Abstracts - 2007
Belief Dynamics and Cultural Shifts
This project is a large, 5 yr, multi-university, interdisciplinary effort initiated in 6/05, and funded by AFOSR. The official title is “Computational Models for Belief Revision, Group Decisions, and Cultural Shifts.”
Rather little is known about how individuals gather into small groups to help each other reach a common goal. What is known suggests marked differences across cultures, especially depending upon the nature of the goal. To model these behaviors, we must consider the roles of trust, beliefs, attitudes, and sacred values within a culture, and how they interact with institutional constraints and perceived external pressures. Because these factors may differ considerably across cultures, an individual’s view of what constitutes a rational act can vary markedly, as is highlighted by the current Western vs Middle Eastern clashes. The objective of the MURI is to understand better and to model the structure, evolution and dynamics of small groups engaged in achieving a shared goal.
To reach this objective, we have assembled the following interdisciplinary team:
Scott Atran, CUNY (John Jay), Univ. Mich. & CNRS, Anthropology
Other Collaborators include: Marc Sageman, consultant, Alex Pentland, Media Lab, MIT, Barry Silverman, ESE, Univ. Penn., Rajesh Kasturirangan, Nat. Inst. Adv. Studies, Bangalore.
The spectrum of our expertise is critical to the success of our project. We are using use a multi-prong, interdisciplinary attack. First, the field data available on the effects of cultural change on belief dynamics and behavior are brought to the attention of the collaborating modelers, each expert in quite different types of computational models. These include probabilistic models, Bayes nets, agent-based networks, graphical models, causal and analogical models, and game-theoretic models. This range of models provides us with different representational frameworks appropriate for different levels of analysis, and subsequently, for the integration of models across the individual, group and social levels. In parallel, we are engaging in further field studies to increase our database. Thirdly, we use a VR microworld in the laboratory to probe the more universal factors underlying trust, cooperation, belief revision, and leadership in group decision-making and strategic planning.
Over the past year, we have been developing and applying six main types of computational models to understand belief revision, decision-making, and network structure and evolution. These are:
(i) Infinite Relational Model (Tenenbaum, Kemp) which is aimed at recovering social network structure from data about beliefs in the society (e.g. Medin-Atran Guatemalan studies) and an extension on Learning Relational Concepts.
(ii) Story Structure & Causal Models (Forbus, Winston, Finlayson) whose long-term target is automatic abstraction of themes, role-models, etc. from traditional stories in a culture.
(iii) Consistency-Conformity Model (Page, Bednar) which show conditions for convergence to stability for groups with heterogeneous interests;
(iv) Probabilistic Social Network Model (Richards, Atran, Kasturirangan) which is designed to explore the fragility of a network subject to both internal and external pressures.
(v) Multi-agent Influence Diagrams (Pfeffer, Gal), which is a framework for understanding actions and action sequences in strategic settings.
(vi) Ideal Performance Measures (Stankiewicz), which allows an experimenter to evaluate the optimality of performance in sequential decision-making tasks.
We also have been engaged in collecting data from field studies on behaviors and cultural shifts observed in Guatemalian and Native Indian societies. An equally important focus has been studies of terrorist networks and the belief revisions underlying sacred vs secular beliefs.
Summaries of these results appear in our Annual Progress report available at http://groups.csail.mit.edu/belief-dynamics/.
Our aim is to provide the first integrated computational models of how individual beliefs can influence group decision-making, how groups begin to coalesce into larger entities that lead to cultural shifts, and how these in turn lead to revisions in individual beliefs. Such models will provide predictive tools that can advance adversarial decision-making. For example, they will (1) identify key variables in social networks within different cultures that can be manipulated to revise beliefs; (2) show how social networks may be shattered or made to cohere through cultural shifts or external pressures; (3) show analogies and differences between cultural mind-sets that will provide better perspectives on how societies in different cultures will react to new pressures or to institutional change, and (4) give insights into how trust and belief revision can impact the effectiveness group decision-making.
Supported under AFOSR Contract# FA9550-05-1-0321