Collective Cognition

1
Collective CognitionMathematical Foundations
ofDistributed IntelligenceSanta Fe Institute,
22-26 January 2002

• Welcome!
• Focus Adaptation and Learning in Collectives
• Goals
• Open problems definitional, methodological?
• Promising application areas?
• Canonical problems tractable, but capture
  essentials
• What fields do we need to learn?
• Atmosphere Rethink your assumptions
• Schedule
• Wednesday evening Dinner at El Farol
• Thursday afternoon Self-organizing

2
Collective CognitionMathematical Foundations
ofDistributed IntelligenceSanta Fe Institute,
22-26 January 2002

• Workshop Sponsors
• Intel Corp., Research Division
• Cisco Systems
• DARPA Agent-Based Computing Program, TASK
• SFI
• Robustness Program Packard Foundation
• Evolutionary Dynamics Program Keck Foundation
• Core Funds
• NSF
• MacArthur Foundation
• Business Network
• Thanks to SFI Staff Andi, Sarah, ...

3
Foundations and Open Questions

• What would a foundational approach look like?
• Dynamics of Learning
• Emergence of Distributed Adaptation
• Strategies
• Prototype Models Simple enough to analyze, but
  nontrivial
• Simulation Phenomenology of learning dynamics,
  collective behavior
• Theory Basic constraints/observables,
  quantitative predictions
• Experiment Natural and artificial systems

4
Foundations and Open Questions

• Methods add your favorites here ...
• Dynamical Systems and Pattern Formation Theory
• Statistical Mechanics and Thermodynamics
• Machine Learning and Modern Statistical
  Inference
• Information and Computation Theories
• Game Theory (Economics , Population Dynamics
  (Ecology, Evolution)

5
What is an Intelligent Agent? Some Open Questions

• Learning Agents
• Learning a dynamical system of successive
  predictive models
• States Current knowledge of world and self
• Dynamic Sensory input internal computation ?
  update this knowledge
• Valuation Minimize prediction error, resource
  use, ...

6
What is an Intelligent Agent? Some Open Questions

• What state-space structures required for
  learning?
• Attractors Memories
• Basins of attraction Robust memories
• Internal models Attractor-basin portrait in
  subspace
• Adaptation Bifurcations
• Define and measure agent cognitive abilities
• Develop mathly analyzable, simulatable models

7
What is a Collective? Some Open Questions

• Learning-Agent Collectives
• Group versus individual function
• Define and measure degree of cooperation
• Population size trade-off One smart agent v.
  many less capable?
• Group structure
• Topology, yes, but what to measure (number,
  semi-group, )
• Collective function versus communication topology

8
What is a Collective? Some Open Questions

• Groups necessarily must be better (?)
• Empirical fact Sub-emergent groups
  (submergence)
• Group less functional than the least competent
  member
• Forward v. Inverse v. Inference Problems

9
Obstacles

• What are we up against?
• High-dimensional systems
• Complicated dynamics
• Nonstationary and large-deviation statistics
• Patterns, Organization, Structure?
• Their spontaneous emergence?
• Individual goals and interaction
• Group
• Coordination and Cooperation Functionality
• How to define and quantify these?
• What observables are relevant, usable, and
  general?

10
Applications

• Applications
• Design of distributed and agent-based software
  systems, control systems, and robotics.
• Design of collaboration software and
  groupware'' cf. SlashDot and related web sites.
• Design (or redesign) of institutions and
  organizations to better support collective
  cognition. (Cf. knowledge management'')
• Improved understanding of the behavior of social
  animals.
• Improved understanding of human society, culture,
  and science.
• ...

11
Making and Grasping

• Engineering versus Science
• A necessary tension, due to different goals
• Engineering Design, Build
• Science Understand, Analyze
• Collective Cognition
• Particularly susceptible to confusing these two?
• AI (failures and fallacies) all over again?
• Drew McDermott, Artificial Intelligence meets
  Natural Stupidity (197X).

12

• Optimists
• Only recently, we believe, have the tools
  fallen into place to initiate a rigorous,
  quantitative science of collective cognition.
  Moreover, it appears that soon there will be a
  real practical need for such a science.
• Hence, the workshop.

13

• End of the Week Discussion
• Did we reach our goals?
• New, more appropriate goals developed?
• Plans for the Future
• Proceedings
• Website
• Working groups
• Workshops