Rule learning approaches for symmetric multiplayer games

This paper describes a series of experiments involving rule learning for Samurai, a symmetric multiplayer strategy game. Rule based artificial intelligence is commonly used in a wide variety of computer strategy games. Automated rule learning processes have been used to derive rules both to improve the quality of AI play and to illuminate possible strategies and tactics for the games. This research explores the effects of using seed AI rules and bootstrap AI for rule learning on the outcome of the learning process. It is demonstrated that good quality sets of rules can be generated for the Samurai game with or without seed rules. It is also shown that the use of a bootstrap AI produces rule sets better tuned to face that AI, although not necessarily better in general.