essay 1 autopoeisis

8/9/2019 Essay 1 Autopoeisis

1/11

Essay 1

Autopoiesis and a

Biology of Intentionality

Francisco J. Varela

CREA, CNRSEcole Polytechnique,

Paris, France.

Parts of this text have been published in (Varela 1991).

4


2/11

Biology of Intentionality Francisco J. Varela

1.1 Introduction

As everybody here knows, autopoiesis is a neolo-gism, introduced in 1971 by H. Maturana and my-self to designate the organization of a minimal livingsystem. The term became emblematic of a view ofthe relation between an organism and its medium,where its self constituting and autonomous aspectsare put at the center of the stage. From 1971, untilnow much has happened to reinforce this perspec-

tive. Some of the developments have to do with thenotion of autopoiesis itself in relation to the cellularorganization and the origin of life. Much more hasto do with the autonomy and self-organizing qual-ities of the organism in relation with its cognitiveactivity. Thus in contrast to the dominant cogni-tivist, symbol-processing views of the 70s today wewitness in cognitive science a renaissance of the con-cern for the embeddedness of the cognitive agent,natural or artificial. This comes up in various labelsas nouvelle-AI (Brooks 1991c), the symbol ground-ing problem (Harnad 1991), autonomous agents inartificial life (Varela & Bourgine 1992), or situatedfunctionality (Agree 1988), to cite just a few self-explanatory labels used recently.

Any of these developments could merit a full talk;obviously I cannot do that here. My intentionrather, profiting from the position of opening thisgathering, is to try to indicate some fundamentalor foundational issues of the relation between au-topoiesis and perception. Whence the title of mytalk: a biology of intentionality. Since the crisis ofclassical cognitive science has thrown open the issueof intentionality, in my eyes autopoiesis provides anatural entry into a view of intentionalty that is

seminal in answering the major obstacles that havebeen addressed recently. Ill came back to that atthe end. Let me begin at the beginning.

1.2 Cognition and

Minimal Living Systems

1.2.1 Autopoiesis as the

skeletal bio-logic

The bacterial cell is the simplest of living sys-tems because it possesses the capacity to produce,

through a network of chemical processes, all thechemical components which lead to the constitutionof a distinct, bounded unit. To avoid being triv-ial, the attribute living in the foregoing descriptionmust address the process that allows such consti-tution, not the materialities that go into it, or an

enumeration of properties. But what is this basicprocess? Its description must be situated at a veryspecific level: it must be sufficiently universal to al-low us to recognize living systems as a class, withoutessential reference to the material components. Yetat the same time it must not be too abstract, thatis, it must be explicit enough to allow us to see suchdynamical patterns in action in the actual living sys-tem we know on earth, those potentially to be foundin other solar systems, and eventually those createdartificially by man. As stated by the organizer of ameeting on artificial life: Only when we are able toviewlife-as-we-know-itin the larger context oflife-as-it-could-bewill we really understand the natureof the beast (Langton 1989b, p. 2).

Contemporary cell biology has made it possiblefor some years now to put forth the characteriza-tion of this basic living organizationa bio-logicas that of an autopoieticsystem (from Greek: self-producingMaturana & Varela 1980; Varela et al.1974). An autopoietic systemthe minimal livingorganizationis one that continuously produces the

components that specify it, while at the same timerealizing it (the system) as a concrete unity in spaceand time, which makes the network of productionof components possible. More precisely defined: Anautopoietic system is organized (defined as unity) asa network of processes of production (synthesis anddestruction) of components such that these compo-nents:

(i) continuously regenerate and realize the networkthat produces them, and

(ii) constitute the system as a distinguishable unityin the domain in which they exist.

Thus, autopoiesis attempts to capture the mech-anism or process that generates the identity of theliving, and thus to serve as a categorical distinc-tion of living from non-living. This identity amountsto self-produced coherence: the autopoietic mecha-nism will maintain itself as a distinct unity as longas its basic concatenation of processes is kept in-tact in the face of perturbations, and will disappearwhen confronted with perturbations that go beyonda certain viable range which depends on the specificsystem considered. Obviously, all of the biochemi-cal pathways and membrane formation in cells, canbe immediately mapped onto this definition of au-topoiesis.

A different exercisewhich I do not pursue hereat allis to see how this basic autopoietic orga-nization, present at the origin of terrestrial life(Fleischaker 1988), becomes progressively complexi-

5


3/11


fied though reproductive mechanisms, compartmen-talization, sexual dimorphism, modes of nutrition,symbiosis, and so on, giving rise to the variety ofpro- and eukaryotic life on Earth today (Margulis1981; Fleischaker 1988). In particular, I take herethe view that reproduction is not intrinsic to theminimal logic of the living. Reproduction must beconsidered as an added complexification superim-posed on a more basic identity, that of an autopoi-etic unity, a complexification which is necessary dueto the constraints of the early conditions on a turbu-lent planet. Reproduction is essential for the viabil-ity of the living, but only when there is an identitycan a unit reproduce. In this sense, identity haslogical and ontological priority over reproduction,although not historical precedence.

We do not pursue here these historical complexi-fications, neither do I pursue another equally perti-nent empirical question: Can a molecular structuresimpler than the already intricate bacterial cell, sat-isfy the criteria of autopoietic organization? Thisquestion can be answered by two complementary ap-

proaches: (1) simulation and (2) synthesis of mini-mal autopoetic systems. There are advances in bothfronts. As to the first, there some new results inthe burst of work in artificial life, partly extend-ing our early simulations in tesselation automataof (Varela et al. 1974). The second front, takesthe form of a new cell-centered approach to theorigin of life which seeks chemical embodiments ofminimal autopoietic systems. In fact, the encapsu-lation of macromolecules by lipid vesicles has beenactively investigated as a promising candidate foran early cell (Deamer & Barchfeld 1982; Lazcano1986; Baeza et al. 1987; see Deamer 1986). Luisi

& Varela (1989) make the case that a reverse micel-lar system can come close to the mark for being aminimal autopoietic system. In particular, they dis-cuss the case of a reverse micellar system hosting inits aqueous core a reaction which leads to the pro-duction of a surfactant, which is a boundary for thereverse micellar reaction. The interest of this caseis that much is known about these chemical systemsmaking it possible to actually put into operation aminimal autopoietic system. But I must leave thesefascinating issues to return to my chosen topic here.

1.2.2 Identity of the living

and its world

Autopoiesis addresses the issue of organism as aminimal living system by characterizing its basicmode of identity. This is, properly speaking, toaddress the issue at an ontological level: the ac-cent is on the manner in which a living system be-comes a distinguishable entity, and not on its spe-cific molecular composition and contingent histori-

cal configurations. For as long as it exists, the au-topoietic organization remains invariant. In otherwords, one way to spotlight the specificity of au-topoiesis is to think of it self-referentially as thatorganization which maintains the very organizationitself as an invariant. The entire physico-chemicalconstitution is in constant flux; the pattern remains,and only through its invariance can the flux of real-izing components be ascertained.

I have addressed here only the minimal organiza-tion that gives rise to such living autonomy. As Ihave said, my purpose is to highlight the basic bio-logic which serves as the foundation from which the

diversity visible in current organisms can be consid-ered: only when there is an identity can elaborationsbe seen as family variations of a common class of liv-ing unities. Every class of entities has an identitywhich is peculiar to them; the uniqueness of the liv-ing resides in the kind of organization it has.

Now, the history of biology is, of course,marred by the traditional opposition betweenthe mechanist/reductionists on the one hand andholist/vitalists on the other, a heritage from the bi-ological problem-space of the XIXth century. Oneof the specific contributions of the study of self-organizing mechanismsof which autopoiesis is aspecific instanceis that the traditional oppositionbetween the component elements and the globalproperties disappears. In the simple example of thecellular automaton illustrated above, it is preciselythereciprocal causalitybetween the local rules of in-teractions (i.e. the components rules, which are akinto chemical interactions) and the global propertiesof the entity (its topological demarcation affectingdiffusion and creating local conditions for reaction)which is in evidence. It appears to me that this re-ciprocal causality does much to evacuate the mecha-nist/vitalist opposition, and allows us to move into a

more productive phase of identifying variousmodesof self-organization where the local and the globalare braided together explicitly through this recip-rocal causality. Autopoiesis is a prime example ofsuch dialectics between the local component levelsand the global whole, linked together in reciprocal

6


4/11


relation through the requirement of constitution ofan entity that self-separates from its background. Inthis sense, autopoiesis as the characterization of theliving does not fall into the traditional extremes ofeither vitalism or reductionism.

A second, complementary dimension of basic bio-logic that is central to focus our discussion is thenature of the relationship between autopoietic au-tonomous unities and their environment. It is ex-hypothesis evident that an autopoietic system de-pends on its physico-chemical mileu for its conserva-tion as a separate entity, otherwise it would dissolveback into it. Whence the intriguing paradoxicalityproper to an autonomous identity: the living systemmust distinguish itself from its environment, whileat the same timemaintaining its coupling; this link-age cannot be detached since it is against this veryenvironment from which the organism arises comesforth. Now, in this dialogic coupling between theliving unity and the physico-chemical environment,the balance is slightly weighted towards the livingsince it has the active role in this reciprocal cou-

pling. In defining what it is as unity, in the verysame movement it defines what remains exterior toit, that is to say, its surrounding environment. Acloser examination also makes it evident that thisexteriorization can only be understood, so to speak,from the inside: the autopoietic unity creates aperspective from which the exterior is one, whichcannot be confused with the physical surroundingsas they appear to us as observers, the land of phys-ical and chemical laws simpliciter, devoid of suchperspectivism.

In our practice as biologists we switch betweenthese two domains all the time. We use and ma-

nipulate physico-chemical principles and properties,while swiftly shifting to the use ofinterpretationandsignificance as seen from the point of view of theliving system. Thus a bacteria swimming in a su-crose gradient is conveniently analyzed in terms ofthe local effects of sucrose on membrane permeabil-ity, medium viscosity, hydromechanics of flagellarbeat, and so on. But on the other hand the sucrosegradient and flagellar beat are interesting to ana-lyze only because the entire bacteria points to suchitems as relevant: their specific significance as com-ponents of feeding behavior is only possible by thepresence and perspective of the bacteria as a totality.

Remove the bacteria as a unit, and all correlationsbetween gradients and hydrodynamic properties be-come environmental chemical laws, evident to us asobservers but devoid of any special significance.

I have gone into this lengthy harangue becauseI believe that this truly dialectical relationship is a

key point. In fact, it might appear as so obvious thatwe dont appreciate its deep ramifications. I meanthe important distinction between the environmentof the living system as it appears to an observer andwithout reference to the autonomous unitywhichwe shall call hereafter simply theenvironmentandthe environment for the system which is definedin the same movement that gave rise to its iden-tity and that only exists in that mutual definitionhereinafter the systems world.

The difference between environment and world isthe surplus of significationwhich haunts the under-standing of the living and of cognition, and whichis at the root of how a self becomes one. In otherwords, this surplus is the mother of intentionality.It is quite difficult in practice to keep in view the di-alectics of this mutual definition: neither rigid isola-tion, nor simple continuity with physical chemistry.In contrast, it is easy to conflate the units worldwith its environment since it is so obvious that weare studying this or that molecular interaction in thecontext of an autonomous cellular unit, and hence

to miss completely the surplus addedby the organ-isms perspective. There is no food significance insucrose except when a bacteria swims upgradientand its metabolism uses the molecule in a way thatallows its identity to continue. This surplus is obvi-ously not indifferent to the regularities and texture(i.e. the laws) that operate in the environment,that sucrose can create a gradient and traverse acell membrane, and so on. On the contrary, the sys-tems world is built on these regularities, which iswhat assures that it can maintain its coupling at alltimes.

What the autopoietic system doesdue to its

very mode of identityis to constantly confront theencounters (perturbations, shocks, coupling) withits environment and treat them from a perspec-tive which is not intrinsic to the encounters them-selves. Surely rocks or crystal beads dont beckonsugars gradients out of all the infinite possibili-ties of physico-chemical interactions as particularlymeaningfulfor this to happen a perspective froman actively constituted identity is essential. It istempting, at this point, to slide into some vaporousclouds about meaning reminiscent of the worstkind of vitalism of the past or informational jargonof the present. What I emphasize here is that what

is meaningful for an organism is precisely given byits constitution as a distributed process, with an in-dissociable link between local processes where an in-teraction occurs (i.e. physico-chemical forces actingon the cell), and the coordinated entity which is theautopoietic unity, giving rise to the handling of its

7


5/11


environment without the need to resort to a centralagent that turns the handle from the outsidelikean elan vitalor a pre-existing order at a particu-lar localizationlike a genetic program waiting tobe expressed.

I would like to rephrase this basic idea by turn-ing it upside down as it were. The constant bring-ing forth of signification is what we may describeas a permanent lack in the living: it is constantlybringing forth a signification that is missing, notpre-given or pre-existent. Relevance must be pro-videdex nihilo: distinguish relevant from irrelevantmolecular species, follow a gradient uphill and notdownhill, increase the permeability to this ion andnot to that one, and so on. There is an inevitablecontretemps between an autonomous system and itsenvironment: there is always something which thesystem must furnish from its perspective as a func-tioning whole. In fact, a molecular encounter ac-quires a significance in the context of the entireop-erating system and of many simultaneous interac-tions.

The source for this world-making is always thebreakdowns in autopoiesis, be they minor, likechanges in concentration of some metabolite, ormajor, like disruption of the boundary. Due tothe nature of autopoiesis itselfillustrated in themembrane repair of the minimal simulated exampleaboveevery breakdown can be seen as the initi-ation of an action on what is missing on the partof the system so that identity might be maintained.I repeat: no teleology is implied in this so that:thats what the self-referential logic of autopoiesisentails in the first place. The action taken will bevisible as an attempt to modify its worldchange

from place of different nutrients, increase in the flowof a metabolite for metabolic synthesis, and so on.

In brief, this permanent, relentless action on whatis lacking becomes, from the observer side, the ongo-ingcognitiveactivity of the system, which is the ba-sis for the incommensurable difference between theenvironment within which the system is observed,and the world within which the system operates.This cognitive activity is paradoxical at its veryroot. On the one hand the action that brings fortha world is an attempt to reestablish a coupling withan environment which defies the internal coherencethrough encounters and perturbations. But such ac-

tions, at the same time, demarcate and separate thesystem from that environment, giving rise to a dis-tinct world.

The reader may balk at my use of the term cog-nitive for cellular systems, and my cavalier slidinginto intentionality. As I said above, one of my main

points here is that we gain by seeing the continu-ity between this fundamental level of self and theother regional selves, including the neural and lin-guistic where we would not hesitate to use the wordcognitive. I suppose others would prefer to intro-duce the word information instead. Well, thereare reasons why I believe this even more problem-atic. Although it is clear that we describe an Xthat perturbs from the organisms exteriority, X isnot information. In fact, for the organism only isa that, a something, a basic stuff to in-form fromits own perspective. In physical terms there is stuff,but it is for nobody. Once there is bodyeven inthis minimal formit becomes in-formed for a self,in the reciprocal dialectics I have just explicated.Such in-formation is never a phantom significationor information bits, waiting to be harvested by asystem. It is a presentation, an occasion for cou-pling, and it is in this entre-deux that significationarises (Varela 1979, 1988; Castoriadis 1987).

Thus the term cognitive has two constitutive di-mensions: first itscouplingdimension, that is, a link

with its environment allowing for its continuity asindividual entity; secondby stretching language, Iadmititsimaginarydimension, that is, the surplusof significance a physical interaction acquires due tothe perspective provided by the global action of theorganism.

1.3 Perception-action and

basic neuro-logic

1.3.1 Operational closure of the

nervous system

In the previous Section, I have presented the fun-damental interlock between identity and cognitionas it appears for a minimal organism. In this Sec-tion I want to show how the more traditional levelof cognitive properties, involving the brains of mul-ticellular animals, is in some important sense thecontinuation of the very same basic process.

The shift from minimal cellularity to organismwith nervous system is swift, and skips the com-plexity of the various manners in which multicellu-lar organisms arise and evolve (Margulis & Schwartz1988; Buss 1987; Bonner 1988). This is a transition

in units of selection, and one that implicates the so-matic balance of differentiated populations of cellsin an adult organism, as well as crafty developmentpathways to establish a bodily structure. As Busshas stated recently: The evolution of developmentis the generation of a somatic ecology that mediates

8


6/11


potential conflicts between cell and the individual,while the organism is simultaneously interacting ef-fectively with the extrasomatic environment (Buss1987).

For most vertebrates, this somatic ecology isbound together through the network of lympho-cytes that constitute the core of the immune sys-tem. Again, a discussion of an immunological selfis not my purpose here. I cannot resist the temp-tation, nevertheless, to point out, for completenesssake, that elsewhere I have presented in extenso anetwork approach to the immune system and its rolein the establishment of a flexible cellular/molecularself during the ontogeny of mammals (see Varelaet al. 1988; Varela & Coutinho 1991). In my viewthis identity is not, as traditionally stated, a de-marcation of self as defense against the non-self ofinvading antigens. It is a self-referential, positive as-sertion of a coherent unitya somatic ecologymediated through free immunoglobulins and cellularmarkers in a dynamical exchange. Immune reactionsagainst infections, although clearly important, are

mediated by a peripheral immune system, a dif-ferent sub-population of lymphocytes mobilized notthrough network but clonal expansion mechanisms,like a reflex reactivity acquired through evolution.But enough of this excursus. For my purposes hereI will expeditiously assume the identity of a mul-ticellular organism, distinctly different from an au-topoietic minimal entity in its mode of identity, butsimilar in that it demarcates an autonomous entityfrom its environment.

Now, whats the specific place of the nervoussystem in the bodily operation of a multicellular?Whenever motionis an integral part of the lifestyle

of a multicellular, there is a corresponding develop-ment of a nervous system linking effector (muscles,secretion) and sensory surfaces (sense organs, nerveendings). The fundamental logic of the nervous sys-tem is that of coupling movements with a streamof sensory modulations in a circular fashion. Thenet result are perception-action correlations arisingfrom and modulated by an ensemble of interveningneurons, theinterneuronnetwork. Correspondingly,neurons are unique among the cells of a multicellu-lar organism in their axonal and dendritic ramifica-tions permitting multiple contacts and extending forlarge distances (relative to cellular soma sizes) pro-

viding the essential medium for this intra-organismicsensor-effector correlation.

Contrary to current habit, I wish to emphasizefrom the start the situatednessof this neuro-logic:the state of activity of sensors is brought aboutmosttypically by the organisms motions. To an impor-

tant extent, behavior is the regulation of perception.This does not exclude, of course, independent per-turbations from the environment. But what is typ-ically described as a stimulus in the laboratory,a perturbation which is deliberately independent ofthe animals ongoing activity, is less pertinent (out-side the laboratory) for understanding the biologyof cognition.

The perceptuo-motor coherencies we describe ex-ternally as behavior disguises the arising, within theinterneuron net, of a large sub-setanensemble asis usually saidof transiently correlated neurons.These ensembles are both the source and the resultfrom the activity of the sensory and effector surfaces.What changes is the amount of mediating interneu-rons, and the specific architecture of the respectivenervous system, containing various cortical regions,layers and nuclei. In humans some 1011 interneu-rons interconnect some 106 motoneurons which re-late to 107 sensory neurons distributed in receptorsurfaces throughout the body. This is a ratio of10 : 100, 000 : 1 of interneurons mediating the cou-

pling of sensory and motor surfaces. The rise anddecay of neuronal self-organization, say, in the mod-estAplysiasiphon withdrawal (Zecevicet al. 1989)is all the more valid in larger brains. Thus for in-stance a study in the cat (John et al. 1986) findsthat 5100 million neurons are active throughout thebrain during a simple visuo-motor task of pressinga lever. Such neural assemblies arise in a patchworkof regional areas, evincing the enormous distributedparallelism proper to vertebrate brains.

The neuronal dynamics underlying a perceptuo-motor task is, then, a network affair, a highly coop-erative, two-way system, and not a sequential stage-

to-stage information abstraction. The dense inter-connections among its sub-networks entails that ev-ery active neuron will operate as part of a large anddistributed ensemble of the brain, including localand distant regions. For example, although neuronsin the visual cortex do have distinct responses tospecific features of the visual stimuli (position, di-rection, contrast, and so on), these responses occuronly in an anesthetized animal with a highly sim-plified (internal and external) environment. Whenmore normal sensory conditions are allowed, and theanimal is studied awake and behaving, it has becomeincreasingly clear that the stereotyped neuronal re-

sponses to features are highly labile and contextsensitive. These have been shown, for example, forthe effect of bodily tilt or auditory stimulation. Fur-thermore, the response characteristics of most neu-rons in the visual cortex depend directly on otherneurons localized far from their receptive fields (see

9


7/11


e.g. Allman et al. 1985); even a change in posture,while preserving the same identical sensorial stimu-lation, alters the neuronal responses, demonstratingthat even the supposedly downstream motorium isin resonance with the sensorium (Abeles 1984).

If I may continue to use vision as an example,I can take the previous discussion up one levelof generalization, to note that in recent years re-search has become the study, not of centralizedreconstruction of a visual scene for the benefitof an ulterior homunculus, but that of a patch-work of visual modalities, including at least form(shape, size, rigidity), surface properties (color,texture, specular reflectance, transparency), three-dimensional spatial relationships (relative positions,three-dimensional orientation in space, distance),and three-dimensional movement (trajectory, rota-tion). It has become evident that these different as-pects of vision are emergent properties of concurrentsub-networks, which have a degree of independenceand even anatomical separability, but cross-correlateand work together so that a visual percept is this

coherency.This kind of architecture is strongly reminiscentof a society of agents to use Minskys (1987)metaphor. This multi-directional multiplicity iscounterintuitive but typical of complex systems.They are counterintuitive because we are used to thetraditional causal mode of input-processing-outputdirectionality. Nothing in the foregoing descriptionsuggests that the brain operates as a digital com-puter, with stage-by-stage information processing;such popular descriptions for a system with this typesimply goes against the grain. Instead, to the net-work and parallel architecture corresponds a differ-

ent kind of operation: there is a relaxation time ofback and forth signals until everybody is settled intoa coherent activity. Thus the entire cooperative ex-ercise takes a certain time to culminate, and this isevident in that, behaviorally, every animal exhibitsa natural temporal parsing. In the human brainthis flurry of cooperation typically takes about 200-500 msec, the nowness of a perceptuo-motor unity.Contrary to what it might seem at first glance eitherethologically or in our own introspection, cognitivelife is not a continual flow, but is punctuated by be-havioral patterns which arise and subside in chunksof time. This insight of recent neuroscienceand

cognitive science in general in factis fundamen-tal for it relieves us from the tyranny of searchingfor a centralized, homuncular quality to a cognitiveagents normal behavior.

Let me backtrack a moment and reframe our dis-cussion on cognitive self alongside that of a mini-

mal molecular self. I am claiming that contempo-rary neuroscienceslike cell biology for the case ofthe living organizationgives enough elements toconceive of the basic organization for a cognitiveself in terms of the operational (not interactional!)closureof the nervous system (Maturana & Varela1980; Varela 1979). I speak of closure to high-light the self-referential quality of the interneuronnetwork and of the perceptuo-motor surfaces whosecorrelations it subserves. The qualification opera-tional emphasizes that closure is used in its math-ematical sense of recursivity, and not in the senseof closedness or isolation from interaction, whichwould be, of course, nonsense. More specifically,the nervous system is organized by the operationalclosure of a network of reciprocally related modularsub-networks giving rise to ensembles of coherentactivity such that:

(i) they continuously mediate invariant patterns ofsensory-motor correlation of the sensory and ef-fector surfaces;

(ii) give rise to a behavior for the total organismas a mobile unit in space.

The operational closure of the nervous systemthen brings forth a specificmodeof coherence, whichis embedded in the organism. This coherence is acognitive self: a unit of perception/motion in space,sensory-motor invariances mediated through the in-terneuron network. The passage to cognition hap-pens at the level of a behavioral entity, and not, asin the basic cellular self, as a spatially bounded en-tity. The key in this cognitive process is the nervoussystem through its neuro-logic. In other words thecognitive self is the manner in which the organism,through its own self-produced activity, becomes adistinct entity in space, but always coupled to itscorresponding environment from which it remainsnevertheless distinct. A distinct coherent self which,by the very same process of constituting itself, con-figures an external world of perception and action.

1.3.2 Cognitive self and

perceptual world

The nature of the identityof the cognitive self justdiscussed is, like that of the basic cellular self, one

of emergence through a distributed process. Theemergent properties of an interneuron network are,however, quite different in their properties and likelyto be much more rich in possibilities. What I wishto emphasize here is recent insights into the easi-ness with which lots of simple agents having sim-ple properties may be brought together, even in a

10


8/11


haphazard way, to give riseto what appears to anobserver a purposeful and integrated whole, withoutthe need for a central supervision. We have alreadytouched on this theme when discussing the nature ofthe autopoietic process and cellular automata mod-elling, and later when discussing the constant aris-ing and subsiding of neuronal ensemble underlyingbehavior. This issue of emergent properties is cru-cial for my whole argument here, although I basemy conclusions on contemporary studies from vari-ous biology-inspired complex systems (Farmer et al.1986; Langton 1989a).

What is particularly important is that we can ad-mit that (i) a system can have separate local compo-nents which (ii) there is no center or localized self,and yet the whole behaves as a unit and for the ob-server it is as if there was a coordinating agent vir-tually present at the center. This is what I meantwhen referring to aselflessselfwe could also pos-tulate a virtual self: a coherent global pattern thatemerges through simple local components, appear-ing to have a central location where none is to be

found, and yet essential as a level of interaction forthe behavior of the whole unity.The import of such current models, formalisms

and case studies of complex systems (i.e. emergentproperties through coordinated simple elements) is,in my eyes, quite profound for our understanding ofcognitive properties. It introduces an explicit alter-native to the dominant computationalist/cognitivisttradition in the study of cognitive properties forwhich the central idea is that of syntax independentof materiality which can support a semantics for anenvironment. This is also becoming more and moretrue for the researchers of artificial cognitive sys-

tems, as the current connectionist schools have madeit clear by now. What we find in brains is a promis-cuous tinkering of networks and sub-networks givingno evidence for a structured decomposition from topto bottom as is typical of a computer algorithm. Ac-cordingly, one of the first messages from the studyof artificial neural networks in modern connection-ist terms is the absence of a principled distinctionbetween software and hardware, or more, preciselybetween symbols and non-symbols. In fact, all wefind in modern artificial neural network machinesare relative activities between ensembles underly-ing the regularities we call their behavior or perfor-

mance. We may see that some of these ensemblesrecur regularly enough to describe them as beingprogram-like, but this is another matter. Althoughartificially built, such emerging ensembles cannot becalled computations in the sense that their dy-namics cannot be formally specifiable as the imple-

mentation of some high-level algorithm. Neural net-works even in their fine detail arenotlike a machinelanguage, since there is simply no transition betweensuch elemental operational atoms with a semanticsand the larger emergent level where behavior oc-curs. If there were, the classical computer wisdomwould immediately apply: ignore the hardware sinceit adds nothing of significance to the actual compu-tation (other than constraints of time and space). Incontrast, in distributed, network models these de-tails are precisely what makes a global effect possi-ble, and why they mark a sharp break with traditionin AI. Naturally this reinforces the parallel conclu-sions that apply to natural neural networks in thebrain, as we discussed before.

I have raised this point to caution the readeragainst the force of many years of dominance ofcomputationalism, and the consequent tendency toidentify the cognitive self with some computer pro-gram or high level computational description. Thiswill not do. The cognitive selfisits own implemen-tation: its history and its action are of one piece.

Now this demands that we clarify now the secondaspect of the self to be addressed: its mode of rela-tion with the environment.

1.3.3 Intentionality and neuro-logic

Ordinary life is necessarily one of situated agents,continually coming up with what to do facedwith ongoing parallel activities in their variousperceptuo-motor systems. This continual re-definition of what to do is not at all like a plan,stored in a repertoire of potential alternatives, butenormously dependent on contingency, improvisa-

tion, and more flexible than planning. Situatednessmeans that a cognitive entity hasby definitionaperspective. This means that it isnt related to itsenvironment objectively, that is independently ofthe systems location, heading, attitudes and his-tory. Instead, it relates to it in relation to theperspective established by the constantly emergingproperties of the agent itself and in terms of the rolesuch running redefinition plays in the systems entirecoherence.

Again, as we did for the minimal cellular self, wemust sharply differentiate between environment andworld. And again the mode of coupling is double.

On the one hand, such body-in-space clearly hap-pens through the interactions with the environmenton which it depends. These interactions are of thenature of macrophysical encounterssensory trans-duction, muscle force and performance, light and ra-diations, and so onnothing surprising about them.

11


9/11


However this coupling is possible only if the encoun-ters are embracedfrom the perspectiveof the systemitself. This amounts, quite specifically, to elaborat-ing a surplus signification relative to this perspec-tive. Whatever is encountered must be valued oneway or anotherlike, dislike, ignoreand acted onsome way or anotherattraction, rejection, neutral-ity. This basic assessment is inseparable from theway in which the coupling event encounters a func-tioning perceptuo-motor unit, and it gives rise toan intention (I am tempted to say desire), thatunique quality of living cognition (Dennett 1987).

Phrased in other terms, the nature of the environ-ment for a cognitive self acquires a curious status:it is that which lends itself (es lehnt sich an. . . ) toa surplus of significance. Like jazz improvisation,environment provides the excuse for the neuralmusic from the perspective of the cognitive sys-tem involved. At the same time, the organism can-not live without this constant coupling and the con-stantly emerging regularities; without the possibilityof coupled activity the system would become a mere

solipsistic ghost.For instance, light and reflectance (among manyother macrophysical parameters such as edges andtextures, but let us simplify for the argumentssake), lend themselves to a wide variety of colorspaces, depending on the nervous system involvedin that encounter. During their respective evolu-tionary paths, teleost fishes, birds, mammals, andinsects have brought forth various different colorspaces not only with quite distinct behavioral sig-nificance, but with different dimensionalities so thatit is not a matter of more or less resolution of col-ors (Thompson et al. 1992). Color is demonstra-

bly not a property that is to be recovered fromthe environmental information in some uniqueway. Color is a dimension that shows up only inthe phylogenetic dialogue between an environmentand the history of an active autonomous self whichpartly defines what counts as an environment. Lightand reflectances provide a mode of coupling, a per-turbation which triggers, which gives an occasionfor the enormous in-formative capacity of neuralnetworks for constituting sensori-motor correlationsand hence to put into action their capacity for imag-ining and presenting. It is only after all this hashappened, after a mode of coupling becomes regu-

lar and repetitive, like colors in oursand othersworlds, that we observers, for ease of language, saycolor corresponds to or represents an aspect of theworld.

A dramatic recent example of this surplus signifi-cance and the dazzling performance of the brain as

the generator of neural narratives is provided bythe technology of the so-called virtual realities.Visual perception and motions thus give rise to reg-ularities which are proper to this new manner ofperceptuo-motor coupling. What is most significantfor me here is theveracityof the world which rapidlysprings forth: we inhabit a body within this newworld after a short time of trying this new situa-tion (i.e. 15 minutes or so), and the experience is oftruly flying through walls or of delving into fractaluniverses. This is so in spite of the poor quality ofthe image, the low sensitivity of the sensors, and thelimited amount of interlinking between sensory andimage surfaces through a program that runs in a per-sonal computer. Through its closure, the nervoussystem is such a gifted synthesizer of regularitiesthat any basic material suffices as an environmentto bring forth a compelling world.

This very same strategy of the situatedness of anagent which is progressively endowed with richer in-ternal self-organizing modules is becoming a pro-ductive research program even for the very prag-

matically oriented field of artificial intelligence. Toquote R. Brooks, one of the main exponents of thistendency at some length:

I . . . argue for a different approach to cre-ating Artificial Intelligence:

We must incrementally build up thecapabilities of intelligent systems ateach step of the way and thus auto-matically ensure that the pieces andtheir interfaces are valid.

At each step we should build completeintelligent systems that we let loose in

the real world with real sensing andreal action. Anything less provides acandidate with which we can deludeourselves.

We have been following this approach andhave built a series of autonomous mobilerobots. We have reached an unexpectedconclusion (C) and have a rather radicalhypothesis (H).

C : When we examine very simple levelintelligence we find that explicit rep-resentations and models of the worldsimply get in the way. It turns out tobe better to use the world as its ownmodel.

H : Representation is the wrong unit ofabstraction in building the bulkiestparts of intelligent systems.

12


10/11


Representation has been the central issuein Artificial Intelligence work over the last15 years only because it has provided aninterface between otherwise isolated mod-ules and conference papers.

Brooks (1987, p. 1)

When the synthesis of intelligent behavior is ap-proached in such an incremental manner, with strict

adherence to the sensory-motor viability of an agent,the notion that the world is a source of informationto be represented simply disappears. The auton-omy of the cognitive self comes fully in focus. Thusin Brookss proposal his minimal creatures join to-gether various activities through a rule of cohab-itation between them. This is homologous to anevolutionary pathway through which modular sub-networks intertwined with each other in the brain.The expected result are more truly intelligent au-tonomous sense-giving devices, rather than brittleinformational processors which depend on a pre-given environment or an optimal plan.

It is interesting to note that in this paper Brooksalso traces the origin of what he describes as thedeception of AI to the tendency in AI (and inthe rest of cognitive science as well) to abstraction,i.e., for factoring out situated perception and motorskills. As I have argued here (and as Brooks arguesfor his own reasons), such abstraction misses theessence of cognitive intelligence, which resides onlyin its embodiment. It is as if one could separatecognitive problems in two parts: that which can besolved through abstraction and that which cannotbe. The second is typically perception-action andmotor skills of agents in unspecified environments.

When approached from this self-situated perspec-tive there is no place where perception could de-liver a representation of the world in the traditionalsense. The world shows up through the enactmentof the perceptuo-motor regularities. Just as thereis no central representation there is no central sys-tem. Each activity layer connects perception to ac-tion directly. It is only the observer of the Creaturewho imputes a central representation or central con-trol. The creature itself has none: it is a collectionof competing behaviors. Out of the local chaos oftheir interactions there emerges, in the eye of theobserver, a coherent pattern of behavior (Brooks1986, p. 11).

To conclude, the two main points that I have beentrying to bring into full view in this Section de-voted to the cognitive self are as follows. First, Ihave tried to spell out the nature of itsidentityas abody in motion-and-space through the operational

closure of the interneuron network. This activity isobservable as multiple sub-networks, acting in paral-lel and interwoven in complex bricolages,giving riseagain and again to coherent patterns which manifestthemselves as behaviors. Secondly, I have tried toclarify how this emergent, parallel and distributeddynamics is inseparable from the constitution of aworld,which is none other than the surplus of mean-ing and intentions carried by situated behavior. Ifthe links to the physical environment are inevitable,the uniqueness of the cognitive self is this constantgenesis of meaning. Or, again to invert the descrip-tion, the uniqueness of the cognitive self is this con-stitutivelackof signification which must be suppliedfaced with the permanent perturbations and break-downs of the ongoing perceptuo-motor life. Cogni-tion is action about what ismissing,filling the faultfrom the perspective of a cognitive self.

This view amounts to a biology of intentional-ity. In fact, it answers without ambiguity two keyproblems: the symbol (Harnad 1991) and the syn-tax grounding problems (Searle 1990). The first one

refers to the mystery of the origin of significationof natural symbols, since in the classical cognitivistoption there is an intrinsic need for an arbitrary se-mantic assignment. The answer provided by thisapproach is that the signification arises in the emer-gence of a viewpoint proper to the autonomous con-stitution of the organism at all its level, startingwith its basic autopoiesis. The syntax groundingproblem claims that all syntactic operations in asymbol system are observer-dependent. Our answeris precisely that the constitution of an autonomousunit provides the means for regularities to appearwhich are the bases of composionality. This can

manifest at the cellular level as with the celebratedgenetic code for protein sysnthesis, or at the brainlevel with compositional properties of neural ensem-bles. There is nothing mysterious in the emergenceof such composable regularities. Thus contrary tomost philosophical debate today (be this Searle,Harnad, or Dennett) we do not need to have an arbi-trary observer-dependent assiginment of either sig-nificance or compositionality. The key is in the iden-tity properties generated by the self-constitution ofthe organism.

1.4 Organisms doubledialectics

Organism, then, is a key center for cognitive science,and it cannot be broached as a single process. Weare forced to discover regions that interweave in

13


11/11


complex manners, and, in the case of humans, thatextend beyond the strict confines of the body intothe socio-linguistic register.

Further, what I have argued is that behind thismeshwork of the various selves we carry around, isthat all of these selves share a common and funda-mental logic while differing in their specificity. Thisis a case of what Wittgenstein would have calledfamily resemblances: rather than any character-istic being common to all instances, we deal with acluster of overlapping characteristics. We may alsospeak of this cluster of common characteristics asa shared dialectic, since we are dealing here withdouble-sided process, where co-definition is at thecore of the matter. In fact, I submit that the organ-ismic dialectic of self is a two-tiered affair: We haveon the one hand the dialectics of identity of self;on the other hand the dialectics through which thisidentity, once established, brings forth a world froman environment. Identity and knowledge stand in re-lation to each other as two sides of a single process:that forms the core of the dialectics of all selves.

First, a dialectics of identity establishes an au-tonomous agent, a for-itself (pour soi). This iden-tity is established through a bootstrapping of twoterms:

(i) adynamicalterm which refers to an assembly ofcomponents in network interactions and whichare capable of emergent properties: metabolicnets, neural assemblies, clonal antibody net-works, linguistic recursivity;

(ii) a globalterm which refers to emerging proper-ties, a totality which conditions (downwardly)the network components: cellular membranes,

sensory-motor body in space, self/non-self dis-crimination, personal I.

These two terms are truly in a relation of co-definition. On the one hand the global level can-not exist without the network level since it comesforth through it. On the other hand the dynamicallevel cannot not exist and operate as such withoutit being contained and lodged into an encompassingunity which makes it possible.

Second, a dialectics of knowledge establishes aworld of cognitive significanceforthis identity. Thiscan only arise from the perspective provided by thisidentity, which adds a surplus of significance to theinteractions of the environment proper to the con-stituting parts.

The key point, then, is that the organism bringsforth and specifies its own domain of problemsand actions to be solved; this cognitive domain

does not exist out there in an environment thatacts as a landing pad for an organism that some-how drops or is parachuted into the world. In-stead, living beings and their worlds of meaningstand in relation to each other through mutual spec-ification or co-determination. Thus what we de-scribe as significant environmental regularities arenot external features that have been internalized, asthe dominant representationalist tradition in cog-nitive scienceand adaptationism in evolutionarybiologyassumes. Environmental regularities arethe result of a conjoint history, a congruence whichunfolds from a long history of co-determination. InLewontins (1983) words, the organism is both thesubject and the object of evolution.

This second tier of the organisms dialectics, then,is also established through the bootstrapping of twoterms:

(i) a significance term which refers to the neces-sary emergence of a surplus meaning properto the perspective of the constituted self: cel-lular semantics, behavioral perception and ac-

tion, self/non-self as somatic assertion, personalidentity,

(ii) a coupling term which refers to the neces-sary and permanent embeddedness and depen-dency of the self on its environment, since onlythrough such coupling can its world be broughtforth: physico-chemical laws for the cellularworld, macroscopic physical properties for cog-nitive behavior, molecular interaction for im-mune self, socio-linguistic exchanges for oursubjective selves.

Double dialectics: the nature of an identity andthe nature of a relation to a world. Double para-doxicality: Self-production by dependent contain-ment; autonomy of knowledge through environmen-tal coupling. Both dialectics give rise to the shiftingnature of organism, ineluctably forming itself andin-forming where it is, and equally ineluctably im-plicated in the background from whence it springsforth. Organisms, those fascinating meshworks ofselfless selves, no more nor less than open-ended,multi-level circular existences, always driven by thelack of significance they engender by asserting theirpresence.

Acknowledgments

The financial support of CNRS, Fondation de France(Chaire Scientifique) and the Prince Trust Fund isgratefully acknowledged.

14

essay 1 autopoeisis

Documents