On prediction in Theorist (Q685340)

Theorist is a technical term for both a theoretical framework and a running system which is capable of drawing non-monotonic, defeasible conlcusions [\textit{D. Poole}, A logical framework for default reasoning, Artif. Intell. 36, No. 1, 27-47 (1988; Zbl 0647.68094)]. The research field of non-monotonic or default reasoning established itself during the last decade as an important branch in Intellectics, i.e. Cognitive Science and Artificial Intelligence. It aims at adequately modelling commonsense reasoning which, as has been observed, is often based on uncertain knowledge or is unsound and wooly. The basic idea underlying Theorist is to classify knowledge about everyday life using various categories, to perform a theory formation process, and then to apply classical deduction. The methodology provides mechanisms for dealing with both credulous reasoning, which is sufficient for explaining observations, as well as skeptical reasoning required in case of predicting properties. Its corrent implementation, developed by \textit{David Poole} [Artif. Intell. 36, No. 1, 27-47 (1988; Zbl 0647.68094)] and co-workers, uses an algorithm for prediction which is suitable only for a restricted part of Theorist. In more general cases, the system produces incorrect results regarding skeptical conclusions. We present an extension of this algorithm which is shown to be correct within the entire Theorist framework.