Title:  A New Architecture for Negotiating Agents Based on Punishment and Pessimism

Ronald P. Loui
Department of Computer Science and Engineering
Washington University
St. Louis
USA

Abstract

This talk briefly comments on ideas of Jennings-Sierra-Parsons and
Sandholm for an AI theory of negotiation.  Then it describes a new
architecture for an agent:  this agent is driven purely by the expectation
of settlment and a nonstandard utility which rewards the dialogue-act of
unilaterally breaking down on an uncooperative negotiating partner.  The
resulting behavior includes successively improving offers, accepting an
offer that was initially unacceptable, and breaking down as a punishment
on procedural grounds.  We define the laissez-faire path as the path taken
by two such agents with fixed parameters and no initiative, and view the
joint problem as an acceleration of the schedule of expectation-based acts
so as to avoid trajectories that lead to breakdown.  Finally, we discuss
briefly what are the desiderata for an AI theory of negotiation that
aims at competence rather than performance.