Mechanical Discovery of Classes of Problem-Solving Strategies

Abstract
This research assumes that a problem-solving method has an applicabdtty condmon which specifies the properties of "good" problem-dependent parameters for the method Such a condition Is used as the basis of a computer program that mechamcally generates good parameters for the method to use in solving the problem Such problem-dependent parameters for a method constitute a problem-solving strategy. To show the feastbdity of this approach, it is used on two different methods' the GPS method for solving problems and the Nim-hke method for playing games That as, for each is constructed a program that generates good problem-dependent parameters for the input problem (or game) speclficaUon. These programs and their empirical results on a number of different problems and games are described The programs appear to be reasonably efficient and powerful In pamcular, each discovered some nonobvlous strategies Categories and Subject Descriptors 1.2 6 (Artificial Intelligence) Learning--parameter learning 1 2 8 (Artificial Intelligence) Problem Solving, Control Methods and Search--heunsuc methods

This publication has 4 references indexed in Scilit: