Negotiation with incomplete information about worth: strict versus tolerant mechanisms

Abstract

In the domain of distributed AI (DAI), two negotiation mechanisms (one strict and one tolerant), and a negotiation strategy, for the situation where agents' goals are common knowledge, but the worths they attach to those goals are private are presented. All possible prototypical classes of (non-conflict) encounters are analyzed, and it is shown where the strategy and mechanisms result in stable and/or efficient solutions. In most cases, the negotiation strategy is both stable and efficient. These are two situations in which the strategy, in combination with both mechanisms, results in a conflict (a non-efficient result). The reason for this inefficiency may be the failure to include conflict-resolution techniques. It is conjectured that enhancing the mechanisms with such techniques would result in a stable and efficient outcome in those two situations as well.<>