Running Head Learning Nested Agent Models..pdf

Learning Nested Agent Mo dels in an Information Economy Jose M Vidal and Edmund H Durfee Articial Intelligence Lab oratory University of Michigan Beal Avenue Ann Arb or MI fjm vidal durfee gumichedu May Running Head Learning Nested Agent Mo dels Abstract We present our approach to the problem of how an agent within an economic MultiAgent System can determine when it should b ehave strategically ie learn and use mo dels of other agents and when it should act as a simple pricetaker We provide a framework for the incremental implementation of mo deling capabilities in agents and a description of the forms of knowledge required The agents were implemented and dierent p opulations simulated in order to learn more ab out their b ehavior and the merits of using and learning agent mo dels Our results show among other lessons how savvy buyers can avoid b eing cheated by sellers how price volatility can b e used to quantitatively predict the b enets of deep er mo dels and how sp ecic typ es of agent p opulations inuence system b ehavior Intro duction In op en multiagent systems agents can come and go without any central control or guidance and thus how and which agents interact with each other will change dynamically Agents might try to manipulate the interactions to their individ ual b enets at the cost of the global eciency To avoid this the proto cols