About logic-based A.I. systems that must handle incoming symbolic knowledge

Abstract

The focus in this paper is on logic-based Artificial Intelligence (A.I.) systems that must accommodate some incoming symbolic knowledge that is not inconsistent with the initial beliefs but that however requires a form of belief change. First, we investigate situations where the incoming knowledge is both more informative and deductively follows from the preexisting beliefs: the system must get rid of the existing logically subsuming information. Likewise, we consider situations where the new knowledge must replace or amend some previous beliefs. When the A.I. system is equipped with standard-logic inference capabilities, merely adding this incoming knowledge into the system is not appropriate. In the paper, this issue is addressed within a Boolean standard-logic representation of knowledge and reasoning. Especially, we show that a prime implicates representation of beliefs is an appealing specific setting in this respect.