Negotiation in state-oriented domains with incomplete information over goals

State Oriented Domains (SODs) are domains where agents are concerned with moving the world from an initial state into one of a set of target states. Negotiation in this environment was explored by Rosenschein and Zlotkin [9], who provided an analysis of incentive compatible mechanisms over a variety of two-agent, singleencounter types. Their model included the concept of an agent's worth (the agent's benefit from achieving its goal), using it as a baseline for utility calculation of a negotiation's outcome. One scenario left unexamined, however, was the case where agents know one another's worths, but not one another's goals. This situation creates the possibility of agents' lying to one another solely about goals, to influence the outcome of a negotiation. In this paper, we explore this specific case of known worths and unknown goals in two-agent State Oriented Domains, in a variety of encounter types. Through analysis and examples, it is shown that an agent can benefit from declaring less costly goals, but that there are certain limits to the lies an agent can beneficially declare. We also analyze the connection of this work to classic game theory results, including general work on incentive compatible mechanisms and the revelation principle.