Sandip Sen, University of Tulsa, USA
If participating agents in a multiagent system can be assumed to be cooperative in nature, coordination mechanisms can be used that will realize desirable system performance. Such assumptions, however, are untenable in open systems. Agent designers have to design agents and agent environments with the understanding that participating agents will act to serve their self-interests instead of working towards group goals. We investigate the choice of interaction strategies and environmental characteristics that will make the best self-interested actions to be cooperative in nature. We analyze the inadequacy of traditional deterministic reciprocity mechanisms to promote cooperative behavior with a fair distribution of the workload. A probabilistic reciprocity mechanism is introduced and shown to generate stable and cooperative behavior among a group of self-interested agents. The resultant system exhibits close to optimal throughput with a fair distribution of the workload among the participating agents.