The detection and correction of errors in problem-solving systems

Abstract

This dissertation is an investigation into the world of error detection and correction. Methods of detecting and correcting errors in the knowledge bases of problem-solvers are investigated. Errors are classified as mistakes and blunders. Mistakes are errors which can be corrected; while blunders are errors which cannot be corrected. Two systems are discussed which are designed to detect and correct errors in their knowledge bases and input. The first system, CUMBA, implemented in LISP, was a robotic problem-solver which detected and corrected errors in its data base and also in the problems which it was given to solve. The other system, CUMBAR, was designed to detect and correct errors in sets of instructions in a top-down manner. It analyzed its input, called RECIPES, for syntax and semantic errors, corrected them, when it could, and produced a corrected RECIPE for the user. The two systems discussed in this work demonstrate the feasibility of incorporating error correction techniques in problem-solving systems. The necessity of including these techniques in expert systems is discussed in light of the critical nature of the domains in which expert systems are employed. The underlying thesis of the work is that a rich knowledge base is required to detect and correct errors in knowledge bases. As in hardware systems, redundancy of information is required in software systems for error detection and correction.