Agent Modeling Methods Using Limited Rationality

Book One

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 12

Student Abstracts

To decide what to do in a multiagent world, an agent should model what others might simultaneously be deciding to do, but that in turn requires modeling what those others might think that others are deciding to do, and so on. The Recursive Modeling Method (RMM) [I] provides representations and algorithms for developing these nested models of beliefs and using them to make rational choices of action. However, because these nested models can involve many branches and recurse deeply, making decisions in time-constrained multiagent worlds requires methods for inexpensive approximation and for metareasoning to balance decision quality with decisionmaking cost.