Role of structured induction in expert systems
Alen David Shapiro · OpenGrey (Institut de l'Information Scientifique et Technique) · 1984
A "structured induction" technique was developed and tested using a rules- from -examples generator together with a chess -specific application package. A drawback of past experience with computer induction, reviewed in this thesis, has been the generation of machine -oriented rules opaque to the user. By use of the structured approach humanly understandable rules were synthesized from expert supplied examples. These rules correctly performed chess endgame classifications of sufficient complexity to be regarded as difficult by international master standard players. Using the "Interactive ID3" induction tools developed by the author, chess experts, with a little programming support, were able to generate rules which solve problems considered difficult or impossible by conventional programming techniques. Structured induction and associated programming tools were evaluated using the chess endgames Icing and Pawn vs. King (Black -tomove) and King and Pawn vs. King and Rook (White -to -move, White Pawn on a7) as trial problems of measurable complexity. Structured solutions to both trial problems are presented, and implications of this work for the design of expert systems languages are assessed.