Semantical considerations on some primitives for agent specification

Abstract

Intelligent agents, invented in artificial intelligence (AI), are finding application in a number of traditional areas. Classical AI notions such as knowledge and intentions can serve as natural primitives for the specification of agents. However, in order for them to live up to their promise, these notions must be given rigorous definitions. We propose formal definitions for intentions, knowledge, and know-how in a general model of actions and time. Our definitions are conceptually simple and are designed to be modular, in the sense of being orthogonal to one another. Using these definitions, we are able to prove a success result for agents that is akin to the notion of liveness in traditional computing. Others have been able to prove similar results only with the support of rather strong additional assumptions.