Applause : an implementation of the Collins-Michalski theory of plausible reasoning

The ability to make decisions in situations not encountered before characterizes human reasoning. This paper discusses a pilot implementation of the computational model of human plausible reasoning based on a core theory by Collins and Michalski. The theory assumes that human knowledge can be represented as objects or concepts that are related by similarity, generalization, and specialization relations, and that are arranged into hierarchies. Facts about the world are represented as traces linking nodes of different hierarchies. The building of the hierarchies and construction of the links is an integral part of the learning process undergone by human beings. Plausible reasoning is an ability to draw inferences when direct links between concerned objects are not available. This involves perturbation of established traces, traversal through the concerned hierarchies, inheritance of the properties along the way, and combination of ev idences for selection of the best inference.

The core theory has been operationalized and expanded to use confidence parameters, dynamic learning of dependencies and implications, automatic find ing of context for reasoning, and combination of evidence. A pilot version of the theory of plausible reasoning has been implemented in a system called applause (APproximate/pLAUSiblE reasoning). Some key operations are illustrated with examples, and the plausible reasoning process, including discovery of useful dependencies, is demonstrated on a problem in the domain of the chemical periodic table.