Games computers play: simulating characteristic function gameplaying agents with classifier systems

摘要

Many game theorists are turning to evolutionary simulations tomodel the behavior of boundedly rational agents. This new methodologyallows researchers to observe purely adaptive behaviors in games, toobserve differences of behavior due to changes in the games' parameters,to discover equilibria in games that are too complex to calculateanalytically, and to discover new strategies for playing the games. Iextend this methodology to a more complex class of games than hadpreviously been attempted. I create a coevolutionary environment inwhich three agents, represented by classifier systems, play acharacteristic function game. Although the agents have no computationalcapabilities, they learn to adapt reasonably intelligent behavior