Abstract

Our proposal is to apply a Came Theoretic approach to the games played in Agent Reputation and Trust Final Competitions. Using such testbed, three international competitions were successfully carried out jointly with the last AAMAS international Conferences. The corresponding way to define the winner of such competitions was to run a game with all the participants (16). Our point is that such game does not represent a complete way to determine the best trust/reputation strategy, since it is not proved that such strategy is evolutionarily stable. Specifically we prove that when the strategy of the winner of the two first international competitions (2006 and 2007) becomes dominant, it is defeated by other participant trust strategies. Then we found out (through a repeated game definition) the right equilibrium of trust strategies that is evolutionarily stable. This kind of repeated game has to be taken into account in the evaluation of trust strategies, and this conclusion would improve the way trust strategies have to be compared.