ABSTRACT Reasoning about Rationality and Beliefs

In order to succeed, agents playing games must reason about the mechanics of the game, the strategies of other agents, other agents ’ reasoning about their strategies, and the rationality of agents. This paper presents a compact, natural and highly expressive language for reasoning about the beliefs and rationality of agents ’ decision-making processes in games. It extends a previous version of the language in a number of important ways. Agents can reason directly about the rationality of other agents; agents ’ beliefs are allowed to conflict with one another, including situations in which these beliefs form a cyclic structure; agents ’ play can deviate from the normative game theoretic solution. The paper formalizes the equilibria that holds with respect to agents ’ models and behavior, and provides algorithms for computing it. It also shows that the language is strictly more expressive than that of Bayesian games