Cooperation for DAI through Common-Sense Knowledge

The goal of the Antares Project at MCC is to develop methods that enable separately developed knowledge-based systems to cooperate in solving problems beyond the capabilities of any one of the knowledge-based systems. There have been many previous attempts to develop distributed systems of cooperating experts (see [8] for a comprehensive summary of these). For the most part, these attempts have been ad hoc and have met with few successes. The resultant systems have been domain dependent and did not generalize. Further, it has not been sufficient that each knowledge-based system be written in the same language [9] or have access to a global data structure, such as a blackboard [5, 6]. The fundamental hypothesis of our research is that these systems lack the common-sense understanding of the world that is necessary for cooperation and that is possessed by cooperating humans. Our approach for engendering cooperation among distributed knowledgebased systems utilizes 1) a large global base of common-sense knowledge that permits enforcement of semantic consistency and 2) an explicit characterization of self-interest to control global problem solving while allowing autonomy. It has been shown [1, 4] that autonomous systems can cooperate without the explicit exertion of control if they are self-interested. A general notion of self-interest for an autonomous knowledge-based system must include representations for the goals and problem-solving state of this system, as well as models of the goals and problem-solving states of the other systems with which it interacts. However, interaction and communication