Agents, a broker, and lies

摘要

Virtual enterprises need reliable and efficient cooperation mechanisms to carry out transactions between autonomous agents with conflicting interests. Available cooperation mechanisms either use bilateral multi-step negotiation or auctioning. Negotiations encourage agents to reason about the interests of their opponents. Thus, negotiations suffer from counter speculations. Auctions apply to asymmetric trading only; they either favor the auctioneer or the bidders. Both mechanisms do not promote agents to tell the truth. Therefore, we propose to use a trustbroker to mediate between the agents. We introduce three symmetric, negotiation free one-step protocols to carry out a sequence of decisions for agents with possibly conflicting interests. The protocols achieve substantially better overall benefit than random or hostile selection, and they avoid lies. We analyze the protocols with respect to informed vs. uninformed lies, and with respect to beneficial vs. malevolent lies, and show that agents are best off to know and announce their true interests. Analytical proofs and simulations substantiate our results.