RESOLVING ADVERSARIAL CONFLICTS: AN APPROACH INTEGRATING CASE-BASED AND ANALYTIC METHODS

This thesis addresses the problem of finding compromise solutions to multi-agent conflicts. This is a difficult problem since the compromise choices that a problem solver has for continuum-valued goals are infinite, and the agents need to be persuaded to partially abandon goals during problem solving.

To deal with these difficulties, we propose an integration of: (1) Heuristic Methods. Use of past cases similar to the current problem; (2) Analytic Methods. Application of multi-attribute utility theory to many decision makers. Past problem solving episodes similar to the current one are used to focus on the relevant parts of a problem, form a basis for analogical reasoning, avoid past mistakes, and suggest argumentation strategies. Utility theory is used to identify feasible compromises, evaluate whether a contemplated solution is an improvement on a previously rejected one and provide a computational formalism for persuasive argumentation.

We present the processes mentioned above and the knowledge sources that support them. Our examples are taken from the domain of labor mediation and are implemented in a computer program, called the PERSUADER that functions as a mediator in hypothetical labor negotiations. Using the methods described, it suggests appropriate settlements to the disputants. If a suggested compromise is rejected, the PERSUADER attempts to either modify the settlement or the opposing party's "view" of the settlement using the same methods.