Evolutionary detection of new classes of equilibria: Application in behavioral games

Abstract

Standard game theory relies on the assumption that players are rational decision makers that try to maximize their payoffs. Experiments with human players show that real people rarely follow the predictions of normative theory. Our aim is to model the human behavior accurately. Several classes of equilibria (Nash, Pareto, Nash-Pareto and fuzzy Nash-Pareto) are considered by using appropriate generative relations. Three versions of the centipede game are used to illustrate the different types of equilibrium. Based on a study of how people play the centipede game, an equilibrium configuration that models the human behavior is detected. This configuration is a joint equilibrium obtained as a fuzzy combination of Nash and Pareto equilibria. In this way a connection between normative theory, computational game theory and behavioral games is established. © 2010 Springer-Verlag.