A Formal Account of Opportunism in Multi-agent Systems

Abstract

Opportunistic behavior is a selfish behavior that takes advantage of knowledge asymmetry and results in promoting agents' own value but demoting other agents' value. It is commonly existing in business transactions and social interactions, thereby gaining much attention and investigation from social science. In the context of multi-agent systems, it is normal that knowledge is distributed among different agents, which creates the opportunity for agents to perform opportunistic behavior to other agents. Since opportunistic behavior has undesirable results for other agents in the system, the aim of this thesis is to eliminate such a selfish behavior from the system. We first propose a formal account of opportunism based on the situation calculus, capturing the features of opportunism: knowledge asymmetry, intention and value opposition. Because opportunistic behavior has undesirable results for other agents in the system but cannot be observed indirectly, there has to be a monitoring mechanism that can detect the performance of opportunistic behavior. We secondly provides a logical framework to specify monitoring approaches for opportunism. We investigate how to evaluate agents' actions to be opportunistic with respect to different forms of norms when those actions cannot be observed directly, and study how to reduce the monitoring cost for opportunism. In order for monitoring and eliminating mechanisms to be put in place, it is important to know in which context agents will or are likely to perform opportunistic behavior. Therefore, we develop a framework to reason about agents' opportunistic propensity. Opportunistic propensity refers to the potential for an agent to perform opportunistic behavior. We characterize the situation where agents will perform opportunistic behavior and the contexts where opportunism is impossible to occur. Finally, we reach our goal through designing two mechanisms for eliminating opportunism: in the epistemic approach an agent's knowledge gets updated so that the other agent is not able to perform opportunistic behavior, and in the normative approach the system is updated with a norm so that it is not optimal for an agent to perform opportunistic behavior.