Defeasible reasoning in dynamic domains

Abstract

The design of intelligent agents is a key issue for many applications. Since there is no universally accepted definition of intelligence, the notion of rational agency was proposed by Russell as an alternative for the characterization of intelligent agency. A rational agent must have models of itself and its surroundings to use them in its reasoning. To this end, this paper develops a formalism appropriate to represent the knowledge of an agent. Moreover, if dynamic environments are considered, the agent should be able to observe the changes in the world, and integrate them into its existing beliefs. This motivates the incorporation of perception capabilities into our framework. The abilities to perceive and act, essential activities in a practica! agent, demand a timely interaction with the environment. Given that the cognitive process of a rational agent is complex and computationally expensive, this interaction may not be easy to achieve. To solve this problem, we propase inference mechanisms that rely on the use precompiled knowledge to optimize the reasoning process.