Persuasion Based on Exchanging Proposals for Cooperative Scheduling Takayuki Ito and Toramatsu Shintani Department of Intelligence and Computer Science, Nagoya Institute of Technology, Nagoya, Japan 466-8555 SUMMARY Software agents, which can act autonomously and cooperatively in network environments on behalf of their users, are being actively investigated. Reaching a consensus is one of the most important activities of agents. The coop- eration of the agents in creating a global schedule (a sched- ule being a vector of events), we call cooperative scheduling. In this article, we formalize cooperative scheduling as a bargaining problem in game theory, then we propose a persuasion method for solving the bargaining problem among agents. We propose a persuasion method based on exchanging proposals between agents. Namely, each agent receives the others proposal in compensation for the others acceptance of his own proposal. In our experiments, we use agenda scheduling as an example of cooperative scheduling and demonstrate the efficiency of the persuasion method proposed here. ©1999 Scripta Tech- nica, Syst Comp Jpn, 30(7): 18, 1999 Key words: Multi-agent systems; persuasion; reaching a consensus; game theory; negotiation. 1. Introduction In recent years, software agents [2, 7] have been investigated very actively. These can act autonomously and cooperatively in network environments on behalf of their users. A multiagent system is a system in which multiple agents autonomously and cooperatively solve local and global problems. Reaching a consensus [10, 11, 17] is a central issue for multiagent systems. In this article, we assume a situation in which each agent has a local schedule and tries to reach a consensus in order to create a global schedule. The cooperation of the agents in creating a global schedule (a schedule being a vector of events) we call cooperative scheduling. For example, when scheduling agendas for a meeting, each agenda is an event. Each agent has individual preference for a schedule. We call individual preferences local prefer- ences. Local preferences should be reflected on a global schedule. Here, we formalize a method for cooperative scheduling. We then propose a persuasion method for cre- ating a global schedule in which local preferences are reflected. In real society, there exist a variety of methods for persuasion [8]. We have modeled some persuasion methods and applied them to negotiation among software agents. In a process of persuasion, an agent who persuades another agent is called a persuader and an agent who is persuaded by a persuader is called a compromiser. The outline of the persuasion protocol can be shown as follows. (Request) the persuader sends a proposal to the compromiser in order to reach a consensus. (Compromise) the compromiser re- ceives the proposal. If the compromiser is able to accept the proposal, he (or she) need not revise his (or her) belief. If he is unable to accept the proposal, the compromiser tries to revise his belief (needs, decisions, or preference) in order to accept the proposal. (Reply) as the result of the belief revision, if the compromiser is able to accept the proposal, he replies in the form of an agreement message. If not, the compromiser replies by sending a reject message. CCC0882-1666/99/070001-08 © 1999 Scripta Technica Systems and Computers in Japan, Vol. 30, No. 7, 1999 Translated from Denshi Joho Tsushin Gakkai Ronbunshi, Vol. J81-D-I, No. 9, September 1998, pp. 10991106 1