RI: Medium: Collaborative Research: Game Theory Pragmatics

Stanford University, Stanford CA

Investigators

Abstract

The past decade has seen unprecedented interaction between artificial intelligence and game theory, with exciting intellectual problems, technical results, and potentially important applications. However, this thriving interaction has thus far not challenged in a fundamental way some of the basic assumptions of game theory, to include various forms of equilibrium as the fundamental strategic concept. Equilibrium specifies conditions under which the strategic choices of agents are in some sense stable. Equilibria are clever and beautiful constructs, but they embody strong idealized assumptions and as a result their applicability to complex, realistic games (i.e., formalizable 'social' interactions) is limited. Arguably computer science can provide alternative modeling foundations, or at least significantly contribute to them. This project explores several complementary directions, to include: alternatives to equilibrium as game solution criteria; replacing analysis of large, complex games with analysis of their abbreviated or approximate versions; using machine-learning techniques to model the extent to which agent behavior is strategic, adaptive, or otherwise intelligent; investigating the role of strategic reasoning in controlled but rich environments, such as Computational Billiards, which involves continuous state and actions spaces as well as control uncertainty. One of the outreach and educational components of this project is organizing and participating in an annual Computational Billiards competition. Applications range from electronic commerce to social networks to peer-to-peer systems to online games, and in general all settings in which individual interests intertwine with computational elements.