Towards Efficient Negotiation Mechanisms for Collaboration

Book One

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 17

Student Abstracts

This paper addresses the problem of resolving disagreements among groups of self-interested, rational agents that are engaged in a collaborative activity. It argues that current AI approaches to negotiation do not adequately accommodate important features of collaborative planning. For collaborative groups, efficiency of outcomes and the consideration of costs associated with calculating preferences in disagreements are more important than the common game-theoretic design goal of strategy stability. A new mechanism, Blind Mediation, that makes the tradeoff between efficiency and the cost of calculating preferences explicit and that accommodates privacy concerns is presented. By experiment, Blind Mediation is shown to perform significantly better than full revelation of preferences and negotiations among people.