Establishing Cooperation in Highly-Connected Networks Using Altruistic Agents

This paper addresses the importance and challenges of establishing cooperation among self-interested agents in multiagent systems (MAS). We study MAS operating on highly-connected random and scale-free (SF) networks. However, we emphasize SF networks as these are prevalent in society and nature. Existing imitation-based approaches for cooperation have been shown to not fare very well in these highly-connected networks. Motivated by studies that show the advantage of altruistic privacy buddies in online social networks to provide better privacy guarantees in highly-connected networks, we present a stochastic influencer altruistic agent (StIAA) mechanism for cooperation. In StIAA, a small proportion of altruistic agents which irrespective of their payoff, always cooperate with their neighbors are introduced into a network of self-interested agents that try to maximize their payoff by imitating the wealthiest agents in their neighborhood. To determine optimality of their action choices, the self-interested agents imitate the cooperative action of their altruistic neighbors (should there be one) with a small exploration probability. We show, both analytically and experimentally, that StIAA leads to significantly higher cooperation in highly-connected networks than the existing imitation-based approaches. We also conduct a comprehensive study on the performance of StIAA and the results indicate that it is both robust and scalable.