Axtell - ONR-Atlanta presentation - GTRI Website

Document Sample
Axtell - ONR-Atlanta presentation - GTRI Website Powered By Docstoc
					 Complex Adaptive Systems,
Computational Social Science
     Irregular Warfare
                     Rob Axtell
   Chair, Dep’t. of Computational Social Science
            George Mason University
      External Professor, Santa Fe Institute
Complex Adaptive Systems (5 minutes)

Computational Social Science (5 minutes)

Computational Public Policy (5 minutes)

Regular Warfare as Simple (5 minutes)

Irregular Warfare as Complex (5 minutes)

Modeling Irregular Warfare (5 minutes)
Complex Systems Thinking

 Nonlinear dynamical systems (math):

   Chaos (high irregularity from low-
   dimensional, deterministic system)

 Distributed, decentralized systems (CS):

   Complexity (simple regularity from
   high-dimensional, stochastic system)
Early Ideas of Complexity
        (Santa Fe Institute style)

  Physicists + computer scientists + economists +
  biologists (evolutionary, ecological)

  Artificial life (von Neumann -> Burks -> Langton)

  Self-organization, emergence, spontaneous order
  (including self-organized critical systems)

  Evolutionary algorithms, genetic programming

  Artificial markets, artificial societies
Social Complexity: Micro

     Heterogeneous agents

     Bounded rationality

     Interaction through networks

     Action away from equilibrium

     Realized computationally
Social Complexity: Macro

Social systems are multi-level systems

Agent modeling is a macro-scope (C Langton)

Can’t infer properties of the macro from the micro
specification: fallacy of composition

Can’t infer properties of the micro (agents) from what
emerges at the macro-level: fallacy of division

Social sciences are the hard sciences (H Simon)
 Simple ➜ Complex
 (20th C) (21st C)
Single decision-maker      Multiple agents
Scalar value function,     Heterogeneous utilities,
first-order conditions,    purposive behavior,
numerical solutions        agent models+simulation
Decision theory            Game theory
Mean field, averages       Networks, extremes
Equilibrium, fixed point   Volatility, adaptation, co-
theorems                   evolution
Continuous, smooth         Discrete mathematics,
mathematics                computation
Command-and-control        Bottom-up emergence
    Agent Computing

Population of adaptive software objects...

...That interact socially

...Following (simple) rules of behavior

...Macrostructure EMERGES from the interactions of
the agents
        OR ⊂ MAS
Operations Research
  Single formal representation
  Extremize scalar value function
  Yields normative prescriptions (policies)

Agent modeling/Multi-agent Systems
  Each agent has its own internal model...
  ...and acts to improve its value function
  Key: what emerges at the social level?
  Both positive and normative aspects
   Agents in Practice
Abstract models: Sugarscape

Empirical models: ‘Artificial Anasazi’

Policy-relevant models: 3 1/2 successes

 Abstract model for developing intuitions

Simple agents who
forage for resources
on a landscape

Emergent order
         Artificial Anasazi
   Empirical model for generating explanations

Simple maize-growers
who occupied the
Colorado Plateau for
1000 years

Disappeared c 1300
       Policy model for making predictions

Pre 9-11: Mathematical
Post 9-11: Computational
  Smallpox (White House)
  Flu (HHS)

Policy-makers have driven
the adoption of
computational tools
NASDAQ used to trade
in 1/8s and 1/16s

SEC req’d

NASDAQ management
commissioned an agent
simulation in ADVANCE
of the rule change

Predicted 6 of 8
statistical changes
Regular Warfare: Simple

Lanchester equations
model force-on-force

Can be calibrated for
hand-to-hand combat,
armored combat (e.g.,
tank v tank), etc.

Warfare has evolved!
   Regular vs Irregular
Regular warfare is like a
parade: choreographed,
scripted, orchestrated

Irregular warfare is like
city life outside the
parade: distributed, wild

Need to understand the
society in which irregular
war happens
  Irregular Warfare is a
Complex Adaptive System
 Agent-based combat an
 active area of research
 Agent models are new
 way to do social science
 They are capable of
 representing ALL the
 main ideas of complex
 adaptive systems
 But we are very far away
 from complete models in
 social science
         Agent Models for
Origin and growth of
the Taliban
IED construction,
placement, sensing

Most GIS-savvy
        What is Needed?
Better behavioral,
cognitive, neural models:
experimental economics,
cog sci, neuroscience
Better social process
models: tribal dynamics,
identity choice, growth
More HSCB research
Coherent 10 year
research effort req’d.
 Digression on Research
  Funding in Economics
  Math/Physics: $1.5B
  Ocean + Polar: $1.0B
  Social, Behavioral and
  Economic Sciences: $250M
  Social+Economic Sci:
  Economics: $30M
  Macro+Finance: $5-6M
Fed: $50-100M                Cost of the Financial
Treasury + SEC: $0             Crisis: $2-20T
What is Not Yet Possible
 in Irregular Warfare
High-fidelity models
Prediction of point
outcomes, e.g.,
specific event @ (x, t)
Small N forecasting
comparison of tactics
Long run
How to Tell the Possible
 from the Impossible
Project team is majority
physicists, computer scientists
and engineers
Discussion focuses on IT
Behavior, is ‘representative,’
not variegated (e.g., SD)
Output is mostly ‘eye candy’
No matter what you ask for
they can model it (i.e., no
sense of hard vs easy tasks)
Unwillingness to ‘open source’
their code
Agent models, grounded behaviorally, are the way
social science will be done, eventually
They are ‘normal science’: abstract, empirical and
predictive/policy versions all possible
Problem is not that the methodology is limiting--
there is nothing it can’t do!
Bottleneck is cognitive/behavioral foundations
Snake oil salesmen are out there!
A large, many-year research effort is needed