The Utility of Communicati Intelligent A

When intelligent agents who have different knowledge and capabilities must work together, they must communicate the right information to coordinate their actions. Developing techniques for deciding what to communicate, however, is problematic, because it requires an agent to have a model of a message recipient and to infer the impact of a message on the recipient based on that model. We have developed a method by which agents build recursive models of each other, where the models are probabilistic and decision-theoretic. In this paper, we show how an agent can compute the impact of a message in terms of how it increases (or decreases) its expected utility. By treating the imperfect communication channel probabilistically, our method allows agents to account for risk in committing to nonintuitive courses of action, and to compute the utility of acknowledging messages.