the work of Devitt (e.g. 2005) and Kornblith (e.g. 2007).I respond elsewhere on behalf of my kind of a priorist to Devitt (Jenkins forthcoming)and to Kornblith (Jenkins MS).One thing to note before I get started is that when I call some knowledge or

justification ‘a priori’, I mean that it is epistemically independent of empirical evidence.This does not entail that it is empirically indefeasible.1 Similarly, it is not to be assumedthat a priori knowledge or justification as I understand them are related in anyparticularly interesting way to infallibility, incorrigibility, or indubitability. Nor shouldit be assumed that a priori knowledge/justification is always of/for necessary truths, oralways of/for analytic truths.Some of the authors with whom I engage in this chapter may be best interpreted as

having some different notion(s) of a prioricity in mind. My comments, then, are to beinterpreted as an investigation of whether the challenges they mount against a priorismas they understand it have any force when considered/reframed as challenges to apriorism as I understand it. I do this in the sprit of trying to give a priorism its bestchance of success in the prima facie hostile environment of naturalism.

2 ‘Naturalism’

The term ‘naturalism’ covers a multitude of virtues and a few sins. In this section I try toconvey some of the diversity in contemporary philosophical usage of this term. For easeof expression, I shall assume throughout this chapter that ‘naturalism’ is sometimescorrectly used to refer to fairly specific positions (and stances, etc.), but that it also hasa semantically general use on which these various specific things can all be trulydescribed as ‘kinds of naturalism’. Nothing much hangs on this terminological choiceexcept ease of non-metalinguistic expression.Most (perhaps all) naturalisms have something to do with natural science. But it is hard

to identify much by way of more substantive commonalities. For example, the relevantkind(s) of natural science may be more or less narrowly delimited, and may be actual,possible, or ideal. And science may be playing very different roles in relation to thedifferent forms of naturalism: for example, natural science may play a role in the

NATURALISTIC CHALLENGES TO THE A PRIORI 275

characterization of some forms of naturalism, may be appealed to in the motivation forothers, and may be only of historical relevance for others.

An initial point to note is that while ‘naturalism’ often refers to a doctrine, it does notalways do so. Sometimes, for example, ‘naturalism’ can refer to a stance, an approach, amethodology or an injunction (e.g. an injunction to use a certain kind of method-ology). Maddy, for example, characterizes naturalism as ‘an approach’, namely that of‘working within science to understand, clarify and improve science’ (2000, p. 108).Sometimes naturalism is characterized by an attitude of ‘respect for the findings ofscience, and for the progress that has been made by science’ (Nolan 2005, p. 10). Inother senses of ‘naturalism’, mere failures to do something may be sufficient for being anaturalist. For instance, it seems that according to Sider (2011, pp. 292–3) the mereabsence of certain notions from one’s ideology suffices for a kind of naturalism.Even when we focus on the doctrines, there are very important differences between

different naturalisms. One relatively commonly drawn distinction (see e.g. Papineau2007) is that between ontological and methodological naturalism. Ontological natur-alism says (approximately) that only natural things—that is, the kinds of things positedby the natural sciences—exist. Methodological naturalism says (approximately) thatphilosophical enquiry is to be conducted in the manner of natural science. As far ascontemporary analytic philosophy is concerned, these certainly have been two of themore significant kinds of naturalism. But there is a lot more variety out there than iscaptured by this distinction alone.

Doctrine naturalisms can be usefully divided into the first-order philosophical andthe metaphilosophical. Metaphilosophical naturalistic theses are theses about philoso-phy; first-order naturalistic theses are philosophical theses but are not about philosophy(at least, not in particular). The first-order philosophical naturalisms include quitegeneral views in ontology, such as the aforementioned view that only natural thingsexist. There are also more subdiscipline-specific ontological naturalisms, such as theview that ‘the ontology of mathematics is the mathematical ontology of our bestnatural science’ (Paseau 2008, }6.1). But there are also views in metaphysics morebroadly (i.e. not merely ontological views) that are called ‘naturalism’, such as ‘thedoctrine that reality consists of nothing but a single all-embracing spatio-temporalsystem’ (Armstrong 1978, p. 261), the view that ‘the world is the way that best sciencesays it is’ (Jenkins 2007, p. 259), and the view that ‘all facts are natural facts’ (Lenman2006, }1). Insofar as there is (or could be, or could be thought to be) more to reality,the world and/or the facts than ontology alone, these more broadly metaphysicalnaturalisms should be distinguished from the narrower ontological kind.

The word ‘naturalism’ is also associated with a wide variety of other first-orderviews, some as different from the foregoing metaphysical naturalisms as the view (at theintersection of ethics and the philosophy of language) that ‘moral words have the samemeaning as certain combinations of non-moral expressions’ (Hare 1995, p. 340) or theview (in the philosophy of religion) that ‘science, properly so-called, cannot involve

276 C. S. I . JENKINS

religious belief or commitment’ (Plantinga 1996, p. 178). There are all sorts of first-order naturalisms that aren’t squarely ontological or even metaphysical.For our purposes, a particularly important tranche of such naturalisms consists of the

first-order epistemological naturalisms. These include the view that ‘[i]t is rational tobelieve (only) our best scientific theories’ (Colyvan 2009, }4), the view that ‘all epistemicfacts are natural facts’ (Feldman 2001, }4), and the view that there is no a prioriknowledge (Devitt 2005). These, I think, are better classified as epistemological thanmethodological (as in the more familiar ontological–methodological distinction), sincethey most directly concern epistemological notions (knowledge, justification, andrationality) rather thanmethod. First-order methodological naturalisms I take to includethe view that ‘themode of inquiry typical of the physical sciences will provide theoreticalunderstanding of theworld, to the extent that this sort of understanding can be achieved’(Stoljar 2009, }12). Here the focus is directly on method or ‘mode of inquiry’.The metaphilosophical naturalisms can be helpfully subdivided into two bundles,

the descriptive and the normative. Descriptive metaphilosophical naturalisms includethe view that philosophy is literally and already part of the scientific enterprise, as wellas views to the effect that the enterprise of philosophy is in some subtler way continuouswith that of science. Some descriptive metaphilosophical naturalisms are specificallymethodological, such as the view that ‘at bottom philosophy and science have just thesame aims and methods, namely, to establish synthetic knowledge about the naturalworld, in particular knowledge of laws and causal mechanisms, and to achieve this bycomparing synthetic theories with the empirical data’ (Papineau 2007, }2.1).Lastly for the purposes of this non-comprehensive survey, there are the normative

metaphilosophical naturalisms, such as the view that philosophy should be part of thescientific enterprise (whether or not it already is). Some normative metaphilosophicalnaturalisms are methodological too. For example, the view that philosophers shoulddefer to scientists (or some scientists) on certain issues (see e.g. Maddy 1997), and theview that conceptual analysis, the practice of consulting intuitions, and the like are notgood methods for the pursuit of philosophy (see e.g. Kornblith 2007), both fall in thenormative, metaphilosophical, methodological category.The focus of this chapter is on ways in which a variety of naturalisms could be used

to mount challenges to the existence and/or philosophical significance of a prioriknowledge or justification. Most obviously, some of the first-order epistemologicalnaturalisms will rule out a prioricity. Indeed, some forms of first-order epistemologicalnaturalism amount simply to a denial of the existence of a priori knowledge and/orjustification. But these kinds of naturalism cannot really be said to motivate a challengeto the a priori. (They might themselves be motivated by the adoption of other, related,naturalisms, however. Certain of the views expressed by Devitt in his (2005) andelsewhere might be regarded in this light; see Jenkins forthcoming for discussion.) It isalso important to be clear that not all first-order epistemological naturalisms needmotivate or involve any anti-a priorist leanings. For example, one could quite com-fortably adopt the naturalism according to which ‘it is rational to believe (only) our best

NATURALIST IC CHALLENGES TO THE A PRIORI 277

scientific theories’ (Colyvan 2009, }4) alongside the view that mathematical theoriesare both scientific and a priori knowable.

Let me conclude this preliminary section by noting that, in addition to variation inthe range of kinds of naturalism that might motivate a challenge to the a priori, thereare important differences in the nature of the challenges that can be so motivated. One keydifference is between the deniers and the belittlers. Some naturalists argue that weshould deny that there is any a priori knowledge or justification. Others acknowledgeits existence but argue that it is trivial or insignificant, or cannot do the kind of work itmight have been expected to do. I shall explore challenges of each of these kinds in theremaining sections of this chapter. I discuss Quine and Maddy first (deniers), thenPapineau (a belittler).

3 QuineMany people’s prototype of a naturalistic challenge to the a priori is (some version of)Quine’s. In this section, I shall reconstruct an argument that there is no a prioriknowledge based on views that Quine expresses in his (1951). I should stress that thisreconstruction is not supposed to be uncontroversial Quine exegesis; it is rather anattempt to (re-)interpret a Quinean naturalistic objection to the a priori in a way thatdoes not succumb to certain well-rehearsed objections (e.g. those of Grice andStrawson 1956).2 Let’s call the defender of this argument Quine*, without prejudiceas to whether Quine* = Quine.

Quine*’s argument goes as follows:

1. (Premise) There is no ‘special rational faculty’ that could give rise to a prioriknowledge.

2. (From 1) Anything knowable a priori would have to be analytic.3. (Premise) The notion of analyticity assumes a smaller unit of meaning (and

sameness-of-meaning) than a total theory.4. (Premise) There are no smaller units of empirical confirmation than total theories.5. (Premise)The units of empirical confirmation and the units ofmeaning are the same.6. (From 4 and 5) There are no smaller units of meaning than total theories.7. (From 2, 3, and 6) Nothing is a priori knowable.

Naturalism of some kind is supposed (by Quine*) to motivate premise 1. This premisecould be motivated, for example, by the view that we are justified in believing onlywhat we have scientific evidence for (together with the view that there is no scientificevidence for any special rational faculty giving rise to a priori knowledge). In terms ofthe survey offered in }2 above, this motivating naturalism would be a form of first-order epistemological naturalism. Alternatively, premise 1 could be motivated by theview that knowledge is only achieved through scientific methods, and scientificmethods do not include the use of any special rational faculty giving rise to a prioriknowledge. This is a form of first-order methodological naturalism.


Interim conclusion 2 does not follow from premise 1 without some further work,but one way of trying to make this step look reasonable is to argue that since we needexperience to gain knowledge of anything which is independent of us (a view whichmight itself be motivated by or identical to some kind of naturalism), the only way tohave a priori knowledge is for it to be knowledge of something that depends on us. Thenone can argue that analytic truths are made true by meaning, and hence dependent onus in the kind of way that would make them suitable to be known a priori. One needs afurther assumption to the effect that there are no other relevant ways in which a truthcould depend on us and hence be a priori knowable, but maybe this can be motivatedby considering that other kinds of knowledge of truths which depend on us don’t seemto be well-described as ‘a priori’; knowledge of what one is thinking or feeling at themoment, for example, seems different in relevant respects from the typical examples ofputatively a priori knowledge (knowledge of arithmetic, knowledge of logic, etc.).Premise 3 derives from the thought that in order to hold that (say) the sentence

‘Bachelors are unmarried’ is analytic, one must be prepared to make claims about themeaning of ‘bachelor’ and/or ‘unmarried’ (e.g. that ‘bachelor’ means the same as‘unmarried man’), and one can only do this if one allows that these subsentential-sized units (words and phrases) have meanings. (The existence of subsentential mean-ings may not be entailed by the analyticity claim; perhaps one could try to argue that it isin virtue of the meaning of the whole sentence, and nothing smaller, that the sentenceis true. But I’m not sure how one would go about doing that, and certainly it isn’t thenatural way to think about how analyticity works.)Premise 4 is Quinean confirmational holism; the arguments for and against that are

pretty well known and well discussed, and I won’t rehearse any here. Premise 5 ismotivated by the thought that meaning cannot be allowed to be too radically divorcedfrom experience. This same underlying thought is often associated with verificationism(see e.g. Schlick 1936: ‘ . . . there is no way of understanding any meaning withoutultimate reference to ostensive definitions and this means, in an obvious sense,reference to “experience” or “possibility of verification”’), which itself is stronglyassociated with naturalism of some kind: members of the Vienna Circle were naturalistsat least in the sense that they adopted a stance of respect for (that which they consideredto be genuine) science, and some were inclined to regard philosophy as at leastcontinuous with (and perhaps a branch of) science. Verificationism was stronglyassociated with the rejection of certain areas of metaphysics on these sorts of grounds.(See Uebel 2006 for an overview of the Vienna Circle’s history and views.)So we also can with some plausibility think of premise 5 as motivated (perhaps

indirectly) by some kind of naturalism, a cousin of the kind(s) motivating the verifica-tionists. Interim conclusion 6 follows validly from 4 and 5, and the main conclusion 7follows validly from 2, 3, and 6.

The Quine*an argument is open to criticism of various kinds at various points, muchof which I would be sympathetic to. But here I want to look at criticisms of it whichclearly do not seek to challenge any of the motivating naturalisms. The most obvious


way to do this is to go after premise 3 and/or premise 4, since it is premises 1 and 5 (andthe questionable step from 1 to 2) that are most plausibly motivated by naturalisticconsiderations. I will focus on premise 4 here; as mentioned previously, I’m not surehow I’d go about challenging premise 3.

There are some very familiar ways of resisting confirmational holism (see e.g. Maddy1992; Sober 1993), but I’mnot yet convinced by any of these and in any case don’t wantmerely to churn up well-trodden ground here. Instead, let me go so far as to grant toQuine that propositions cannot be confirmed by experience one at a time, but only as partsof whole theories. (If there is any way of arguing that this holistic view is also naturalism-motivated, then by making this move I am able to respect an additional element inQuine’s naturalistic motivation.) It remains a possibility—or so I have argued—thatthere is such a thing as empirical confirmation at the sub-propositional level. Confirm-ation of this kind is not (at least, not in the first instance) the kind of thing that conferspositive epistemic status on any propositional belief, but rather the kind of thing thatrenders sub-propositional mental representations, or concepts, epistemically respectable.

The epistemic respectability of a concept (or group of concepts) amounts to its beingepistemically respectable for a subject to rely on that concept (or those concepts) whenengaged in conceptual examination and related processes. For example, when weexamine our arithmetical concepts and thereby arrive at arithmetical beliefs, we aretrusting the concepts in question as guides to the nature of arithmetical reality. Weneed those concepts to be trustworthy guides in order for this trust to be appropriate.The suggestion of Jenkins (2008a) is that it is the relationship between concepts andexperience that—in good cases—renders concepts trustworthy. When our conceptsare sensitive to experience (and thus to the world as it affects us through experience),those concepts can become trustworthy guides to the world.

I don’t want to rehash too many of the details of this suggestion here, but it may behelpful to sketch a quick caricature to help the reader see what is intended. The centralidea is (very roughly) that as we get sensory input from the world, this input determinesthe contours of our conceptual scheme: either we acquire concepts in response tosensory input because they are useful for dealing with it, or as sensory input comes inwe filter out such pre-existing concepts as are not useful in dealing with it. (Or, perhaps,both.) The idea is that the concepts rendered useful by sensory input (and henceacquired/retained) are probably the ones whose contours mirror or map out thestructure of the mind-independent world. (Why else would these concepts be theuseful ones?3) Having secured concepts which map the structure of the world, we are atliberty to conduct conceptual examinations (or use our ‘intuitions’,4 or whatever othersimilar concept-guided processes are at our disposal) which can then reveal to us howthings stand in reality because they are working with concepts that map that reality. Forexample, having secured trustworthy concepts of round and square on the basis of myexperience, I can examine those concepts to learn that nothing is a round square.

By the time such conceptual examination has led to my believing some proposition,I have (if everything is going well) secured positive epistemic status for a (proposition-


shaped) belief. But I have achieved this by way of securing subpropositional groundingfor the concepts under examination.The salient point about this sort of view, for current purposes, is that there is nothing

about it that is at all in tension with the naturalisms that motivate the Quine*an argumentunder consideration. Certainly there is no appeal to a special rational faculty; indeed,the whole point of the suggestion is to try to respect a (naturalistically motivated)empiricist view that experience would need to be crucially involved in any process ofsecuring knowledge of the mind-independent world through conceptual examination.The proposal is in this way very different from the more traditional, non-empiricistaccounts of how possessing the right concepts can enable one to secure knowledge (seee.g. Peacocke 2000; Bealer 2000). The naturalistic thought motivating Quine*’spremise 5, that meaning should not be too far divorced from what experience candeliver, is also respected: small (concept-sized) units of meaning are postulated, butalongside small (concept-sized) units of empirical confirmation.5

If we are sometimes able to secure justified belief and/or knowledge through thesesorts of channels, I suggest in my (2008a), then the justification and/or knowledge soattained is properly categorized as empirical. Nevertheless, it will also bear many of thetrademark features of the ‘a priori’ as traditionally conceived. Most importantly, it willnot be dependent on the subject’s empirical evidence, since the role played by experi-ence in grounding our concepts is not an evidential one: evidence is evidence for aproposition, and concepts are sub-propositional.6 For this reason, I have argued that itis appropriate to describe this special kind of justification and knowledge as ‘a priori’,despite the fact that they are also empirical.7

And if this is right, then a commitment to priori justification and knowledge can bemade consistent with the robustly naturalistic theses that motivate the Quine*anargument discussed in this section. This is how I propose to resist, on behalf of the apriorist, this first kind of naturalistic challenge.

4 MaddyThe second challenge comes from Maddy (2000), whose naturalism is presented as ‘adescendant of Quine’s’ (p. 92). Although she is sometimes at pains to characterize it as‘an approach’ rather than ‘a theory’ (p. 108), this naturalism does evidently involvecommitment to some theses: for example, that there is no ‘first philosophy’, that any‘two-level’ approach to enquiry on the model of Carnap’s is a mistake, and that allenquiry, including the kind of philosophical enquiry traditionally thought to beamenable to a priori treatment, is in fact just part of the empirical scientific enterprise.In terms of the classificatory structure in }2, Maddy’s naturalism encompasses first-order methodological and metaphilosophical methodological theses.Maddy clearly takes her naturalism to have first-order epistemological consequences:

she thinks that it leaves no room for any special a priori knowledge. She states that weshould adopt ‘a one-level view that fails to underwrite the a priori–a posteriori


distinction’ (p. 114), and says that although ‘naturalism may doom the time-honourednotion of a priori knowledge’, she is ‘quite willing to accept’ this cost (p. 114).

Since my purpose here is to be as concessive as possible to the kinds of naturalismthat are supposed to motivate challenges to the a priori, I shall grant Maddy all hernaturalistic starting points, and question only the step from these to the rejection of thea priori. Obviously, they leave no room for a priori methods which operate separatelyfrom empirical/scientific enquiry. But we still have the option of adopting a view onwhich the a priori exists and does not operate in any such removed way.

Maddy herself considers two ways in which the a priori might be incorporated into aone-level picture wherein there is no ‘first philosophy’ (nor anything else operatingseparately from empirical/scientific enquiry). First, she considers that ‘even the one-level naturalist can say that an analytic statement is unrevisable short of a change inlanguage, and in that sense, might be considered a priori’ (p. 112). Maddy’s only replyto this is that it wouldn’t deliver a ‘viable’ account of a priori knowledge, by which shemeans one that will render ‘some (seemingly) contentful scientific or mathematicalclaims ( . . . the sort of thing Kant might count as synthetic) as a priori’ (fn. 43). Shethinks this because she ‘sincerely doubt[s] that . . . “continuous manifold” is part of themeaning of the term “space”, or that the axiom of replacement is part of the meaningof the term “set”’.

This strikes me as an unsuccessful reply. For one thing, Maddy’s doubts concerningthe meaning of ‘space’ and ‘set’ may be unfounded—she offers us no reasons to sharethem—and for another they form an inadequate induction base for her general claimthat no ‘seemingly contentful’ scientific or mathematical claim is analytic in the relevantsense. Moreover, even if that general claim is right, there could be a substantial andinteresting role for the a priori in delivering propositions which are not ‘seeminglycontentful’ or ‘the sort of thing Kant might count as synthetic’. Why should thestandard for ‘viability’ in an account of the a priori have anything to do with whatKant might count as synthetic? Maddy doesn’t say. Although she advertises a challengeto there being any a priori knowledge (the a priori is supposed to be potentially‘doomed’ by her naturalism), at best she might here be giving an argument againstthere being any a priori knowledge of propositions that are ‘contentful’ in some to-be-specified way.

However, the kind of a prioricity that Maddy is proposing here consists in a certainkind of unrevisability, and hence isn’t the kind of a prioricity I’m working with (see }1).So the envisaged appeal to the consistency of that kind of unrevisability with a one-level approach of the kind required by Maddy’s naturalism wouldn’t in any caseprovide an acceptable way for me to argue that the kind of a priorism I think is mostpromising is compatible with a ‘one-level’ approach.Maddy’s second effort on behalf of the one-level apriorist is to suggest that a

methodological distinction within science itself, between ‘conventional/pragmatic’aspects of scientific methodology and ‘theoretical/empirical’ aspects, could give riseto an a priori/a posteriori distinction. Not much detail is offered, but if I understand


Maddy correctly the idea is roughly that scientists adopt some hypotheses, such as thehypothesis that space-time is continuous (Maddy’s example), for conventional and/orpragmatic reasons, while they adopt others, such as the hypothesis that atoms exist (alsoMaddy’s example) for theoretical and/or empirical reasons (i.e. because there isevidence for their truth). The one-level apriorist might therefore try arguing thatsome or all of the hypotheses adopted (purely) for conventional and/or pragmaticreasons count as known a priori, while those adopted for theoretical and/or empiricalreasons are known a posteriori (if at all).This option Maddy rejects on the grounds that it is far from clear how a hypothesis

adopted purely for conventional and/or pragmatic reasons could count as being knownat all. I think this is a reasonable concern, and am also somewhat suspicious of therequired methodological distinction in the first place. So I won’t try to defend thissecond proffered route to one-level a priorism either.Instead, I propose that the suggestion sketched in }3 can be understood as being a

one-level a priorism of the requisite kind. If that suggestion is correct, all knowledge,whether a priori or a posteriori, is ultimately empirical. And there is no obvious reasonto regard any of it as on a ‘higher level’ than the rest. There is no obvious reason tothink, for example, that a priori empirical justification cannot be outweighed orundermined by a posteriori empirical evidence for a contrary view. There is no obviousreason to doubt that physicists regularly employ a priori empirical methodologies orthat philosophers regularly employ a posteriori ones. There is no Carnapian commit-ment to ‘external questions’ about (say) ontology, which are only resolvable via themerely pragmatic adoption of some linguistic framework. All questions, whether ornot a priori methods can be used to address them, are to be answered throughappropriate sensitivity to empirical input. This is what Maddy’s naturalism requires,and the kind of apriorist I have in mind can accommodate that demand.

5 PapineauIn this section I shall consider some naturalistic moves made by Papineau in his 2009 indefence of the view that there is no significant role for a priori knowledge in philoso-phy. The naturalism that Papineau expounds is the view that ‘philosophical investi-gation is like scientific investigation’ in three respects (p. 2):

(1) ‘[T]he claims made by philosophy are synthetic, not analytic . . . ’(2) ‘[P]hilosophical knowledge is a posteriori, not a priori: the claims established by

philosophers depend on the same kind of empirical support as scientific theories.’(3) ‘[T]he central questions of philosophy concern actuality rather than necessity:

philosophy is primarily aimed at understanding the actual world studied byscience, not some further realm of metaphysical modality.’

In terms of the }2 classifications, Papineau’s naturalism is metaphilosophical anddescriptive, and although he describes it as ‘methodological’, it is elucidated as consist-


ing of metaphilosophical theses in the philosophy of language, epistemology, andmetaphysics respectively.

The challenge to the a priori that emerges through Papineau’s defence of (2) isclearly different in kind from Quine*’s and Maddy’s challenge. Papineau does notwant to deny that there is a priori knowledge, he just wants to deny that anyphilosophical knowledge is a priori. Trivially, I cannot consistently respect all ofPapineau’s naturalism, defined as (1)–(3), while resisting his anti-a priorist claim (2).I can, however, resist the arguments Papineau offers for (2) without simply rejectingthe naturalistic premises on which his arguments are based.

Before doing that, let me note an important qualification which Papineau makes toclaims (1) –(3). Papineau says that he is—at least for the purposes of his (2009)—targetingonly ‘theoretical philosophy’, which he takes to include metaphysics, epistemology, andthe philosophies of mind and language. He says that he is not (yet) trying to defend(1)–(3) with respect to ‘areas of philosophy that trade in normative claims, or mathemat-ical claims, or logical or modal claims’ (p. 3). This last exclusion is surprising in that itthreatens to render (3) trivially true, but perhaps we can charitably interpret Papineau asintending here to exclude only areas of philosophy that trade in explicitly modal claims.

Even so, the restriction is very puzzling. What are we to make, for instance, of thefact that epistemology—which is ruled in—explicitly and centrally trades in normativeclaims (e.g. about what is justified) as well as in modal claims (e.g. about what isknowable), and hence appears to be simultaneously ruled out? How are we to categor-ize, for example, the central areas of metaphysics and philosophy of language (ruled in)associated with normativity and modality (ruled out)?

Even supposing some reasonable cut-off of the envisaged kind can be drawn, thejustification offered for drawing it is puzzling. Papineau says that the claims of theoret-ical philosophy are ‘more easily interpretable’ (p. 3) and so the task of defending (1)–(3)is easier for them. For what it’s worth, I find the interpretation of many epistemologicaland metaphysical claims extremely taxing, and that of at least some ethical and modalclaims comparatively straightforward. But be all this as it may, I shall here examinePapineau’s arguments for (2) granting him some stable conception of theoreticalphilosophy of the kind he envisages, and, following his usage, I shall refer to theoreticalphilosophy in this sense as ‘philosophy’ for the rest of this section.Papineau rightly acknowledges (p. 20) that it does not follow from the alleged

syntheticity of philosophical claims, i.e. (1), that philosophical knowledge is not apriori. Instead, he offers an argument to the effect that the ‘intuitions’ on whichphilosophy’s putatively a priori methodology supposedly rests in fact stand in need ofa posteriori empirical justification before they can be relied upon.

According to Papineau, philosophical intuitions—by which he means the sorts ofthings we pump with traditional philosophical thought experiments—are claims thatare ‘encapsulated’ in the cognitive mechanisms employed in the thought-experimentalprocesses, such as the mechanisms used in deciding whether or not the subject hasknowledge in a Gettier case. That is to say, these intuitions are a sort of ‘implicit


“assumption”’ (p. 17) which the use of the relevant mechanisms brings into play, andwhich we typically don’t challenge.Papineau then says that encapsulated assumptions in general are epistemologically

shaky; they are better regarding as the mind’s quick-and-dirty rule-of-thumb guide-lines than as anything epistemically trustworthy. He notes that they ‘may lead us astrayin certain cases’ (p. 21). He concludes that we cannot trust them until we have securedproper a posteriori justification for them. Papineau considers that this point isreinforced by recent ‘findings’ in experimental philosophy: he takes reported variationin responses to certain thought-experimental situations to be evidence of the unreliabilityof the relevant encapsulated assumptions. If there is variability, he reasons, somebodyhas to be wrong; so it is possible for intuitions to lead us astray. Hence we must back upall these assumptions with proper a posteriori justification before they can count asjustified (p. 22).We can express Papineau’s main argument as follows:

i. (Premise) Intuitions are encapsulated assumptions.ii. (Premise) Encapsulated assumptions may lead us astray in certain cases.iii. (From i and ii) We cannot trust intuitions until we have a posteriori justification

for them.iv. (From iii) Philosophical knowledge is a posteriori.

I have many concerns about this argument. One is that Papineau does not defendeither premise. The view expressed in premise i is introduced as a ‘way of seeing thematter’ on p. 17, not as a way we should see the matter. In Jenkins (MS), I distinguishtwo major strands in the ways philosophers use the term ‘intuition’, and argue that onlyone of the two is strongly associated with a prioricity. In this sense of intuition,I suggest, intuitions may be understood as the (reasonably reliable) deliverances ofconceptual examination and related processes, to be construed in the way outlined in}3 (and more thoroughly in Jenkins 2008a). This is another way of seeing the matter,and I explain in Jenkins (2008a and MS) what some of its appealing features are. I don’tyet see why I should prefer Papineau’s way.Likewise, premise ii is merely stated, not defended. Papineau says that ‘[t]he function

of [the relevant] cognitive mechanisms . . . is to deliver judgments about particular casesquickly and efficiently’, but no evidence for this claim is offered or indicated. There isjust an analogy with assumptions that are encapsulated in the visual system: we have aquick visual mechanism that always interprets sharp intensity changes as edges, whichcan’t be uniformly relied upon. ‘I would like to say the same about . . . intuitions . . . ’,says Papineau (p. 18). But he doesn’t give readers a reason to join him.

However, I am in the business of granting Papineau as much as possible by way ofstarting points that could be considered naturalistic in spirit. And perhaps it could beargued that premise i is motivated by a kind of naturalism that says we need a scientificunderstanding of everything we believe in, and thus if we are to countenance intuition,we need a scientific understanding of how it works. It’s still a big step from there to


saying that intuitions are encapsulated assumptions, but let’s very concessively imaginesome argument to the effect that all other attempts to make scientifically respectablesense of how intuition works are failing. And let’s very concessively grant that premise iican be motivated as a finding of science and hence something that a naturalisticphilosopher—in the sense of one who respects scientific findings and considers thempotentially relevant to philosophy—should pay attention to.

Even with premises i and ii in place, I find the argument unconvincing. One reasonis that in the step from ii to iii, Papineau seems to be arguing from the possibility oferror to the absence of trustworthiness. Unless we are infallibilists about justification orknowledge, the mere fact that it is possible for us to be led astray in trusting intuition willnot mean that when we trust intuition we acquire no justification for or knowledge ofthat which it delivers. A wide-ranging scepticism threatens if we think that way, sinceall the standard sources of a posteriori justification are pretty fallible too. We are givenno grounds to think that the unreliability of the relevant encapsulated assumptions is ofa bad enough kind to entail that we have no justification for them.

A second problem is that Papineau does not consider whether other sources of apriori justification might be available in philosophy besides what he is calling ‘intu-itions’ (i.e. the deliverances of thought experiments). The methodology known as theCanberra Plan, for example, suggests a role for the a priori in philosophy which is (atleast prima facie) different, but it is not considered by Papineau. (See e.g. Braddon-Mitchell 2009 for more on the Canberra Plan and the a priori.)

A third is that the unreliability of encapsulated assumptions in general does not entailthat the kinds of encapsulated assumptions relied upon in philosophical thoughtexperiments are unreliable. (Compare: the Internet in general is not a reliable sourceof information about contemporary analytic philosophy. But the Stanford Encyclopediaof Philosophy is pretty good.) Also worth noting at this point is the fact that the supportwhich Papineau claims to get from experimental philosophy is shaky at best, since thereare various ways of explaining the relevant variability data (even supposing it to berobust) without attributing any differences in the underlying intuitions. (See e.g. Sosa1998 and 2009, }3.)

Some might hope to reconstruct a different argument for the a posteriority ofintuitions/encapsulated assumptions from Papineau’s two arguments for their synthe-ticity. Papineau himself admits that syntheticity does not entail a prioricity, but maybethere are ways to bridge that gap. The prospects are not good for this kind of project,however. Even if an argument from syntheticity to a posteriority could be mustered,this line of argument leaves open the aforementioned possibility that the a priori couldget in on the act in other ways than via the things Papineau calls ‘intuitions’/encapsu-lated assumptions. And moreover Papineau’s arguments for the syntheticity of theseencapsulated assumptions are themselves questionable, as I shall now argue.

In the first of them, Papineau argues that we should expect cognitive mechanisms toencapsulate synthetic information. Since ‘[t]he whole point of these mechanisms is tostart with limited information and deliver further conclusions’, they would not be fit


for purpose if they were ‘restricted to analytic inferences and precluded from engagingin ampliative ones’ (p. 19).One problem with this argument is that its conclusion is not the one Papineau needs.

In order to have a hope of justifying (2) by these means, he needs the conclusion thatthe encapsulated assumptions are all synthetic. This argument delivers, at best, theconclusion that some of them are. It could be that along with the useful syntheticassumptions, a lot of not-terribly-useful analytic assumptions are also encapsulated (as aby-product, maybe), and these can give rise to a priori philosophical justification andknowledge.A second issue concerns the unremarked slide from talking about analytic assumptions

to talking about analytic inferences. I’m not quite sure what Papineau means by an‘analytic inference’, but from context one might assume he means a non-ampliativeone. But it’s not clear without further discussion why a mechanism that delivers onlyanalytic propositions as ‘intuitions’ shouldn’t also be capable of ‘engaging in’ ampliativeinferences. In general, it’s best not to assume that the relationship between propositionsand inferences (and that between correlative cognitive states and processes) will bestraightforward (witness, for example, the importance of distinguishing between ruleand statement versions of logical principles like modus ponens).A third point is that the argument assumes that whatever is reached bymeans of a non-

ampliative inference is not a ‘further conclusion’. This imports some heavyweightassumptions about the identity conditions of items of information. And however thoseissues play out, it isn’t unreasonable to suppose that it would be useful to have amechanism for delivering conclusions which are non-obviously non-ampliatively entailedby our limited starting information, whether in the end these count as ‘further conclu-sions’ or just helpful re-statements of information that we (in some sense) already had.Papineau’s second argument for the syntheticity of the relevant encapsulated

assumptions is that thought experiments would be ‘much less interesting’ if theresultant intuitions were analytic, because then ‘they would tell us about the structureof our concepts, but they wouldn’t help us understand the rest of the world’ (p. 19). I’llsay three things about this. The first is that on at least some very influential construals of‘analyticity’, analytically true propositions are not propositions about concepts, butpropositions which are true in virtue of concepts (or sometimes merely knowable in virtueof conceptual competence, as on Boghossian’s 1997 conception of ‘epistemic analyticity’).Secondly, even if intuitions are a guide to the nature of our concepts in the first instance,one might treat the structure of one’s concepts as in turn being a guide to (certainelements of) the structure of the world that is mapped by the contours of thoseconcepts. (This is one of the key commitments of the kind of a priorism sketched in}3.) Thirdly, it looks as if the conclusion of Papineau’s argument here is once again notwhat he needs. At best, he might have shown that we need our encapsulated assump-tions to include some syntheticities, so that the process of thought experimentation isnot too boring. But there is no obvious reason why they couldn’t also include someanalyticities, which could give rise to a priori philosophical knowledge.


So Papineau’s challenge to the a priori can be met too (and indeed, most of the workof meeting it can be done without reliance on any particular account of how a prioriknowledge might work).

6 Concluding RemarksQuine* and Maddy argue from naturalistic premises to the conclusion that there is no apriori. In reply, I have argued that the a priori is susceptible to no such objections. Thisis not because the naturalistic premises are wrong (though perhaps they are; I have notconsidered that question here), but just because the a priori, seen in the right light, isn’tincompatible with them.

Papineau argues from naturalistically motivated premises to the conclusion that the apriori has no place in (theoretical) philosophy. In reply, I have argued that the groundson which he bases his argument are inadequate to support its conclusion.

In general, I am a lot more optimistic about the existence and significance of the apriori than many who self-identify as ‘naturalists’ appear to be. The full range ofpossible naturalistic challenges to the a priori is large and diverse; I certainly don’tclaim to have given anything like a recipe for defusing all such challenges in thepreceding sections. My hope is rather that by addressing this sample set of fairlycharacteristic challenges, I can encourage others to share my optimism about theprospects for detente between the a priori and naturalism.

Notes* Thanks to members of an audience at the University of Leeds, and especially to Jonathan

Ichikawa, for comments and discussion. Thanks also to two anonymous readers for OUP, fortheir comments on a near-final draft.

1. I discuss reasons for focusing on a notion of a prioricity which does not require empiricalindefeasibility in Jenkins (2008b, }1). In brief, adding an indefeasibility requirement makes ittoo easy to reject the a priori for uninteresting reasons.

2. The reconstruction draws on earlier work (see Jenkins 2008a, }7.5).3. That question, to be sure, has various possible answers which the less caricatured version of

this view must address.4. I discuss the relationship between ‘intuition’ and concepts in Jenkins MS.5. This even leaves room for a concept-level analogue of the verificationist constraint on

meaning to be adopted if desired (see Jenkins 2008a, p. 176).6. See Jenkins (2008a, p. 148) for more details.7. See Jenkins (2008a, }9.5) for more details.

ReferencesArmstrong, D. 1978. ‘Naturalism, Materialism and First Philosophy’, in Philosophia 8, pp. 261–76.Bealer, G. 2000. ‘A Theory of the A Priori’, in Pacific Philosophical Quarterly 81, pp. 1–30.


Braddon-Mitchell, D. 2009. ‘Naturalistic Analysis and the A Priori’, in D. Braddon-Mitchell andR. Nola (eds), Conceptual Analysis and Philosophical Naturalism, Cambridge, MA: MIT Press,pp. 23–44.

Boghossian, P. 1997. Analyticity, in B. Hale and C. Wright (eds), Companion to the Philosophy ofLanguage, Oxford: Blackwell.

Colyvan 2009. ‘Naturalising Normativity’, in D. Braddon-Mitchell and R. Nola (eds), Concep-tual Analysis and Philosophical Naturalism, Cambridge, MA: MIT Press, pp. 303–13.

Devitt, M. 2005. ‘There Is no A Priori’, in Steup, M. and Sosa, E. (eds), Contemporary Debates inEpistemology, Oxford: Blackwell, pp. 105–15.

Feldman, R. 2001. ‘Naturalized Epistemology’, in E. Zalta (ed.), Stanford Encyclopedia of Philoso-phy, available at <http://plato.stanford.edu/entries/epistemology-naturalized/>.

Grice, H. P. and Strawson, P. 1956. ‘In Defence of a Dogma’, in The Philosophical Review 65,pp. 141–58.

Hare, R. M. 1995. ‘A New Kind of Ethical Naturalism’, in Midwest Studies in Philosophy 20,pp. 340–56.

Jenkins, C. S. 2007. ‘Epistemic Norms and Natural Facts’, in American Philosophical Quarterly 44,pp. 259–72.

Jenkins, C. S. 2008a. Grounding Concepts: An Empirical Basis for Arithmetical Knowledge. Oxford:Oxford University Press.

Jenkins, C. S. 2008b. ‘A Priori Knowledge: Debates and Developments’, in Philosophy Compass3, pp. 436–50.

Jenkins, C. S. forthcoming. ‘What Can We Know A Priori?’, to appear in R. Neta (ed.) CurrentControversies in Epistemology, London: Routledge.

Jenkins, C. S. MS. ‘ “Intuition”, Intuitions, Concepts and the A Priori’, draft available at <http://www.carriejenkins.co.uk/research/work-in-progress>.

Kornblith, H. 2007. ‘Naturalism and Intuitions’, in Grazer Philosophische Studien 74, pp. 27–49.Lenman, J. 2006. ‘Moral Naturalism’, in E. Zalta (ed.), Stanford Encyclopedia of Philosophy,available at <http://plato.stanford.edu/entries/naturalism-moral/>.

Maddy, P. 1992. ‘Indispensability and Practice’, in The Journal of Philosophy 89, pp. 275–89.Maddy, P. 1997. Naturalism in Mathematics. Oxford: Oxford University Press.Maddy, P. 2000. ‘Naturalism and the A Priori’, in P. Boghossian and C. Peacocke (eds), New

Essays on the A Priori, Oxford: Oxford University Press, pp. 92–116.Nolan, D. 2005. David Lewis. Chesham: Acumen.Papineau, D. 2007. ‘Naturalism’, in E. Zalta (ed.), Stanford Encyclopedia of Philosophy, available at<http://plato.stanford.edu/entries/naturalism>.

Papineau, D. 2009. ‘The Poverty of Analysis’, in Aristotelian Society Supplementary Volume 82,pp. 1–30.

Paseau, A. 2008. ‘Naturalism in the Philosophy of Mathematics’, in E. Zalta (ed.), Stanford Encyclo-pedia of Philosophy, available at <http://plato.stanford.edu/entries/naturalism-mathematics>.

Peacocke, C. 2000. ‘Explaining the A Priori: The Programme of Moderate Rationalism’, inP. Boghossian and C. Peacocke (eds), 2000, New Essays on the A Priori, Oxford: ClarendonPress, pp. 255–85.

Plantinga, A. 1996. ‘Methodological Naturalism?’, in J. van der Meer (ed.), Facets of Faith andScience volume 1, Lanham, MD: University Press of America, pp. 177–221.


http://www.carriejenkins.co.uk/research/work-in-progress

http://www.carriejenkins.co.uk/research/work-in-progress

http://plato.stanford.edu/entries/naturalism-moral/

http://plato.stanford.edu/entries/naturalism

http://plato.stanford.edu/entries/naturalism-mathematics

http://plato.stanford.edu/entries/epistemology-naturalized/

Quine, W. 1951. ‘Two Dogmas of Empiricism’, in The Philosophical Review 60, repr. in his From aLogical Point of View: Nine Logico-Philosophical Essays, 1953, edn. of 1980, Cambridge, MA:Harvard University Press, pp. 20–46.

Schlick, M. 1936. ‘Meaning and Verification’, in The Philosophical Review 45, pp. 339–69.Sider, T. 2011. Writing the Book of the World. Oxford: Oxford University Press.Sober, E. 1993. ‘Indispensability and Mathematics’, in The Philosophical Review 102, pp. 35–57.Sosa, E. 1998. ‘Minimal Intuition’, in M. De Paul andW. Ramsey (eds),Rethinking Intuition: ThePsychology of Intuition and Its Role in Philosophical Inquiry, Oxford: Rowman & Littlefield,pp. 257–69.

Sosa, E. 2009. ‘A Defence of Intuitions’, in D. Murphy and M. Bishop (eds), Stich and his Critics,Oxford: Blackwell, pp. 101–12.

Stoljar, D. 2009. ‘Physicalism’, in E. Zalta (ed.), Stanford Encyclopedia of Philosophy, available at<http://plato.stanford.edu/entries/physicalism/>.

Uebel, T. 2006. ‘Vienna Circle’, in E. Zalta (ed.), Stanford Encyclopedia of Philosophy, available at<http://plato.stanford.edu/entries/vienna-circle/>.


http://plato.stanford.edu/entries/physicalism/

http://plato.stanford.edu/entries/vienna-circle/

13

How Deep is the Distinctionbetween A Priori and A PosterioriKnowledge?1

Timothy Williamson

1The distinction between a priori and a posteriori knowledge can be introduced thebottom–up way, by examples. I know a posteriori whether it is sunny. I know a priorithat if it is sunny then it is sunny. Such examples are projectible. We learn from themhow to go on in the same way, achieving fair levels of agreement in classifying newcases without collusion. Of course, as well as clear cases on each side there are unclearcases, which elicit uncertainty or disagreement when we try to classify them as a priorior as a posteriori. But virtually all useful distinctions are like that. If we want to be moreprecise, we can stipulate a sharper boundary with the clear cases of the a priori on oneside and the clear cases of the a posteriori on the other. How could such a distinction beproblematic? If some philosopher’s theory puts a clear case on the wrong side of theline, surely that is a problem for the theory, not for the distinction.The risk for the bottom–up method of introduction is that it may make a distinction

of no special significance. On that scenario, our classifications follow similarities anddifferences that, although genuine, are largely superficial, like a taxonomy of plants andanimals based only on colour. If so, epistemologists would do better to avoid thedistinction between the a priori and the a posteriori in their theorizing, because itdistracts them from deeper similarities and differences.The alternative method of introduction is top–down, by a direct statement of the

difference between the a priori and the a posteriori in epistemologically significanttheoretical terms. For instance: a priori knowledge is independent of experience; aposteriori knowledge depends on experience. The risk for the top–down method isthat it may turn out that everything is on the same side of the theoretically drawn line.

If Quine (1951) is right, all knowledge depends at least indirectly on experience. Aswith the risk for the bottom–up method, that would make the distinction epistemo-logically useless, but for a different reason.

Friends of the distinction typically assume that the bottom–up and top–downmethods yield equivalent results and so are mutually supporting, each averting therisk to the other. Of course, the risk here is that the two methods may have incompat-ible results. If so, assuming otherwise leads us into epistemological error.

But are any of the risks realized? In this chapter I will suggest two ways in whichreliance on a distinction between the a priori and the a posteriori does more harm thangood in epistemology. Neither of them is exactly Quine’s. As a framework fordiscussion, I first sketch the top–down distinction as contemporary philosophers tendto conceive it.

2The distinction between the a priori and the a posteriori is primarily a classification ofspecific ways of knowing.2 Away of knowing is a priori if and only if it is independent ofexperience. It is a posteriori if and only if it depends on experience. The relevant sensesof ‘independent’ and ‘experience’ are discussed later. Every specific way of knowing iseither a priori or a posteriori, and not both. One knows p a priori if and only if oneknows p in an a priori way. One knows p a posteriori if and only if one knows p in an aposteriori way. Thus if one knows p, one knows it either a priori or a posteriori.

One may know p both a priori and a posteriori, if one knows it in several ways,some a priori, some a posteriori. Tradition excluded that case on the grounds thatonly necessities (truths that could not have been otherwise) are known a prioriwhereas only contingencies (truths that could have been otherwise) are known aposteriori. But that was a mistake. Here is a simple counterexample. Suppose thatMary is good at mathematics but bad at geography, while John is bad at mathematicsbut good at geography. Both of them can perform elementary deductions. Maryknows a priori by the usual standards that 289 + 365 = 654 and does not know at allthat there are cable cars in Switzerland. John knows a posteriori by the usual standardsthat there are cable cars in Switzerland but does not know at all that 289 + 365 = 654.From the premise that 289 + 365 = 654, Mary competently deduces the disjunctiveconclusion that either 289 + 365 = 654 or there are cable cars in Switzerland (since adisjunction follows from either disjunct), and thereby comes to know the disjunctiona priori by the usual standards, since the logical deduction introduces no dependenceon experience. Meanwhile, from the premise that there are cable cars in Switzerland,John competently deduces the same disjunctive conclusion, and thereby comes toknow it a posteriori by the usual standards, for although the deduction itself is apriori, his knowledge of the conclusion inherits the dependence on experience of hisknowledge of the premise. Thus John and Mary know the same disjunctive truth, butMary knows it a priori while John knows it a posteriori. Since the disjunction that

292 TIMOTHY WILL IAMSON

either 289 + 365 = 654 or there are cable cars in Switzerland inherits necessity fromits first disjunct, John knows a necessary truth a posteriori.The primary distinction between ways of knowing can be used to effect a secondary

classification of things known. A truth p is a priori if and only if p can be known a priori.If we stipulate analogously that p is a posteriori if and only if p can be known aposteriori, then a truth may be both a priori and a posteriori, as with the disjunctionthat either 289 + 365 = 654 or there are cable cars in Switzerland. Alternatively, if westipulate that a truth is a posteriori if and only if it is not a priori, then a truth that cannotbe known a posteriori counts as a posteriori if it also cannot be known a priori. There isno presumption that a truth can be known at all. Perhaps the best fit to current practicewith the term is to stipulate that a truth is a posteriori if and only if it can be known aposteriori but cannot be known a priori. Even this way of drawing the distinction issubject to Kripke’s forceful case (1980) that there are both contingent a priori truthsand necessary a posteriori ones. Although some philosophers reject Kripke’s argu-ments, even they now usually accept the burden of proof to show why the epistemo-logical distinction between the a priori and the a posteriori should coincide with themetaphysical distinction between the necessary and the contingent.To go further, we must clarify the terms ‘independent’ and ‘experience’. One issue is

that even paradigms of a priori knowledge depend in a sense on experience. Forexample, we supposedly know a priori that if it is sunny then it is sunny. But if ourcommunity had no direct or indirect experience of the sun or sunny weather, howcould we understand what it is to be sunny, as we must if we are so much as to entertainthe thought that it is sunny, let alone know that it is so?The standard response is to distinguish between two roles that experience plays in

cognition, one evidential, the other enabling. Experience is held to play an evidential rolein our perceptual knowledge that it is sunny, but a merely enabling role in ourknowledge that if it is sunny then it is sunny: we needed it only to acquire the conceptsunny in the first place, not once we had done so to determine whether it currentlyapplies. Experience provides our evidence that it is sunny, but not our evidence that ifit is sunny then it is sunny; it merely enables us to raise the question. The idea is that ana priori way of knowing may depend on experience in its enabling role but must notdepend on experience in its evidential role.Another issue is how widely to apply the term ‘experience’. It is mainly associated

with ‘outer’ experience, involving perception by the usual five senses, but why shouldit exclude ‘inner’ experience, involving introspection or reflection? After all, one’sknowledge that one is in pain is presumably a posteriori, even though the experienceon which it depends is inner. Excluding inner experience by mere stipulation, withoutreference to any deeper epistemological difference, is liable to make the distinctionbetween a priori and a posteriori knowledge epistemologically superficial. Inner andouter experience will therefore provisionally be treated on an equal footing.One might worry that if inner experience is included, our experience of reflecting

on the proposition that if it is sunny then it is sunny will play an evidential role in our

HOW DEEP IS THE DISTINCTION 293

knowledge that if it is sunny then it is sunny, and that Mary’s experience of calculatingthat 289 + 365 = 654, on paper or in her head, will play an evidential role in herknowledge that 289 + 365 = 654. Presumably, the response is that the role is purelyenabling. The relevant evidence is not the psychological process of reflecting orcalculating, but rather in some sense the non-psychological logical or mathematicalfacts to which that process enables one to have access.

On further thought, however, that response causes more problems than it solves. Forwhat prevents it from generalizing to outer experience? For example, part of theevidence that a massive comet or asteroid collided with the Earth about 250 millionyears ago is said to be that certain sediment samples from China and Japan containcertain clusters of carbon atoms. That those samples contained those clusters of atoms isa non-psychological fact. Of course, in some sense scientists’ outer experience played arole in their access to the fact. But, by analogy with the logical and mathematical cases,the relevant evidence is not the psychological process of undergoing those outerexperiences, but rather the non-psychological physical facts to which that processenables us to have access. The role of the outer experience is purely enabling, notevidential. If so, what would usually be regarded as paradigm cases of a posterioriknowledge risk reclassification as a priori.

The threat is not confined to theoretical knowledge in the natural sciences. Even foreveryday observational knowledge, it is a highly controversial move to put thepsychological process of undergoing the outer experience into the content of theperceptual evidence we thereby gain. What we observe is typically a non-psycho-logical fact about our external environment, not a psychological fact about ourselves.

One obstacle to resolving the problem is the unclarity of the terms ‘experience’ and‘evidence’ as many philosophers use them. The top–down way of introducing thedistinction between the a priori and the a posteriori promised to put it on a firmtheoretical footing, but in practice relies on other terms (such as ‘experience’ and‘evidence’) understood at least partly bottom–up, through examples and prototypes.Although bottom–up understanding often serves us well enough, in the present case itleaves us puzzled all too soon.

Of course, many attempts have been made to explicate the a priori–a posterioridistinction by introducing new theoretical apparatus. My aim here is not to discussthose attempts separately. Instead, I will address the distinction more directly, bycomparing what would usually be regarded as a clear case of a priori knowledge withwhat would usually be regarded as a clear case of a posteriori knowledge. I will arguethat the epistemological differences between the two cases are more superficialthan they first appear. The conclusion is not that the cases are borderline. I do notdeny that they really are clear cases of a priori and a posteriori knowledge respectively,at least by bottom–up standards. In any case, showing that a distinction has borderlinecases does not show that it is unhelpful for theoretical purposes. Rather, the appropriateconclusion is that the a priori–a posteriori distinction does not cut at the epistemo-logical joints.


An analogy may be helpful with an argument of the same kind about a politicaldistinction. If one aims to criticize the distinction between liberal and non-liberalpolicies, one achieves little by producing examples of policies that are neither clearlyliberal nor clearly non-liberal. Every useful political distinction has borderline cases.But if one can produce an example of a clearly liberal policy that is politically onlysuperficially different from a clearly non-liberal policy, then one has gone some waytowards showing that the liberal–non-liberal distinction does not cut at the politicaljoints.3

3Here are two truths:

(1) All crimson things are red.(2) All recent volumes of Who’s Who are red.

On the standard view, normal cases of knowledge of (1) are clearly a priori, because bydefinition crimson is just a specific type of red, whereas normal cases of knowledge of (2)are clearly a posteriori, because it takes direct or indirect experience of recent volumes ofthe British work of reference Who’s Who to determine their colour (that is, thepredominant colour of their official cover). But let us describe two cases in more detail.Suppose that Norman acquires the words ‘crimson’ and ‘red’ independently of each

other, by ostensive means. He learns ‘crimson’ by being shown samples to which itapplies and samples to which it does not apply, and told which are which. He learns‘red’ in a parallel but causally independent way. He is not taught any rule like (1),connecting ‘crimson’ and ‘red’. Through practice and feedback, he becomes veryskilful in judging by eye whether something is crimson, and whether something isred. Now Norman is asked whether (1) holds. He has not previously considered anysuch question. Nevertheless, he can quite easily come to know (1), without looking atany crimson things to check whether they are red, or even remembering any crimsonthings to check whether they were red, or making any other new exercise of percep-tion or memory of particular coloured things. Rather, he assents to (1) after briefreflection on the colours crimson and red, along something like the following lines.First, Norman uses his skill in making visual judgments with ‘crimson’ to visuallyimagine a sample of crimson. Then he uses his skill in making visual judgments with‘red’ to judge, within the imaginative supposition, ‘It is red’. This involves a generalhuman capacity to transpose ‘online’ cognitive skills originally developed in perceptioninto corresponding ‘offline’ cognitive skills subsequently applied in imagination. Thatcapacity is essential to much of our thinking, for instance when we reflectively assessconditionals in making contingency plans.4 No episodic memories of prior experi-ences, for example of crimson things, play any role. As a result of the process, Normanaccepts (1). Since his performance was sufficiently skilful, background conditions werenormal, and so on, he thereby comes to know (1).


Naturally, that broad-brush description neglects many issues. For instance, whatprevents Norman from imagining a peripheral shade of crimson? If one shade ofcrimson is red, it does not follow that all are. The relevant cognitive skills must betaken to include sensitivity to such matters. If normal speakers associate colour termswith central prototypes, as many psychologists believe, their use in the imaginativeexercise may enhance its reliability. The proximity in colour space of prototypicalcrimson to prototypical red is one indicator, but does not suffice by itself, since it doesnot discriminate between ‘All crimson things are red’ (true) and ‘All red things arecrimson’ (false). Various cognitive mechanisms can be postulated to do the job. Weneed not fill in the details, since for present purposes what matters is the overall picture.So far, we may accept it as a sketch of the cognitive processes underlying Norman’s apriori knowledge of (1).

Now compare the case of (2). Norman is as already described. He learns the complexphrase ‘recent volumes ofWho’s Who’ by learning ‘recent’, ‘volume’, ‘Who’s Who’, andso on. He is not taught any rule like (2), connecting ‘recent volume ofWho’s Who’ and‘red’. Through practice and feedback, he becomes very skilful in judging by eyewhether something is a recent volume of Who’s Who (by reading the title), andwhether something is red. Now Norman is asked whether (2) holds. He has notpreviously considered any such question. Nevertheless, he can quite easily come toknow (2), without looking at any recent volumes ofWho’s Who to check whether theyare red, or even remembering any recent volumes of Who’s Who to check whetherthey were red, or any other new exercise of perception or memory. Rather, he assentsto (2) after brief reflection along something like the following lines. First, Norman useshis skill in making visual judgments with a ‘recent volume of Who’s Who’ to visuallyimagine a recent volume of Who’s Who. Then he uses his skill in making visualjudgments with ‘red’ to judge, within the imaginative supposition, ‘It is red’. Thisinvolves the same general human capacity as before to transpose ‘online’ cognitive skillsoriginally developed in perception into corresponding ‘offline’ cognitive skills subse-quently applied in imagination. No episodic memories of prior experiences, forexample of recent volumes of Who’s Who, play any role. As a result of the process,Norman accepts (2). Since his performance was sufficiently skilful, background condi-tions were normal, and so on, he thereby comes to know (2).

As before, the broad-brush description neglects many issues. For instance, whatprevents Norman from imagining an untypical recent volume of Who’s Who? If onerecent volume of Who’s Who is red, it does not follow that all are. The relevantcognitive skills must be taken to include sensitivity to such matters. As before, Normanmust use his visual recognitional capacities offline in ways that respect untypical as wellas typical cases. We may accept that as a sketch of the cognitive processes underlyingNorman’s a posteriori knowledge of (2).

The problem is obvious. As characterized earlier, the cognitive processes underlyingNorman’s clearly a priori knowledge of (1) and his clearly a posteriori knowledge of (2)are almost exactly similar. If so, how can there be a deep epistemological difference


between them? But if there is none, then the a priori–a posteriori distinction isepistemologically shallow.One response is to argue that at least one of the cases has beenmislocated in relation to

the a priori–a posteriori boundary. Perhaps Norman’s knowledge of (1) is really aposteriori, or his knowledge of (2) is really a priori (although presumably we did notmake both mistakes). The risks of such a strategy are also obvious. If we reclassifyNorman as knowing (1) a posteriori, we may have to do the same for all or mostsupposed cases of a priori knowledge, perhaps even of basic principles in logic andmathematics (such as standard axioms of set theory). ForNorman’s knowledge of (1) didnot initially seem atypical as a supposed case of a priori knowledge. On the other hand, ifwe reclassify Norman as knowing (2) a priori, we may still lose the distinction between apriori disciplines such as logic and mathematics and a posteriori disciplines such asphysics and geography. Either way, we end up with an a priori–a posteriori distinctionthat cannot do much theoretical work.Another response to the descriptions of how Norman knows (1) and (2) is more

sceptical: it may be suggested that if the cognitive processes are really as described, thenthey are too unreliable to constitute genuine knowledge at all. However, this option isalso unpromising for friends of the a priori–a posteriori distinction, for at least tworeasons. First, it imposes an idealized epistemological standard for knowledge thathuman cognition cannot be expected to meet. None of our cognitive faculties iseven close to being globally infallible. More local forms of reliability may suffice forknowledge; sceptics have not shown otherwise. Second, even if we are sceptical aboutknowledge in such cases, we should still assign belief in (1) and (2) some other sort ofpositive epistemic status, such as reasonableness, to which the a priori–a posterioridistinction should still apply in some form. Norman’s a priori reasonable belief in (1)and his a posteriori reasonable belief in (2) could still be used in a similar way to argueagainst the depth of the new distinction. For purposes of argument, we may as wellaccept that Norman knows (1) and (2).In the terms used in Section 2, the question is whether Norman’s experience plays

an evidential or a merely enabling role in his knowledge of (1) and (2). Even in the caseof (1), the role seems more than purely enabling. Consider Norbert, an otherwisecompetent native speaker of English who acquired the words ‘crimson’ and ‘red’ ascolour terms in a fairly ordinary way, but has not had very much practice with feedbackat classifying visually presented samples as ‘crimson’ or ‘not crimson’. He usually makesthe right calls when applying ‘crimson’ as well as ‘red’ online. By normal standards he islinguistically competent with both words. He grasps proposition (1). However, hisinexperience with ‘crimson’ makes him less skilful than Norman in imagining acrimson sample. As a result, Norbert’s reflection on whether crimson things are redcomes to no definite conclusion, and he fails to know (1). Thus Norman’s pastexperience did more than enable him to grasp proposition (1). It honed and calibratedhis skills in applying the terms ‘crimson’ and ‘red’ to the point where he could carry outthe imaginative exercise successfully. If Norman’s experience plays a more than purely


enabling role in his knowledge of (1), a fortiori it also plays a more than purely enablingrole in his knowledge of (2).5

If the role of Norman’s experience in his knowledge of (1) is more than purelyenabling, is it strictly evidential? One interpretation of the example is that, althoughNorman’s knowledge of (1) does not depend on episodic memory, and he may evenlack all episodic memory of any relevant particular colour experiences, he neverthelessretains from such experiences generic factual memories of what crimson things look likeand of what red things look like, on which his knowledge of (1) depends. By contrast,Norbert fails to know (1) because his generic memories of what crimson things look likeand of what red things look like are insufficiently clear. On this interpretation, Norman’scolour experience plays an evidential role in his knowledge of (1), thereby making thatknowledge a posteriori. But we have already seen that such reclassification is a riskystrategy for defenders of the a priori–a posteriori distinction. Instead, it may be proposed,although colour experience can play an evidential role in a posteriori knowledge of whatcrimson things look like, and so indirectly in a posteriori knowledge of (1), we need notdevelop the example that way. The only residue of Norman’s colour experience activein his knowledge of (1) may be his skill in recognizing and imagining colours.6 Such arole for experience, it may be held, is less than strictly evidential. Let us provisionallyinterpret the example the latter way. In Section 5 we will reconsider, but reject, the ideathat even supposed paradigms of a priori knowledge are really a posteriori.

Norman’s knowledge of (2) can be envisaged in parallel to his knowledge of (1) asjust envisaged. Although experience of recent volumes of Who’s Who can play anevidential role in a posteriori knowledge of what such volumes look like, and soindirectly in a posteriori knowledge of (2), that is not what goes on with Norman. Theonly residue of his experience of recent volumes ofWho’s Who active in his knowledgeof (2) is his skill in recognizing and imagining such volumes. That role for experience isless than strictly evidential. Nor does Norman’s present experience play any more of anevidential role in his knowledge of (2) than it does in his knowledge of (1).

On this showing, the role of experience in both cases is more than purely enablingbut less than strictly evidential. This reinforces a suspicion raised in Section 2, that talkof ‘experience’ and ‘evidence’ does little to help us apply the a priori–a posterioridistinction top–down. Appeals to ‘observation’ as the hallmark of a posteriori know-ledge hardly do better, for they leave us with the question: in what way does Norman’sknowledge of (2) involve observation while his knowledge of (1) does not?

The most salient difference between Norman’s knowledge of (2) and his knowledgeof (1) is that (2) is contingent while (1) is necessary. That difference may be whatinspires the idea that there must be a deep difference in his knowledge of them. ButKripke taught us not to read the epistemology of a truth off its metaphysical status. Ofcourse, these two propositions do differ epistemologically:

(N1) It is necessary that crimson things are red.(N2) It is necessary that recent volumes of Who’s Who are red.


For (N1) is a known truth, while (N2) is false, so not known, indeed presumablyimpossible, so unknowable. But an epistemological difference between (N1) and (N2)does not imply any epistemological difference between (1) and (2). In general, knowing anecessary truth does not imply knowing that it is necessary. For example, John in Section2 knew that either 289+ 365= 654 or there are cable cars in Switzerland by knowing thatthere are cable cars in Switzerland, without knowing that 289 + 365 = 654; he did notknow that it is necessary that either 289+ 365= 654 or there are cable cars in Switzerland,even though it is indeed necessary. Likewise, Norman may know (1) without knowing(N1). He may be a sceptic about necessity, or never even entertain modal questions. Inparticular, Norman can know (1) without knowing (N1) and deducing (1) from it.Indeed, the idea that a precondition of knowing a necessary truth is knowing that it isnecessary generates an infinite regress. For since (N1) is itself necessary, a precondition ofknowing (N1) would be knowing (NN1), and so on ad infinitum:

(NN1) It is necessary that it is necessary that crimson things are red.

A subtler attempt to extract epistemological significance from the difference in modalstatus between (1) and (2) might exploit the modal nature of some proposed constraintson knowledge, such as various versions of reliability, sensitivity, and safety, many ofwhich imply that one knows p only if falsely believing p is in some sense not too live apossibility.7 For since (1) is necessary, false belief in (1) is impossible, whereas false beliefin (2) is possible, although not actual. Of course, since Norman is granted to know both(1) and (2), he satisfies any modal necessary condition for knowledge with respect toboth truths. However, he might still be more reliable, or sensitive, or safe, with respectto (1) than to (2). But how could any such contrast make the difference between apriori and a posteriori knowledge? No constraint that all necessary truths trivially satisfyexplains why some of them are known a priori, others only a posteriori.To consider the point more fully, imagine Gull, who believes whatever his guru tells

him. The guru tosses a coin to decide whether to assert to Gull Fermat’s Last Theorem(FLT), if it comes up heads, or its negation (¬FLT), if it comes up tails. The coin comesup heads, and on the guru’s testimony Gull obediently believes FLT, a necessary truth.It would generally be agreed that Gull does not know FLT. Indeed, in the non-technical senses of the terms, Gull’s belief in FLT does not look particularly reliable, orsafe, or sensitive to the facts. If the coin had come up tails, Gull would have wound upbelieving the necessary falsehood ¬FLT instead, in a parallel way. Any version of areliability, sensitivity, or safety condition on knowledge non-trivially applicable toknowledge of a necessary truth p will concern possibilities of false belief in otherpropositions suitably related to p, not just of false belief in p itself. On such a dimension,we have no reason to expect Norman’s knowledge of (1) to do better than hisknowledge of (2). He may be just as prone to error in his judgments of colour inclusionas in his judgments of the colours of types of book; the distribution of errors in modalspace may be much the same in the two cases. Once again, the difference in modalstatus between (1) and (2) is not what matters epistemologically.


The main effect of the modal difference between (1) and (2) may be to distract usfrom the epistemological similarity. The necessity of (1) prompts us to assimilateNorman’s knowledge of (1) to stereotypes of a priori knowledge, which we canvaguely do because the role of experience is not strictly evidential. The contingencyof (2) prompts us to assimilate his knowledge of (2) to stereotypes of a posterioriknowledge, which we can vaguely do because the role of experience is not purelyenabling. Even having accepted Kripke’s examples of the contingent a priori and thenecessary a posteriori, we may still operate on the default assumptions that knowledgeof necessary truths is a priori and that knowledge of contingent ones is a posteriori.Unlike Kripke’s, cases such as Norman’s may trigger nothing to overturn either default,especially when they are considered separately, so by default we confidently classifyknowledge of the necessary truth as a priori and knowledge of the contingent one as aposteriori, without noticing that there is no significant epistemological difference be-tween them. We can use the qualifiers ‘a priori’ and ‘a posteriori’ that way if we like,but then we should not expect them to do much work in epistemology.

4How widespread is the problem? It might be argued that although the a priori–aposteriori distinction does not mark any deep difference between Norman’s know-ledge of (1) and his knowledge of (2), the example is a special case, and that thedistinction marks a deep difference in a wide range of other cases. A priori knowledgeof logic and mathematics may be contrasted with Norman’s knowledge of (1), and aposteriori knowledge by direct observation, preserved by memory, transmitted bytestimony and extended by deductive, inductive, and abductive reasoning may becontrasted with his knowledge of (2). Thus, it might be claimed, the a priori–aposteriori distinction can still do plenty of useful work in epistemology after all.

I will argue that such an attitude is much too complacent. Many cases of a prioriknowledge are relevantly similar to Norman’s knowledge of (1), and many cases of aposteriori knowledge are relevantly similar to his knowledge of (2).Moreover, althoughepistemologists have become accustomed to treating the category of a priori knowledgeas problematic, we still tend to treat the category of a posteriori knowledge as epi-stemologically explanatory. This attitude is particularly prevalent amongst those whodeny that there is a priori knowledge. They think that it is clear enough how a posterioriknowledge works, but hopelessly obscure how a priori knowledge could work.8 Oncewe appreciate how problematic the distinction itself is, we may rid ourselves of theillusion that we can understand what is going on in a case of knowledge by classifying itas a posteriori. The usual stereotype of a posteriori knowledge is just as epistemologicallyuseless as the usual stereotype of a priori knowledge.

I will not discuss in detail how wide a range of other a posteriori knowledgeresembles Norman’s knowledge of (2). Since there is nothing very special about (2),it is fairly clear that if cognitive skills learnt online but applied offline can generate a


posteriori knowledge of (2), without experience playing a strictly evidential role, thenthey can do likewise for many other truths. Examples include some knowledge ofphysical and practical possibility and of counterfactual conditionals.In the case of (2), colour inclusions may look special. How much other putatively a

priori knowledge resembles Norman’s knowledge of (1)? He uses nothing like theformal proofs we associate with mathematical knowledge. A closer comparison is withknowledge of mathematical axioms, in particular with standard axioms of set theory(such as those of Zermelo–Fraenkel set theory).9

We may take as a typical example the Power Set Axiom, which says that every sethas a power set, the set of all its subsets:

PSA 8x 9y 8zðz 2 y $ z � xÞÞÞHere z � x abbreviates the formula Set(z) & Set(x) & 8u (u 2 z ! u 2 x) (Set(z)means: z is a set).10 Proofs of theorems throughout mathematics routinely and tacitlyrely on PSA. But how do mathematicians know that PSA is true? The problem here isnot how mathematicians know that there are any sets at all, for if there are no sets thenPSA is vacuously true (since both z 2 y and z� x are always false). Rather, the problemis how mathematicians know that if there is a set, it has a power set.Some textbooks motivate PSA in effect by telling readers that, unless they accept it,

they will be unable to do set theory. They do not claim that accepting the axiom isnecessary for understanding the language of set theory, in particular ‘Set’ and ‘2’. Theyintroduced those symbols at an earlier stage of the exposition. Once the axiom has beenstated, readers are treated as grasping its content but potentially still wondering why toaccept it; that is the point of the pragmatic motivation. It might be interpreted as anappeal to authority: take the author’s word for it, once you start working with thisaxiom you will see why it is needed in mathematics.Other expositions of set theory attempt more intrinsic justifications of PSA. For

instance:

If I have a set, then I can think of all possible subsets of this set. It is probably going to be a largercollection, but not so terribly much larger. It is reasonable to think of this as giving us back a set.11

This is an implicit appeal to the principle of limitation of size, that things form a set if andonly if there are not too many of them (fewer than absolutely all the things there are).Sometimes the appeal is backed up by the calculation that a finite set with just nmembers has just 2n subsets. But of course PSA is intended to apply to infinite sets too.12

An alternative to limitation of size is the picture of sets as built up by an iterativeprocess; at each stage one forms all possible sets of things already built up or given. Thispicture too is sometimes used to justify PSA:

[S]uppose x is formed at [stage] S. Since every member of x is formed before S, every subset of xis formed at S. Thus the set of all subsets of x can be formed at any stage after S.13


Of course, the apparently causal and temporal talk of forming sets at earlier or laterstages is intended metaphorically, without commitment to any genuinely constructivistconception of sets. Nevertheless, the point of the metaphor is to appeal to theimagination, enabling us to think about the question in a more vivid, concrete, andperspicuous way, and in particular to convince us that there will be a stage after S,without which the power set never gets formed.14 The metaphor prompts us toundertake an imaginative exercise that makes offline use of our online skill in observingand engaging in processes of physical creation, a skill honed by past experience. This isnot so distant from the imaginative exercise through which Norman came to know (1).

Something similar goes on in the justification from limitation of size. It starts withthe supposition that ‘I have a set’, which already suggests a picture of the set as availableto hand. On that supposition, ‘I can think of all possible subsets of this set’. Of course,none of that is intended to suggest any idealist metaphysics of sets, on which it isessential to them to be thought by a subject. Rather, the aim is again to make us engageimaginatively with the question. The point of calling the subsets ‘possible’ is not toemphasize that they could exist, for it is not in question that they actually do exist; it isto suggest that I could select them. Imagine that I have to hand three objects a, b, and c.They form a set from which I can make eight selections:

fa; b; cgfa; bgfa; cgfb; cgfagfbgfcgfg

Those eight sets are the members of the power set of the set of the original threeobjects. My online experience of making different selections from amongst percep-tually presented objects facilitates my offline imagined survey of all possible selections,and enables me to make the judgment in the quotation, ‘It is probably going to be alarger collection, but not so terribly much larger’. The cognitive tractability of thepower set in such simple cases helps us accept PSA. Again, Norman’s knowledge of (1)is not so far away.

If standard axioms of set theory are justified by general conceptions of the sets, suchas limitation of size or iterativeness, we may wonder how those general conceptions arein turn to be justified. Although the answer is hardly clear, all experience in thephilosophy of set theory suggests that the attempt to make such a general conceptionof sets intuitively compelling must rest at least as heavily on appeals to the imaginationwith metaphors and pictures as do attempts to make intuitively compelling one of thestandard set-theoretic axioms.


An alternative view is that such intrinsic justifications of set-theoretic axioms aresecondary to extrinsic ones from their fruitfulness, their explanatory and unifyingpower. This need not involve Quine’s idea that mathematics is justified by its applica-tions in natural science. Corresponding to the textbook’s implied injunction above tothe mathematical novice, ‘wait and see’, applications in mathematics itself may be morerelevant.15 Thus the strategy does not immediately commit one to an account ofmathematical knowledge as a posteriori, even though the envisaged abductive meth-odology is strongly reminiscent of the natural sciences.Bertrand Russell describes a similar order of proceeding:

[I]nstead of asking what can be defined and deduced from what is assumed to begin with, we askinstead what more general ideas and principles can be found, in terms of which what was ourstarting-point can be defined or deduced.16

He observes:

The most obvious and easy things in mathematics are not those that come logically at thebeginning; they are things that, from the point of view of logical deduction, come somewhere inthe middle.17

This suggests knowledge of the ‘most obvious and easy things inmathematics’ as a bettercandidate than knowledge of the axioms of set theory to fit the stereotype of a prioriknowledge. An example is knowledge that 2 + 2 = 4, an arduously derived theorem ofRussell and Whitehead’s system in Principia Mathematica. But if ordinary knowledge ofelementary arithmetic is not by derivation from logically more basic principles, thenpresumably it is by something more like offline pattern recognition, and we still havenot moved far from Norman’s knowledge of (1). Only the very lazy-minded could becontent with the explanation that we know that 2 + 2= 4 ‘by intuition’. Even if it is truethat we do so in some sense of ‘intuition’, how does saying that constitute a genuinealternative to a view that assimilates our knowledge to Norman’s?Even if experience plays no strictly evidential role in core mathematical practice, the

suspicion remains that its role is more than purely enabling. Although we can insist thatmathematical knowledge is a priori, it is unclear how it differs epistemologically fromsome examples of the a posteriori, such as Norman’s knowledge of (2).Rather than pursuing the epistemology of mathematics further, let us see whether the

stereotype of a priori knowledge fares any better in the epistemology of logic. For a simpleexample, consider the reflexivity of identity, the principle that everything is self-identical:

RI 8x x ¼ x

We are not asking about a priori knowledge that RI is a logical truth. We are just askingabout a priori knowledge of RI itself, knowledge that everything is self-identical.A tempting reaction is that anyone who doubts RI thereby just shows that they do

not understand it. In the jargon, RI may be claimed to be epistemologically analytic. Weneed not discuss whether epistemological analyticity entails a priority, for it is false that


basic logical truths are epistemologically analytic in the relevant sense. I know educatednative speakers of English who deny that everything is self-identical, on the groundsthat material substances that change their properties over time are not self-identical.I regard those speakers as confused, but the understanding they lack is primarily logicalrather than semantic. Although they are mistaken about the logical consequences ofidentity, by normal standards they are not linguistically incompetent with the Englishexpressions ‘everything’, ‘is’, ‘self-’, and ‘identical’, and the way they are put together,nor with their counterparts in a formal first-order language with ‘=’. A language schoolis not the place for them to learn better. We might stipulate a sense of the tricky word‘concept’ in which anyone who doubts RI counts as associating a different ‘concept’with ‘=’ from any logically standard one, but recycling a theoretical disagreement as adifference in ‘concepts’ hardly clarifies the position.18

AlthoughRI is not epistemologically analytic, it of course does not follow that it is notknown a priori. One competent speaker may know a priori what another denies. Buthow is RI known? The universal generalization is unlikely to be an axiom of a formalsystem that develops innately in the human head. In a standard natural deduction system,RI is derived by the introduction rule 8I for the universal quantifier from a formula ofthe form a = a, which is itself a theorem (indeed, an axiom) by the introduction rule=I for the identity sign.19 That is the formal analogue of imagining an object, within thescope of the imaginative supposition judging ‘it’ to be self-identical, and concluding thateverything is self-identical. The 8I rule is subject to the restriction that the term a onwhich one universally generalizes must not occur (free) in any assumption from whichthe premise of the application of 8I was derived, otherwise one could derive 8x Fxfrom Fa. In this case the restriction is vacuously met; a = a is a theorem and has noassumptions. Our informal thinking lacks a comparably clear way of keeping track of itsassumptions. We make a judgment, perhaps within the scope of an imaginative suppos-ition, but we may be unaware of its assumptions or sources. Thus it is often nottransparent to us how far we can generalize. We may be imagining the case in a waythat is less generic or typical than we think. For example, those who deny (howevermistakenly) that a changing thing is self-identical may charge that if we imagine anunchanging thing in evaluating RI we thereby beg the question in its favour.

Such reflections should not drive us into a general scepticism about our putativeknowledge of universal generalizations. That would be an over-generalization of justthe kind against which the reflections warn. They should not even drive us into aparticular scepticism about our putative knowledge of RI. Knowing p does not requireseparately assessing in advance all possible fallacious objections to p. To require us tocheck that the imagined instance is typical of all members of a domain D before weuniversally generalize over D is to impose an infinite regress, for ‘The imaginedinstance is typical of all members of D’ is itself a universal generalization overD. What matters for knowledge may be that we do safely imagine the instance in arelevantly generic way, even though the process is opaque to us. Surely we can quiteeasily know RI. Whether or not something changes is not really relevant to whether it


is self-identical, so it does not matter whether we imagine a changing object or anunchanging one.A resemblance between our knowledge of RI and Norman’s knowledge of (1) is

starting to emerge. Of course, an important aspect of Norman’s knowledge of (1) is hisoffline imaginative use of capacities to apply colour terms calibrated perceptuallyonline. Is there anything similar in our knowledge of RI? Experience involves a processof continually judging numerical identity or distinctness among objects perceived orremembered in a wide variety of guises. This cognitive capacity for judging identityand distinctness in experience is non-logical, for pure logic gives us only the barestformal constraints. If we have a non-logical capacity to make such identity judgments,we need no additional logical capacity corresponding to the rule =I to make identityjudgments of the special syntactic form a = a. After all, we could use the non-logicalcapacity to judge a = b and b = a (for some suitable term b syntactically distinct from a)and then apply the transitivity of identity to deduce a = a. The transitivity of identitydoes not depend on its reflexivity.20 A simpler and more plausible way of using a non-logical capacity to make judgments of identity and distinctness to judge a = a would bedirectly to feed in the term a twice over as both inputs to some device for comparison,which would trivially return a positive result. That can be done online or offline.Even the trivial comparison in a = a can be mishandled. If the name a denotes an

enduring, changing substance, but one associates the first token of awith the propertiesthe object had at a time t (attributed in the present tense, not relativized to t) and thesecond token of awith the properties the object had at another time t* (also attributed inthe present tense, not relativized to t*, and incompatible with the former properties),then the output from the identity test is a false negative. Through experience of materialthings undergoing slow large changes, one becomes less prone to such mistakes,although some adult metaphysicians still manage to make them, and so deny RI.The foregoing remarks are not intended to suggest that knowledge of RI is a

posteriori. Classify it as a priori by all means, but do not let that blind you to howmuch it has in common with a posteriori knowledge of identity and distinctness, just asNorman’s a priori knowledge of (1) has so much in common with his a posterioriknowledge of (2).The salient difference between (1) and (2) is modal rather than epistemological: (1) is

metaphysically necessary, (2) metaphysically contingent. By contrast, a = a and a = b donot differ in modal status if both are true and the terms a and b are proper names or otherrigid designators, for an identity claimwith such terms is metaphysically necessary if trueat all.21 But in that case the salient difference between the formulas a = a and a = b islogical rather than epistemological. For a= a but not a= b is a logical truth. The propertyof logical truth is not demarcated epistemologically but by more formal criteria.22 Just asthe modal difference between (1) and (2) makes us overestimate the strictly epistemo-logical difference between Norman’s knowledge of (1) and his knowledge of (2), sothe logical difference between a = a and a = b makes us overestimate the strictlyepistemological difference between our knowledge of a = a and our knowledge of a = b.


Naturally, far more work would have to be done to confirm the foregoing hintsabout logical and mathematical knowledge. Nevertheless, the indications so far suggestthat what are often counted as the clearest cases of a priori knowledge are much lessdifferent epistemologically than they are usually depicted as being from cases of aposteriori knowledge. The epistemological similarity of Norman’s a priori knowledgeof (1) to his a posteriori knowledge of (2) is no isolated case. The usual stereotype of apriori knowledge is seriously misleading, because it omits a pervasive role of experiencethat is more than purely enabling, although less than strictly evidential.

5The inadequacy of the usual stereotype of a priori knowledge may seem to support theidea that we can make progress by following some Quineans, if not Quine himself, inpositively classifying all knowledge as a posteriori.23 Doing so would at least have thenegative advantage of not putting a distinction where there is no deep difference. Butdoes it also yield some positive understanding of the general nature of knowledge?

On Quine’s picture, a theory faces the tribunal of experience collectively, notsentence by sentence. Taken at face value, the image implies that two consequencesof a theory cannot differ in epistemic status. But that is absurd. For Quine, the totalityof a person’s beliefs constitute a theory (perhaps an inconsistent one), their total theoryof the world, but who thinks that two of a person’s beliefs cannot differ in epistemicstatus? If some of my beliefs constitute knowledge, it does not follow that all of themdo; it does not even follow that all of them are true. One’s beliefs about science andmathematics may be on average epistemically better off than one’s beliefs aboutreligion and politics, or vice versa. Indeed, experience favours the belief that experi-ence favours some beliefs more than others. Quine later restricted his holism to amoderate version that permits some discrimination amongst our beliefs.24 Neverthe-less, at least this mild form of global holism is surely right: no two of our beliefs are inprinciple epistemically insulated from each other.

To make progress, we need a more developed model, on which an individual beliefhas its own epistemic status, but that status depends in principle on the epistemic statusof each other belief. Holism is far more plausible as a claim about the pervasiveinterdependence of epistemic status than as the claim that only whole theories haveepistemic status. The obvious and best-developed candidate for such a model is someform of Bayesian epistemology. It assigns evidential probabilities to individual propos-itions, subject to standard axioms of probability theory, which constrain the overalldistribution of probabilities to all propositions.25 The paradigmatic way of updatingevidential probabilities is by conditionalization on new evidence, encapsulated in aproposition e. The new evidential probability Probnew(p) of any proposition p is the oldconditional evidential probability Probold(p | e) of p on e, which is equal to the ratioProbold(p & e)/Probold(e) whenever Probold(e) > 0. Conditionalization is a globalprocess; one overall probability distribution, Probnew, replaces another, Probold.


However, Bayesian epistemology does not vindicate a Quinean rejection of the apriori. For standard axioms of probability theory constrain every probability distribu-tion to assign probability 1 to any theorem of classical propositional logic, andprobability 0 to its negation. Probabilistic updating on new evidence cannot raise orlower the probability of theorems or anti-theorems. That is not just an optionalconvention. Loosening it deprives probability theory of the mathematical structureon which its utility depends. Although minor concessions to specific non-classicallogics may not destroy that utility entirely, any version of probability theory worthhaving will give such a privileged status to some core of logic.In Bayesian epistemology, logical truths are not the only propositions to enjoy a

good epistemic status that they do not owe to the evidence. Let e conjoin all therelevant evidence, and Probnew be the result of conditionalizing Probold on e as above.We may assume that the evidence was not certain in advance; Probold(e) < 1. Supposethat a hypothesis h is well supported by e, so Probnew(h) is high. Consider the materialconditional e ! h. Since it is a logical consequence of h, Probnew(e ! h) is at least ashigh as Probnew(h). But we can prove that either Probnew(e! h) = Probold(e! h) = 1or Probnew(e! h) < Probold(e! h).26 In other words, either e! h was already certainprior to the evidence, which did not confirm e ! h, or the evidence disconfirmede! h. Thus e! h enjoys a good epistemic status, because Probnew(e! h) is high, but itdoes so despite the evidence or independently of it.When holistic epistemology is made rigorous, the results do not support the idea that

the only way of enjoying high epistemic status is by confirmation through experience;they do the opposite. That is not to deny the strong similarities between the epistemol-ogy of logic and the epistemology of other sciences evident in debates over proposals torevise or extend classical logic. For we cannot assume that those similarities are properlyarticulated on the model of confirmation or disconfirmation through experience.Rather than appealing to formal models, those who claim that all knowledge is

empirical or a posteriori may suggest that we understand the paradigm of suchknowledge, simple cases of observational knowledge, well enough for the informalproposal to assimilate all knowledge to the paradigm to be illuminating and non-trivial.Although there are manifest differences between the paradigm and cases of highlytheoretical knowledge, the idea is that on sufficiently deep analysis they will turn out tobe differences in complexity, not in fundamental nature.How well do we understand the paradigm, simple cases of observational knowledge?

Presumably, the picture is that in such cases sense perception is a channel for a causalconnection between the truth of a proposition p and the agent a’s belief in p, creating astrongly positive local correlation between truth and belief. We may symbolize thecorrelation as p, Bap. The proposition p should paradigmatically concern the state ofthe environment, not the state of the agent, for otherwise the case is too special to be asuitable model for knowledge in general.A first step in making the model less simplistic is to note that the correlation depends

on the receptivity of the agent. If a is too far from the relevant events or shuts her eyes or

has bad eyesight, the preconditions for the correlation may not be met. We cansymbolize this one-way correlation between receptivity and the previous two-waycorrelation as Rap ) (p , Bap).

27 Even at this utterly elementary level, it is clear thata causal connection between truth and belief is not the onlyway to achieve such a set-up.For suppose that p is a necessary truth (as it were, ) p), and that the receptivity of theagent by itself causes the belief (Rap ) Bap). Then we have Rap ) ( p , Bap), eventhough there is no causal connection between p and Bap. The receptivity condition Raphere should not be envisaged as some mystical state of opening one’s soul to Platonicheaven; it may be amundane psychological process, for example of calculation. Thus theRap) (p, Bap) model carries no commitment to conceiving the modelled epistemicstates as all a posteriori. For all it implies, some of them are a priori. Perhaps surprisingly,treating simple observation as the paradigm to which all knowledge must be assimilateddoes not in principle commit one to a uniformly a posteriori epistemology.

As an alternative paradigm of knowledge, many self-described naturalists prefer theexperimentally based findings of the natural sciences. A conception of all knowledge asa posteriori or empirical may be an attempt to assimilate it all to natural science. Ofcourse, one of the most salient obstacles to any such attempt is mathematics. Obviously,theorems of mathematics do not normally have direct experimental support. ForQuine, they have indirect experimental support because mathematics is part of ourtotal scientific theory, which is confirmed as a whole (if at all) by experimental tests.But we have already seen that theories are not the only bearers of epistemic status. Thepractice of the natural sciences themselves requires evaluating the epistemic status ofmuch smaller units: for example, should we believe a report of a given astronomicalobservation or experimental result? Once we ask more discriminating questions aboutthe epistemic status of individual axioms and theorems of mathematics, it becomesmuch harder to tell a plausible story on which they owe that status primarily toexperimental support. Although some axioms and theorems are in a better epistemicposition than others, that has far more to do with considerations internal to mathemat-ics than with experimental support.28 The same holds even more obviously for axiomsand theorems of logic.

On present evidence, the slogan ‘All knowledge is a posteriori’ or ‘All knowledge isempirical’ is defensible only if the term ‘a posteriori’ or ‘empirical’ is emptied of allserious content. Unfortunately, that does not deprive the underlying prejudice ofinfluence. Like other prejudices, it acts selectively, for instance by imposing moresevere demands for external justification on armchair methods in philosophy than onother methods of inquiry.

6We should not react to the inadequacy of the usual stereotype of a priori knowledgeby declaring all knowledge a posteriori. Conversely, of course, we should not react tothe inadequacy of the usual stereotype of a posteriori knowledge by declaring all


knowledge a priori. Since the terms ‘a priori’ and ‘a posteriori’ are not meaningless bynormal standards, some difference between a priori and a posteriori knowledgeremains. But that does not rehabilitate the distinction as of great theoretical value forepistemology. After all, there is a difference between plants that are bushes and plantsthat are not bushes, but it does not follow that that distinction is of great theoretical valuefor botany.29Whenwe start investigating some phenomena, we have little choice but toclassify them according to manifest similarities and differences. As our understandingdeepens, we may recognize the need to reclassify the phenomena on less obviousdimensions of greater explanatory significance. Distinctions that aided progress in theearly stages may hinder it later on. The a priori–a posteriori distinction is a case in point.

Cognitive psychology will have much to offer the epistemologist’s attempt toovercome philosophical prejudices and classify according to deeper and less obvioussimilarities and differences. But that does not mean a reduction of epistemology tocognitive psychology. Epistemological questions are typically at a higher level ofgenerality than those of cognitive psychology: for example, they may concern allknowledge, all epistemic probability, or all rationality. Epistemology also engagesmore fully with evaluative questions about how knowledge is better than mere truebelief, or rationality than irrationality, or whether we ought to proportion our beliefsto the evidence. Of course, some ‘naturalized’ epistemologists abjure such evaluativequestions as ‘unscientific’. In doing so, they seem more under the influence of logicalpositivism, or the prejudice discussed in Section 5, than of actual scientific practice. Inparticular, cognitive psychologists do not abjure all evaluative judgments about whatis rational or irrational, in their experimental studies of irrationality. Moreover,pursuing any scientific inquiry involves making numerous judgments that are evalu-ative in the way characteristic of epistemology. Are these data trustworthy? Is thatargument valid? Which of these theories is better supported by the evidence? Even ifthose questions play a merely instrumental role in the natural sciences, there is nowarrant for the idea that scientific practice somehow discredits the attempt to inquiremore generally into the nature of phenomena like trustworthiness, validity, andevidential support. Indeed, it seems contrary to the scientific spirit to disapprove ofsystematic general inquiry into such matters while nevertheless continually relying onjudgments about them.Beyond the potential contribution of cognitive psychology, we need to develop a

more detailed, precise, and specifically epistemological vocabulary for describing thefine structure of examples such as those in Sections 3 and 4, involving the offlineapplication in the imagination of cognitive skills originally developed in perception,especially when they involve generic imaginary instances used to reach general non-imaginary conclusions.30 That is likely to prove a far more fruitful project forepistemology than yet another attempt to reconstruct the tired-out distinction be-tween the a priori and the a posteriori or to stretch the latter thin enough to cover allknowledge.


Notes1. This paper develops the brief critique of the distinction between a priori and a posteriori

knowledge in Williamson 2007, pp. 165–9. Earlier versions of the material were discussed atKing’s College London, the University of Santiago de Compostela, Moscow State Univer-sity, St Petersburg State University, the University of Hertfordshire, a course on Mind,World, and Action at the Inter-University Centre in Dubrovnik, a ‘master-class’ for theBritish Postgraduate Philosophy Association in London where Giacomo Melis and AsgerSteffensen commented on it, and a seminar at the University of Lisbon. I am grateful tothem, to audiences on those occasions, to Anna-Sara Malmgren, and to an anonymousreferee for Oxford University Press for comments, questions, and discussion.

2. Many contemporary epistemologists, especially those of an internalist bent, treat the distinc-tion as primarily a classification of forms of justification rather than of knowledge. Forreasons explained in Williamson 2000, I treat the classification of forms of knowledge asprimary. Friends of justification should not find much difficulty in reworking the argumentsof this chapter in their terms.

3. I am not actually endorsing the conclusion that the liberal–non-liberal distinction does notcut at the political joints, because I have not actually given an example of a clearly liberalpolicy that is politically only superficially different from a clearly non-liberal policy.

4. See Williamson 2007, pp. 137–78.5. It is even more implausible that experience plays a purely enabling role in the more complex

examples in Williamson 2007, pp. 166–7; (1) is used here for its simplicity.6. Compare the huge debate on Mary’s (a posteriori?) knowledge of what red things look like

( Jackson 1982).7. For some discussion see Williamson 2000, pp. 123–30 and 147–63.8. For an example of this attitude see Devitt 2005.9. The axioms of group theory are simply clauses in the mathematical definition of ‘group’ and

so raise no distinctive epistemological problem; likewise for the axioms for other kinds ofalgebraic or geometrical structure. The primary role of the axioms of set theory in math-ematics is quite different. Proofs in all branches of mathematics rely on their truth. They arenot clauses in the mathematical definition of ‘set’ (there is no such definition analogous tothe mathematical definition of ‘group’). Although they can be adapted to serve as suchclauses in a mathematical definition of ‘cumulative hierarchy’ or the like, that is a secondaryuse; even proofs about all cumulative hierarchies rely on the axioms of set theory in theirprimary role. Some globally structuralist philosophers of mathematics may urge a differentattitude, but mathematicians have not yet seen fit to indulge them, nor is it clear how theycould coherently adopt a globally structuralist attitude.

10. The quantifiers in PSA are not restricted to sets. In applying mathematics, we may need thepower set of a set of concrete objects. If x is a non-set, it has no subsets, so y can be {} (if wewant a set) or x itself (if we want a non-set). If x is a set, then {} � x, so ymust be a set since{} 2 y and only sets have members. If we did not impose Set(z) as a condition on z � x,every non-set would vacuously count as a subset of any set x and so have to belong to thepower set of x.

11. Crossley, Ash, Brickhill, Stillwell, and Williams 1972, p. 61.


12. An infinite set with Œmembers also has 2Œ subsets, but in the infinite case the definition of 2Œ

depends on PSA rather than offering it independent support. For the history of thelimitation of size principle and scepticism about its capacity to motivate PSA, see Hallett1984.

13. Shoenfield 1977, p. 326. For a more philosophical account of the iterative conception of set,see Boolos 1971.

14. The appeal to the imagination is explicit at pp. 323–4 of Shoenfield 1977, in his generalaccount of the process of set formation. Note that he is writing as a mathematician formathematicians and others, not as a philosopher.

15. See Maddy 2011.16. Russell 1919, p. 1. According to Russell, the latter order ‘characterises mathematical

philosophy as opposed to ordinary mathematics’. If Maddy is right, it also characterizes settheory as a branch of ordinary mathematics.

17. Russell 1919, p. 2.18. SeeWilliamson 2007, pp. 73–133, for a more detailed critique of epistemological analyticity.19. Matters are more complicated in free logic, where the term a is not guaranteed to denote

anything in the domain of quantification.20. The standard proof of transitivity uses the elimination rule =E, the indiscernibility of

identicals, but not the introduction rule =I.21. The classic defence of the necessity of identity is, of course, Kripke 1980. On some

metaphysical views, what is necessary in the case of both identities is only that if the objectsexist then they are identical.

22. For present purposes logical truths are treated as formulas rather than propositions. The truthof a = b does not make it the same logical truth as a = a.

23. ‘It is overwhelmingly plausible that some knowledge is empirical, “justified by experience.”The attractive thesis of naturalism is that all knowledge is; there is only one way of knowing.’(Devitt 2005, p. 105).

24. See Quine 1981, p. 71.25. More accurately, the constraints are on a distribution of probabilities to all propositions in

the �-field of propositions that receive probabilities at all.26. Probold(e & ¬h) = Probold(e & ¬h | e)Probold(e) + Probold(e & ¬h | ¬e)Probold(¬e). But

Probold (e & ¬h | ¬e) = 0 and Probold(e & ¬h | e) = Probnew(e & ¬h), so Probold(e & ¬h) =Probnew(e & ¬h)Probold(e). Since Probold(e) < 1, Probold(e & ¬h) < Probnew(e & ¬h) unlessProbnew(e&¬h) = 0. In the former case, Probnew(e! h) = 1� Probnew(e&¬h) < 1� Probold(e& ¬h) = Probold(e ! h). In the latter case, Probold(e & ¬h) = 0 too, so Probold(e ! h) = 1.

27. The correlation between receptivity and the former two-way correlation may itself be two-way (Rap , (p , Bap)), if receptivity is the only way of setting up the first correlation, butthe other direction does not matter for present purposes.

28. For example, in the set theory ZFC, the Axiom of Replacement is usually considered to bebetter established than the Axiom of Choice, but not because it has more experimentalsupport. For other objections to a purely holistic account of the confirmation of mathemat-ics as part of total science see Maddy 2007, pp. 314–17.

29. We are not concerned with more radical redefinitions of the words in the sentence.30. For an introduction to psychological work on the cognitive value of the imagination, the

case specifically relevant to the examples in Section 3, see Harris 2000.

ReferencesBarwise, J. (ed.) 1977. Handbook of Mathematical Logic. Amsterdam: North-Holland.Boolos, G. 1971. ‘The iterative conception of set’, The Journal of Philosophy 68, pp. 215–32.Crossley, J. N., Ash, C. J., Brickhill, C. J., Stillwell, J. C., and Williams, N. H. 1972. What isMathematical Logic? Oxford: Oxford University Press.

Devitt, M. 2005. ‘There is no a priori’, in Steup and Sosa 2005, pp. 105–15.Hallett, M. 1984. Cantorian Set Theory and Limitation of Size. Oxford: Clarendon Press.Harris, P. L. 2000. The Work of the Imagination. Oxford: Blackwell.Jackson, F. 1982. ‘Epiphenomenal qualia’, The Philosophical Quarterly, 32, pp. 127–36.Kripke, S. 1980. Naming and Necessity. Oxford: Blackwell.Maddy, P. 2007. Second Philosophy: A Naturalistic Method. Oxford: Oxford University Press.Maddy, P. 2011. Defending the Axioms: On the Philosophical Foundations of Set Theory. Oxford:Oxford University Press.

Quine, W. V. 1951. ‘Two dogmas of empiricism’, Philosophical Review, 60, pp. 20–43.Quine, W. V. 1981. Theories and Things. Cambridge, Mass.: The Belknap Press of Harvard

University Press.Russell, B. A.W. 1919. Introduction to Mathematical Philosophy. London: George Allen andUnwin.Shoenfield, J. R. 1977. ‘Axioms of set theory’, in Barwise 1977, pp. 321–44.Steup, M. and Sosa, E. (eds) 2005. Contemporary Debates in Epistemology. Oxford: Blackwell.Williamson, T. 2000. Knowledge and its Limits. Oxford: Oxford University Press.Williamson, T. 2007. The Philosophy of Philosophy. Oxford: Blackwell.


Index

a prioriand beliefs about justificational status 221challenge to from experimentalphilosophy 48–9, 50–2, 68, 92, 95–7, 286;see also intuition, experimental challenges to

contingent 88n9, 104, 242explanation of the reliability of a prioribelief 76–7, 80–2

high-grade 130and intuition 34–5low-grade, 111, 115–16negative conception of 77, 104–5particularism and 221phenomenology of 35, 73–4positive conception of 77, 104–5privileged position of 206pure/impure 98, 101–2status of classification judgments 38weak conception of 252

a priori and a posteriori, the distinctionbetween 35, 105, 254, 265, 266–7, 294–7,300, 309

Alston, William 207, 209, 211, 213–14analytic/analyticity 82, 116, 124–5, 135–6, 156,

278, 286–7, 303–4arbitrary partiality argument 211–8armchair methodology/philosophy 12, 37, 45,

48, 105, 111–12, 115–16, 125, 129–30

Bayesian epistemology 228, 306–7Bealer, George 41n12, 73belief-forming process 213, 226, 233, 256,

258–9individuating 257–8justification of the rationality of 239–41

the Benacerraf challenge to the a priori 67–8Berkeley, George 206, 210Boghossian, Paul 136, 287BonJour, Laurence 2, 73, 84, 97, 100–1,

212–13bootstrapping reasoning 194, 235–7, 240–1Burge, Tyler 75, 81, 84, 158–63

Cartesian perspective 206, 250–1Casullo, Albert 100causal-enabling vs. causal-explanatory

roles 175–6Chisholm, Roderick 263classifiers 20–1, 25coherence 186

common sense 142, 189–90, 192, 196evolution and 192–3

competence 30–2, 112–13, 115–17, 127–8,161, 180n8, 187–8, 195, 198–9

concept-constituting beliefs and inferences 69,72–5, 80, 140–2, 145

concepts 21, 52abbreviative 143–8concept possession 28–30, 69constitutive views of 30–1, 71grounded 261–2, 280–1justified 261–2teleology and 140–2theories of 28–30

conceptual necessity 117; see alsoconceptually-grounded necessary truths

conceptually-grounded necessary truths 115,118, 127, 129

content determining conditions; seeconcept-constituting beliefs and inferences

contingent a priori knowledge/justification; see apriori, contingent

debunking explanations 96, 114, 117defeasibility 53–5, 78defeaters 160–1, 178–9, 234, 257–8

experimental philosophy as a defeater for the apriori 50–2, 59–60, 94

undermining 50, 172Descartes, Rene 2, 98, 215–6

see also Cartesian perspectivedetermination theory 69–70Devitt, Michael 67, 277disagreement 49–50, 214doubt 216–17Dretske, Fred 24Dummett, Michael 239–40

entitlement 75, 159–60, 163epistemic possibility 31, 118, 123, 125empirical justification 231–2, 241

insufficiencies of 97–8; see also skepticism, apriori is required to reject skepticism

empirically indefeasible 144–6, 251–3empiricism 24, 37–8, 200n5, 261–2evidence 13–15, 19, 32, 34, 94, 172, 190, 219,

228, 262, 264, 307evidence, first-order vs. second-order questions

of 13, 18, 96–7evidence possession 33–4

experience 73, 77, 187, 200n5, 227, 293–4enabling vs. warranting role of 52–3, 57–8,

159, 175, 266, 271, 293, 297–8as ground of concepts 161, 280; see also

concepts, groundedand necessity 99–101negative vs. positive dependence on 54–5, 94nonevidential epistemic role of 261–2, 266–9

experimental philosophy 12, 16, 46–7, 68, 285as assisting a priori methods 86as improvement of rational abilities 58–62and modality 102–5as modifier of philosophical practice 34, 93as sorter of good intuitions from bad 86, 93–5

externalismand justification 77–8and knowledge 254, 261

fallibilism 35, 75–6, 214, 242foundations 187, 193, 206, 214

Gettier cases 3, 11, 17, 26, 36, 47, 57, 62Goldman, Alvin 23, 36, 71Grice, Paul 54, 135

Hawthorne, John 27, 104, 249, 254–61Horwich, Paul 136, 145Hume, David 206, 210

Ichikawa, J. 36, 56immunity to empirical revision 153; see also

empirically indefeasibleimplicit conceptions 70–1, 76inference 24, 69, 99, 140–1, 143, 149–50,

234–5, 287to the best explanation 78, 96, 111–15,

195, 237inferential justification 53–4, 100–1, 195, 197–8internalism 77–8, 197introspection 37, 84, 169, 173, 187, 207, 213,

218, 293intuitiona priori intuition cannot be entirely

unreliable 31, 215, 218collective intuition and evidence 32–4cultural rationally competent 199definition of 11, 13, 188, 284–5as distinct from experience 188evidential status of 23, 284–6, 303examples of 67, 115, 215, 221experimental challenges to 16–19, 285and experimental philosophy 12, 86–7intuitional methodology 22, 37–9rational 73rationally competent 188, 197, 199as a source of knowledge 197sources of unreliability of 28–30

Jackson, Frank 31, 97, 155Jenkins, Carrie 89n25, 105, 249, 261–5,

274–5, 285justification 93, 198, 206, 230, 234, 263–4, 270and competence; see competencea posteriori 35, 98, 178, 283–4a priori 73–4, 77, 83, 136, 146, 159, 218,250–1, 275

doxastic 55–6, 231foundationalism about 186–7; see alsofoundations

propositional 55–6, 60, 231, 234relationship between a priori and a posteriori;see a priori and a posteriori, the distinctionbetween

Kant, Immanuel 2, 99, 242, 262Kitcher, Philip 54, 249, 251–3, 255, 259knowledgea posteriori 98, 178, 265, 292, 300, 307–8a priori 45, 52, 69, 75, 136, 159, 162–3,250–1, 254–6, 261, 275, 281, 292

safety account of 254, 299senses of 27socio-historical conception of 259–60testimonial 159–61; see also testimonythat is both empirical and a priori 281

Kornblith, Hilary 24, 277Kripke, Saul 17, 102, 118, 120–2, 124, 293, 300

Lewis, David 156logic 303–5a priori justification of 141, 148–53function of logical concepts 142, 148–53truths of 136–9

Machery, Edouard 17, 19, 47Maddy, Penelope 276, 281–3mathematics 93, 136, 142–3, 154–5, 261–2,

300, 301–3meaning 20, 25, 135, 155–6, 279modal knowledge, see necessity, knowledge of

necessary truthsmodal invigoration 102Moore, G.E. 188–90, 194, 201n8, 240, 241Moser, Paul 263–4

Nagel, Jennifer 27–8natural kind terms 115, 120–1naturalismnaturalistic challenges to the a priori 38, 275,278, 282–4, 308

naturalized epistemology 277, 309as a stance 276versions of 275–8

necessityand aims of philosophy 73–4, 283–4

314 INDEX

explaining metaphysical necessities in terms ofconceptual necessities 118, 120, 125

justification of belief in/knowledge ofnecessary truths 115–16, 266–8, 298–9

relation to the a priori 35–6, 73–4, 98–9, 100Neta, Ram 158, 165, 167–8, 173–4non-empirical reasoning 233

observation 97, 193–4, 294, 307–8; see alsoexperience

Papineau, David 283–8Peacocke, Christopher 69–71, 75, 77Plantinga, Alvin 35, 72, 73, 207, 209–10,

213–14, 222, 223n12, 224n16, 277Plato 24pragmatic encroachment 95–6pragmatism 40n9, 138Pryor, James 226, 240–1, 244n19

quantum disjunction 151–3Quine, W.V.O. 140, 146, 306on analyticity 135revisability thesis 134, 136–8on skepticism about the a priori 136, 278

rationalism 24, 73, 95, 200n5reason-independent causation 166Reformed epistemology 208–9Reid, Thomas 206–7, 210–12, 219reliabilism 253, 256reliability 15–19, 31–2, 78, 86, 215, 238, 297,

299; see also defeatersRussell, Bertrand 215, 303

seemings 13, 73, 188, 197–8sets 301–3Silins, Nicholas 241simulation 28, 36skeptical hypothesis 117, 227a priori justification for rejecting a skepticalhypothesis 228, 236–9

skepticism 2, 189, 219a priori is required to reject skepticism 6,228–31

about evidential status of intuitions19, 22

arguments for 198, 226–8avoiding 206conditional 207, 210

Sosa, Ernest 49Speckled Hen, problem of the 187Stich, Stephen 3, 12Strawson, P. 135, 202n15Summerfield, Donna 94synthetic a priori 282, 286–7

taking experience at face value 233–4,239–40; see also entitlement

testimony 159–61, 163, 165, 171–2, 257fast-track model of 176–9

Truetemp 17, 29, 47, 68

understanding 71, 73, 77, 80, 83–5, 126–7,160–1, 188, 196, 199, 304

warrant 35, 53, 55, 130, 158–9, 161–2, 175,177, 252, 259; see also justification

Weatherson, Brian 75, 226Weinberg, Jonathan 3, 17–19, 47, 68, 86White, Roger 228, 238wide reflective equlibrium 96, 111, 113,

115, 127Williamson, Timothy 14, 35–6, 55, 105,

124–5, 249on a priori/a posteriori distinction 36, 266–7,

294–7on armchair methodology 28, 125–30, 265–6his challenge to conceptual truth andanalyticity 79–82, 125–30

Wittgenstein, Ludwig 140, 190–2, 194Wolterstorff, Nicholas 207, 208–9, 211,

213–14, 220Wright, Crispin 234, 241

INDEX 315

albert casullo, joshua c. thurow-the a priori in philosophy-oxford university press (2013)

Documents

university of oxford

university of nebraska

hillprofessor of philosophy

rights department

blank table of contentslist

experimental philosophyjoshua

moral rights

philosophythis page