Problem Solving by Negotiation among Autonomous Knowledge-Processing Systems

Abstract

Distributed expert systems have several advantages compared with ordinary expert systems. Research on such systems belongs to the field of cooperative problem solving, which is a sub-field of distributed artificial intelligence. Cooperative problem solving can be divided into three stages : (1) problem decomposition, (2) subproblem solution, and (3) answer synthesis. Several studies have addressed to the general domain-independent model of the first and second points. However, for the second and third points, most studies have addressed only specific cases of distributed expert systems and have insisted on the generality of their method. Few of these works have presented general domain-independent architectures and their implementation. In order to cope with this problem, we present CoKBS : an architecture for managing several autonomous knowledge-processing systems (called agents here-after). CoKBS has been implemented with the metaphor of negotiation among human experts. In CoKBS, the solution to the problem at work is obtained after negotiation between the involved agents. The agent which is in charge of the problem at work selects two agents and the solution is obtained after a negotiation procedure has been carried out between the two agents. During the negotiation procedure each agent may update dynamically his beliefs, his perspective, his own model, as well as his model of the other agent. The behavior of the agent during the negotiation is determined by his set of meta rules and domain-independent meta rules are supplied by the system. The architecture has been tested on a simple application with an experimental system.