Partner Selection Using Reputation Information in n-player Cooperative Games

Abstract

To study cooperation evolution in populations, it is common to use games to model the individuals interactions. When these games are n-player it might be difficult to assign defection responsibility to any particular individual. In this paper the authors present an agent based model where each agent maintains reputation information of other agents. This information is used for partner selection before each game. Any agent collects information from the successive games it plays and updates a private reputation estimate of its candidate partners. This approach is integrated with an approach of variable sized population where agents are born, interact, reproduce and die, thus presenting a possibility of extinction. The results now obtained, for cooperation evolution in a population, show an improvement over previous models where partner selection did not use any reputation information. Populations are able to survive longer by selecting partners taking merely into account an estimate of others’ reputations.