Using Experience-Based Learning in Game Playing

Abstract

This paper describes some initial experimental results in the use and limits of applying experience-based learning techniques to the area of game playing. A system (GINA) has been developed that uses experience-based learning to improve its own base performance level for the game of Othello. A variety of traditional (non-learning) Othello game playing programs have been collected to provide a rich source of “experiences” for GINA. An initial set of experiments has been designed and run in an attempt to understand better the strengths and weaknesses of experience-based learning from the point of view of skill refinement (taking less time to perform at a given level of play) and from the point of view of skill improvement (higher quality play). Preliminary results suggest that in this context experienced-based learning can be a highly effective means for improving both the speed and quality of play without having to accumulate experience bases of intolerable size.