Multi-Agent Learning in Non-Cooperative Domains

Motivation Previous work in coordination (Bond & Gasser 1988) on multi-agent systems are specific either to cooperative or non-coperative problem domains. Previous work in learning in multi-agent systems have considered agents operating to solve a cooperative task with explicit information sharing and negotiations (Wei 1993, Tan 1993). In a companion paper (Sen, Sekaran, & Hale 1994) we describe a general purpose system which describes a cooperative domain in which two agents work together on a joint task without explicit sharing of knowledge or information. The focus of this poster is to extend this approach to a non-cooperative domain, where the agents have conflicting goals. The strength of this work lies in the fact that there is no explicit knowledge exchange between the agents and no dependencies on agent relationships.