Enhancing artificial intelligence in games by learning the opponent's playing style

Abstract

As virtual environments are becoming graphically nearly realistic, the need for a satisfying Artificial Intelligence (AI) is perceived as more and more important by game players. In particular, what players have to face nowadays in terms of AI is not far from what was available at the beginning of the video games era. Even nowadays, the AI of almost all games is based on a finite set of actions/reactions whose sequence can be easily predicted by expert players. As a result, the game soon becomes too obvious to still be fun. Instead, machine learning techniques could be employed to classify a player's behavior and consequently adapt the game's AI; the competition against the AI would become more stimulant and the fun of the game would last longer. To this aim, we consider a game where both the player and the AI have a limited information about the current game state and where it is part of the game to guess the information hidden by the opponent. We demonstrate how machine learning techniques could be easily implemented in this context to improve the AI by making it adaptive with respect to the strategy of a specific player. © 2008 Springer Science+Business Media, LLC.