Using logic programs to model an agent's epistemic state

The notion of rational agency was proposed by Russell [9] as an alternative characterization of intelligence agency. Loosely speaking, an agent is said to be rational if it perfomns the right actions according to the information it possesses and the goals it wants to achieve. Unfortunately, the enterprise of constructing a rational agent is a rather complex task. Although in the last few years there has been an intense flowering of interest in the subject, it is still in its early beginnings: several issues remain overlooked or addressed under too unrealistic assumptions. As slated by Pollock. [5], a rational agent should have models of itself and its surroundings, since it must be able to draw conclusions from this knowledge that compose its set of beliefs. Traditional approaches rely on multi-modal logics to represent the agent's epistemic state [7. l]. Given the expressive power of these formalisms, their use yields proper theoretical models. Nevertheless, the advantages of these specifications lend to be lost in the transition towards practical systems: there is a tenuous relation between the implementations based on these logics and their theoretical foundations [8]. Modal logics systems suffer from a number of drawbacks, notably the well-known logical omniscience problem [10]. This problem arises as a by-product of the necessitation rule and the K axiom, present in any normal modal system. Together, these ruIes imply two unrealistic conditions: an agent using this system must know all the valid formulas, and its beliefs should be closed under logical consecuence. These properties are overstrong for a resource-bounded reasoner lo achieve them. Therefore, the traaditional modal logic approach is not suitable for representing practical believers [11]. We intend to use logic programs as an alternative representation for the agent's epistemic state. This formalization avoids the aforementioned problems of modal logics, and admits a seamless transition between theory and practice. In the next section we detail our model and highlight its advantages. Next, sectiol1 3 prescnts sume conclusions and reports on the forthcoming work.Eje: Aspectos teóricos de inteligencia artificialRed de Universidades con Carreras en Informática (RedUNCI