Using Gelfond-Przymusinska’s epistemic specifications to justify (some) heuristic methods used in expert systems and intelligent control

Abstract

 We show that epistemic specifications, proposed by M. Gelfond and H. Przymusinska in [4], lead to a natural justification of several heuristic methods used in expert systems and in intelligent control.

This paper is an overview of several results, some of which have already been published before. In contrast to the original publications, that were targeted specifically towards researchers that design and analyze traditional (degree of certainty, MYCIN-type) expert systems and intelligent control systems, we present the corresponding results in such a way that they become understandable (and hopefully, natural) to the logic programming community as well.

The relationship between epistemic specifications and heuristic methods not only makes heuristic methods more reliable, it also adds confidence that the formalism of epistemic specifications, as proposed by M. Gelfond and H. Przymusinska, is an appropriate formalization of knowledge: Indeed, the fact that the language of epistemic specifications, that was originally proposed without any numerical degree of certainty in mind, turns out to be quite consistent with (practically successful) number-valued heuristics, is a strong argument in favor of this language.