An argumentative formalism for implementing rational agents

Abstract

The design of intelligent agents is a key issue for many applications. Although there is no universally accepted de nition of intelligence, a notion of rational agency has been proposed as an alternative for the characterization of intelligent agency. Modeling the epistemic state of a rational agent is one of the most di cult tasks to be addressed in the design process, and its complexity is directly related to the formalism used for representing the knowledge of the agent. This paper presents the main features of Observation-based Defeasible Logic Programming (ODeLP), a formalism tailored for agents that perform defeasible reasoning in dynamic domains. Most agents must have a timely interaction with their environment. Since the cognitive process of rational agents is complex and computationally expensive, this interaction is particularly hard to achieve. To solve this issue, we propose an optimization of the inference process in ODeLP based on the use of precompiled knowledge. This optimization can be e ciently implemented using concepts from pattern matching algorithms.