The Role Of Cooperation in Multi-Agent Learning

We are currently investigating the problem of dynamic adaptation in systems consisting of multiple intelligent agents. An essential characteristic of these systems is that the salient knowledge for learning is distributed amongst the agents. As in human communities, effective learning therefore requires cooperation amongst the constituent members. In this paper we detail the role of cooperation in such learning systems and the problems introduced by it. These ideas have been used in MALE (Multi-Agent Learning Environment) from which an illustrative example is shown.