a garden of models steps toward growing the topology of the possible in public policy modeling
DESCRIPTION
A Garden of Models Steps Toward Growing the Topology of the Possible In Public Policy Modeling. Carl Tollander [email protected] 4th Lake Arrowhead Conference on Human Complex Systems April 25-29, 2007. Modeling Complexity is a Complex Activity!. What we usually start out doing…. - PowerPoint PPT PresentationTRANSCRIPT
A Garden of ModelsSteps Toward Growing the Topology of the Possible In Public
Policy Modeling
Carl [email protected]
4th Lake Arrowhead Conference on Human Complex SystemsApril 25-29, 2007
4/28/2007
2 of 19
Build a modelComposed of statements about
Changing object classesWith changing relationshipsWhen background setting and geometry is dynamic
What we usually start out doing…
…but over time,much more of our task demands…
Simulate a modelComposed of statements
aboutA population of given objectsWith known relationshipsIn some specified geometry.
Modeling Complexity is a Complex Activity!
4/28/2007
3 of 19
Public Policy Making - Some Observations
• Policies are constraints that mediate the evolution of future system (community) physical and social structure.
• New candidate policies must be situated relative to a mix of other existing and contemplated policies.
• In novel situations where new policies are contemplated, the availability and semantics of requisite data are likely to be in some flux.
• Policy mix is cross-jurisdictional and multi-constituency.
• Policy makers no longer directly control information availability, analyses and tempo.
4/28/2007
4 of 19
Multi-source,continuously
refined, environmental
data
Heterogeneous,
Highly dynamic
ConstraintSets
Policy Mix
Constituency Mix
EmergentCommunityStructure
Some Implementation Challenges• Representation of emergent structure• Composability and reusability• Model Maintenance• Validation, verification, calibration
Messy,Messy,ContingentContingentConstantly Constantly EvolvingEvolving
Messy,Messy,ContingentContingentConstantly Constantly EvolvingEvolving
Adaptive Modeling Problem
4/28/2007
5 of 19
Approaching Adaptive Modeling
Improve ways to relate informal notions about models and structure to a variety of formal representations.
Leverage knowledge about growth and regeneration toolkits from developmental biology, industrial design, CAS practice….
Derek Wise (UCR Math) defines mathematical gadgets as:
• Specifying some stuff, • Equipped with structure,• Satisfying some properties
StuffStuffStructureProperties
Two-stage modeling processAdaptiv
e Structur
eModelin
g
ModelsOf
Agents
(structure agents continuously co-create
model)
(domain agents, familiar ABM
methodology and analysis)
4/28/2007
6 of 19
Toolkits and Emergent Structure
Toolkitsmediate the developmentof future system structure
•Switching•Encapsulating•Promoting,•Inhibiting,•Repressing•Repairing
Genetic toolkits, e.g. HOX(Caporale, Margulis, Carrol)
Artificial Genomes for auto styling (BiosGroup, Plektyx)
RNA shape space (Fontana, et al)
Evolution of banking in Renaissance Florence (Padgett)
QuickTime™ and aTIFF (LZW) decompressor
are needed to see this picture.QuickTime™ and a
TIFF (LZW) decompressorare needed to see this picture.
Usually multiple toolkits, overlapping, multi-purpose…
They emerge, evolve, disappear…
4/28/2007
7 of 19
Creating Topologies of the Possible
Agents carrying policy toolkits co-construct (grow) ensembles of possible model topology Adaptive Structure ModelingAdaptive Structure Modeling Models of
AgentsModels of
Agents
Well-situated “spot” ABMs created from this topology when needed for analysis.
Still messy, contingent, constantly
evolving… A Community Resource:• Self-maintaining, easier validatation,• Increased policy transparency, interoperability, componentry.• Faster, more targeted ABM creation.
But…
4/28/2007
8 of 19
Garden of Models: Research Program in Adaptive Modeling
How can heterogeneous populations of structure-buildingagents jointly and continuously create, regenerateand navigate a common model context?
Jointly grownmodel structure
Jointly grownmodel structure
Models runnable in existing modeling frameworksModels runnable in existing modeling frameworks
AAAA
AA
AA
AA
DD
DD
DD
DD
DD
DD
DynamicHeterogeneous
Structure-buildingPolicyAgents
DynamicHeterogeneous
Data
4/28/2007
9 of 19
Resartus Testbed
Purpose: test computational embodiments of Garden of Models research program in order to drive effort towards a well-engineered Policymaker’s Workbench software architecture and implementation.– Models building models– Policies as structure-building agency– Agents with identity and multiple agency– Heterogeneous agents, heterogeneous
policies
4/28/2007
10 of 19
• Background independence of emergent structure
• Stuff, Structure, Properties Category Theory
• Rich Partial Equivalence, detection and navigation
• Structure Agent Scheduling
• Emergent Structure Model Feeds
• Workbench user interaction mechanisms
• Toolkit packaging and exchange in Workbench
Resartus areas of investigation
4/28/2007
11 of 19
Category Theory - a set of gadgets useful for working with complex structure• Category - objects + morphisms (transformations)
that preserve structure of the objects.• Functor - bundle of transformations between
categories: object to object, morphism to morphism.• N-category - category of categories, internal
morphisms all functors.• Natural Transformations - transformations
between paths (functors of functors) that are equivalent.
• Equivalence - items are equivalent if there is a transformation between them (many available kinds of transformations)
Resartus areas of investigation - Category Theory
4/28/2007
12 of 19
Agents use policy constraints to detect, establish, degrade or navigate rich partial equivalence in a model based on policy constraints.
In practical terms, these constraints take the form of one or more N-categories, called Horizons, which describe the depth and scope of the policy.
Comparing two N-categories for equivalence with respect to a Horizon yields a (possibly empty) functor, which constitutes new agent-navigable structure in the growing model.
Properties of a policy horizon determine the role of the equivalence vis-à-vis toolkits, i.e., promote, inhibit, repress, activate, etc.
Resartus areas of investigation - Rich Partial Equivalence
4/28/2007
13 of 19
Aa
An Agent carries one or more horizons, which are its agencies.
Agency can be delegated, rewarded, recombined, etc.
aa
On opportunity,Agent may selectone or more ofits agencies for the situation at hand.
• Policy agents• Constituency agents
The identity of an agent is the sum of its agencies and any heuristics for their application.
Resartus areas of investigation - Agency and Identity
4/28/2007
14 of 19
Finding Equivalence
Aa
aa
C1
C2
Aa
aa
C1
C2
Aa
aa
C1
C2
Ea
Ea
!
Ea
!
??Aa
aa
C1 C2
Aa
aa
C1 C2
Aa
aa
C1 C2
??
Ea
?
Ea
!
Navigating Equivalence
Resartus areas of investigation - Equivalence Examples
4/28/2007
15 of 19
Since we don’t know what local topologies we will find, we must minimize pre-specification of that topology in the scheduler. A
a
aa
Aa
aa
Randomjumps incategory-memoryspace(a là Tierra,StarCat)
Resartus areas of investigation - Scheduling Structure Agents
4/28/2007
16 of 19
Similar to RSS (syndicated) news feeds on the Web.
AggregatorAggregatorFeedFeedFeedFeed
FeedFeed
FeedFeed
FeedFeedAggregatorAggregatorFeedFeed
AggregatorAggregatorFeedFeedFeedFeed
Difference: • Feeds are/deliver Categories (new local model topologies)• Aggregation is (here) an adaptive modeling process containing structure-building agents.
Resartus areas of investigation - Model Feeds
4/28/2007
17 of 19
QuickTime™ and aTIFF (Uncompressed) decompressor
are needed to see this picture.
• CAP Workbench for cross-jurisdictional policy in rural communities
• Policy Component Webs - distributed policy making.• Multiple language implementation of Resartus
elements.• Learning - Hierarchical Reactive Planning
– Agent learns choice of horizon– Path equivalence / plan equivalence
• Advanced Schedulers (e.g. Cartan-geometry based)• Analytic Journalism - modeling possible story
spaces• Model recombination
Future Directions
4/28/2007
18 of 19
Category TheoryVery fast introduction by John Baez: http://math.ucr.edu/home/baez/planck/node5.html
Notions of Equivalence by Barry Mazur: http://www.math.harvard.edu/~mazur/preprints/when_is_one.pdf
Research Programs and MathematicsCorfield, David, “How Mathemeticians may Fail to be Fully Rational”, 2006
http://www.dcorfield.pwp.blueyonder.co.uk/HowMathematicians.pdf
Wise, Derek, “Properties, Structure and Stuff”, UCR Quantum Gravity Seminar notes, Spring, 2004, http://math.ucr.edu/home/baez/qg-spring2004/s04week01.pdf
Developmental and Molecular Biology, Constructive Social Models (Toolkits, etc.)
Caporale, Helena Lynnn , “Darwin in the Genome: Molecular Strategies in Biological Evolution”, McGraw-Hill, 2003
Carroll, Sean B., “Endless Forms Most Beautiful: The New Science of Evo Devo”, W.W. Norton Company, 2005
Margulis, Lynn and Dorion Sagan, “Acquiring Genomes: A Theory of the Origins of Species”, Basic Books, 2002
Padgett, John, and Paul McLean, “Organizational Invention and Elite Transformation: The Birth of Partnership Systems in Renaissance Florence”, AJS Volume 111 Number 5 (March 2006): pp 1463-1568
B. M. R. Stadler, P. F. Stadler, G. Wagner and W. Fontana “The Topology of the Possible: Formal spaces underlying patterns of evolutionary change”, Journal of Theoretical Biology, 213 (2), 241-274 (2001)
References
4/28/2007
19 of 19
How to transform structural asymmetry to ‘gradients’ (non-commutative flows along equivalence)
Organizations of dimensionality vs. ‘levels’?
Next implementation languages after Java?
Relationships vs Transformations?
More meta the model the lower the required cognition?
Qu
ickTim
e™
an
d a
TIF
F (U
ncom
pre
ssed
) decom
pre
sso
rare
need
ed
to se
e th
is pictu
re.
Some General Questions