Modeling and Simulation for National Security Decision Making or “A Plea for a V&V and UQ Community of Practice” Laura A. McNamara, PhD Timothy G. Trucano, PhD Sandia National Laboratories Albuquerque, NM Sandia Review and Approval 2005-7679 C
Modeling and Simulation for National Security Decision Making
or “A Plea for a V&V and UQ Community of Practice”
Laura A. McNamara, PhDTimothy G. Trucano, PhD
Sandia National LaboratoriesAlbuquerque, NM
Sandia Review and Approval2005-7679 C
Part One
Computational Social Simulation,
or“Scaling to China”
DECISION MAKING; or,
Who cares about Balinese Water Temples?
• Lots of computational social science takes place outside of decision context– Anasazi Longhouse
– Balinese Water Temples (Lansing)
Computational Science
Why Models are Fraught with Conflict
CUD STUDY:Mod-Sim, Forecasting, and Decision Making
Decision Making Under Uncertainty
EconomicForecasts
IntelligenceAnalysis
Weather/Climate
Forecasts
StockpileStewardship
Factors:
• Information
• Analytic processes and tools
• Uncertainty and incomplete knowledge
• Risk and confidence
• Decision impact
• Social issues (roles, bureaucracy/org)
Factors:
• Information
• Analytic processes and tools
• Uncertainty and incomplete knowledge
• Risk and confidence
• Decision impact
• Social issues (roles, bureaucracy/org)
Black ClassifiedSocial science dominated
Physical science dominated
Modeling and Simulation in Intelligence Analysis
A Case Study
PART ONE
Models in Physics
Model vs. Simulation
• MODEL represents one or more concepts and the relationships among them.
• SIMULATION reveals what emerges as the relationship is invoked over time.
As A increases, B decreasesAs A increases, B decreases
TIME
INCOME
From Physics Knowledge to Computational Model
PHYSICS KNOWLEDGE
Mathematical Representation(e.g., PDE, ODE)
REAL WORLD EVENT‘Solutions’
ALGORITHM C++, FORTRAN
MACHINE‘CODE’
‘OUTPUT’Graph,
3D Simulation
CODES IN X DIVISION
A Changing Role for Physics and Engineering Codes
• 1945-1992: Computational models for “numerical experiments”• NW Testing program ends, 1992• Science Based Stockpile Stewardship, 1995 • Accelerated Strategic Computing Initiative
– “Virtual test environment” for nuclear physics and engineering problems
• Computational models for – knowledge integration?– Diagnostics– prediction– certification?
• The deterrent will be effected/affected by decisions we make with these models
From Design and Test toVerify, Validate, and Quantify Uncertainty
• VERIFICATION: is the model solving the equations correctly and accurately?
• VALIDATION: does the model output compare well with experimental data?
• UNCERTAINTY QUANTIFICATION: What are the primary sources of uncertainty; are they reducible; and how do these dictate investment in research priorities?
From Physics Knowledge to Computational Model
PHYSICS KNOWLEDGE
Mathematical Representation(e.g., PDE, ODE)
REAL WORLD EVENT‘Solutions’
ALGORITHM C++, FORTRAN
MACHINE‘CODE’
‘OUTPUT’Graph,
3D Simulation
PART TWO
Complexity and Agent Based Modeling
Complexity, complex systems, complex adaptive systems…
• “Something almost entirely random, with practically no regularities, would have effective complexity near zero. So would something completely regular, such as a but string consisting entirely of zeroes. Effectively complexity can be high only in a region intermediate between total order and complete disorder.” *
• Complicated systems aren’t necessarily complex!• A complex system: nonlinear, feedback, iterative
nondeterministic, resists decomposition, emergent, self organizing….
*From Gell-Mann, Murray. 1995. ‘What is Complexity?” in Complexity. New York: Wiley and Sons.
Consider the Boid
Maintain separation
Maintain alignment
Maintain cohesion
BOIDS Film
PART THREE
Agent Based Modeling
And Anthropology
• PRIMATOLOGY: troop behaviors• ARCHAEOLOGY: hunter-gatherer foraging; settlement
patterns in SW Anasazi society; the emergence of states from chiefdoms or tribal societies
• CULTURAL ANTHROPOLOGY: Marriage and stratification in Tonga; Balinese water temples as an emergent system for efficient agriculture
• MORE GENERAL PROBLEMS: Why cooperation or competition emerge as behavioral preferences in societies; one-on-one or spontaneous violence vs. organized warfare as a function of environmental characteristics, population size, other variables
The Anasazi Longhouse Simulation:Why was the valley abandoned?
• Households harvest, consume their nutritional requirements of grain.
• Households store any remaining grain for later consumption.
• Households evaluate availability of grain for following year, based on current harvest and grain stores.
• If the amount is insufficient, the household moves to a new viable location to farm
Dean, Gumerman, Epstein and Axtell, 1998. ‘Understanding Anasazi culture change through agent-based modeling’. Santa Fe Institute Working Papers ref: 98-10-094.
“The most humanistic of the sciences, the most scientific of the humanities…”
• ETHNOGRAPHY– Verstehen– Experiential– Idiographic– “Thick Description”– Narrative– Qualitative– Postmodern– Holistic– Meaning
• ABM– Logical positivism– Distanced– Nomothetic– Abstraction– Simplification– Rules, equations– Modern – Extractive– Observables
Why Computational Social Modeling is Really Hard (1)
“If X occurs, then Y is likely.”
P(y,x) > .5
Part Four
Agent Based Models in
National Security Decision Making
EPISIMS: Day 2
EPISIMS: Day 24
EPISIMS: Day 34
Part Five
The Pen and the Sword
Anthropology and National Security:An “allergic response?”
• The sordid history of anthropology• Applied vs. Pure Research
– Applied research is always done with an end in mind, not with an open mind
• First Amendment Rights– McCarthyism: anthropologists being watched, monitored
for leftist behavior– Hoover’s FBI: academics in general, and anthropologists
in particular, monitored in their classrooms• Moral epistemology of anthropology: work to preserve
cultural integrity, diversity• Protection of human subjects
Cold War Anthropology:The Time of Troubles
• WWII: ‘National Character’ studies – Ruth Benedict, The Chrysanthemum and the Sword
• 1963-1965: Project Camelot– SORO/American University project– Develop models of insurgency/ counter insurgency– Chilean fiasco– Was the work any ‘good’?
• National Student Mobilization Committee– Photocopies of anthropologists’ notes from meetings
with intelligence, military officials– Ethics investigation– AAA statement of ethics in anthropological research,
1973
Part Six
Drawing it all Together
The Strong and the Weak
• STRONG Artificial Intelligence: “I can create a machine that can reason as a human mind.”
• WEAK Artificial Intelligence: “I can create a machine that can mimic human techniques for aggregating information.”
• STRONG Computational Social Science: “I can create a virtual society, in which interacting agents produce cultural forms that in turn shape their behaviors.”
• WEAK Computational Social Science: “Within a bounded problem space, I can program agents with simple rules and see what emerges; I can use evolutionary algorithms so my agents can adapt to the emergent phenomenon.”
Modeling and Simulation in Applied Physics and Decision Making
• What’s the purpose of the model?• Who owns the model? Who runs it? Who interprets
its results?• How transparent is the model? • How is the model verified? Validated?• How does the model account for uncertainty in
inputs and outputs?• What is the social location of the model in the
knowledge production process? In decision making?• Is there a formal basis for TRUST in a given
model/simulation?
Questions to consider when developing, using, considering a model
• Why was the model developed? What does it claim?• Which theory? What is theory doing (description,
explanation, prediction?)• What aspect of reality is the model targeting?• What does it mean for data to ‘represent’ some
targeted aspect of social/cultural reality?• What theoretical framework underlies the conceptual
model? • Is the conceptual model being driven by the
computational technique being used to develop the model?
• How do we measure the ‘fit’ between the (rule based) computational format and the (narrative) conceptual model?
• What will the model be used for? ‘Pure’ knowledge production or decision making?
[from Turnley 2004][from Turnley 2004]