AI Seminar

A Criteria-Driven Framework for Agent-Based Modeling

Michael Wellman

Professor, Department of EECS
University of Michigan
Tuesday, December 01, 2009
4:00pm - 5:30pm
Stained-Glass Conference Room (3725 Beyster Bldg.)

Add to Google Calendar

About the Event

Agent-based modeling (ABM) is a simulation-based approach to studying complex social systems, by tracing the interaction of decision-making agents represented as computer programs. The approach is central to the experimental investigation of multiagent systems in artificial intelligence, and enjoys an expanding following in the social sciences. Although there have been notable successes, in many cases the influence of ABM studies has been limited by unresolved concern about the sensitivity of results on seemingly arbitrary modeling choices, particularly assumptions about agent behavior. One potential remedy is to avoid direct specification of behavior, instead postulating general evaluation criteria for behaviors, and formulating behavior determination as a search problem. Evaluation criteria might include strategic stability (evolutionary or game-theoretic), predictive validity, cognitive realism, satisfaction of computational constraints, or some combination of these. Techniques developed in my research group for empirical game-theoretic analysis (EGTA) can be readily adapted to serve this goal for ABM. I present several examples from this EGTA work that can be interpreted in ABM terms, including applications in finance, privacy and security, and search advertising.

Additional Information

Sponsor(s): Toyota

Open to: Public