Chronological ignorance: Experiments in nonmonotonic temporal reasoning (Q1114431)

The article is concerned with some of the central issues of the theoretical research in the artificial intelligence. It aims to show that mechanical reasoning about rich environments can be done in a manner that is both efficient and rigorous. The author uses formal tools he had introduced in his previous articles: An interval-based temporal logic (the propositional case) and an approach to nonmonotonic logic based on a concept of preferred models (a preference logic). Two nonmonotonic logics are introduced: a logic of chronological minimization and a logic of chronological ignorance (CI). The more general of them, the CI, is based on a monotonic logic of temporal knowledge TK, which is an augmentation of the temporal logic by modal operator \(\square\). The step from TK to CI is via a preference (partial order) relation; if a Kripke interpretation M2 is prefered to M1, it is called chronologically more ignorant. Chronologically maximally ignorant (cmi) models are defined. A entails B in CI iff B is true in all cmi models of A. The ``maximal ignorance of some past sentences'' offers a solution of the qualification problem and of the extended prediction problem. (``These problems subsume, and are better defined than, the infamous frame problem''.) Causal theories (in CI) are investigated. There are two types of axioms in causal theories, the boundary conditions and the causal rules. A generalization of causal theories, called inertial theories, is introduced. It yields a formalization of an intuitive concept of potential histories. The both classes of theories have in a sense unique cmi models. These models can be computed efficiently.