The use of imbedded patterns and canonical forms in a self-improving problem solver

Abstract

The aim of the work reported here was to develop problem-solving programs which would solve non-trivial problems and make use of their experience. In all of the work published previously there is a clearly discernible trade-off between problem-solving power and generality. This deadlock could be broken by the development of (a) programs which can make use of their experience and generalize it, and (b) powerful symbol manipulation languages which can express and execute highly complex processes without detailed programming.