Social Norms of Cooperation in Multiagent Systems

Explaining cooperation between individuals is a fundamental scientific endeavour. Similarly, devising mechanisms that encourage autonomous agents to cooperate with each other is a central topic in multiagent systems. Indirect reciprocity (IR) emerged as the most elaborated mechanism of cooperation discovered, involving the moral assessment of actions, the spreading of reputations and agents that care about the social image of others. In this context, the way actions are judged and reputations are attributed depends on the socially adopted norms that define what actions (and under which contexts) are reckoned as Good or Bad. It remains an open question which norms are able to maintain high levels of cooperation in a multiagent system, composed by a finite number of agents that may often behave unpredictably. We employ computational and mathematical methods (inspired in evolutionary game theory) to explore which social norms are able to promote cooperation in a multiagent system. We seek to extend the state of art by introducing new models of IR that consider a finite number of agents that are given the opportunity to randomly explore the strategy space and that may commit errors of different nature.