[New England
      Complex Systems Institute]
[Home] [Research] [Education] [Current Section: Activities & Events] [Community] [News] [The Complex World] [About Complex Systems] [About NECSI]
International Conference on Complex Systems (ICCS2006)

Evolutionary Games and Social Networks in Adversary Reasoning

Paul Scerri
Carnegie Mellon University

Katia Sycara
Robotics Institute, Carnegie Mellon University

     Full text: PDF
     Last modified: August 14, 2006

Evolutionary Games and Social Networks in Adversary Reasoning
Katia Sycara and Paul Scerri
Carnegie Mellon University
Pittsburgh, PA

In this paper we are exploring use of evolutionary game theory (EGT) [Weibull, 1995] to model the dynamics of adaptive opponent strategies for large population of players and strategies. In EGT, players choose a strategy, play it against random members of a large group of players, then get a chance to update their strategy. Over many iterations, EGT can determine how strategies in a population change over time and, e.g., whether some strategies will be evolutionary stable. EGT fits well for developing adaptive adversary models that evolve over time since it can model the game dynamics, bounded rationality, locality and scale to large populations, all properties missing in traditional game theory. Previous EGT work has produced interesting, and sometimes, counter-intuitive results[Frey, 2002]. For example [d’Artigues, 2003] shows how terrorism may be a possible, rational response to competition between states.

Whereas in traditional EGT there is no direct information propagation among the players, we are interested in exploring effects of information propagation through social networks in EGT. The social network describes the connections that exist between agents, e.g., because they are relatives, friends or colleagues. Such networks and how information moves across these networks has been extensively studied. For example, Travers and Milgram (1969) showed that social networks have small worlds property where any two people are connected by a surprisingly small sequence of acquaintances. Carley and other have shown the importance of these networks in the performance of large organizations (Lin and Carley, 1995).
The key underlying phenomena that the information diffusion aims to capture is that the experiences of acquaintances can also be leveraged to speed up learning that takes place in the agent society. Movement of information is likely to change the dynamics of strategies over time, but precisely how it changes the dynamics is a research issue. Notice that the effect of the social network is more subtle than just faster information propagation, specifically, information will spread between acquaintances leading to some sub-groups having distinctly different knowledge than other sub-groups.
We will present experimental results from agent-based simulations that show the impact of information diffusion through social networks on the player strategies of an evolutionary game.

Conference Home   |   Conference Topics   |   Application to Attend
Submit Abstract/Paper   |   Accommodation and Travel   |   Information for Participants

Maintained by NECSI Webmaster    Copyright © 2000-2005 New England Complex Systems Institute. All rights reserved.