Learning from user experience in games

Several concepts from traditional research on Artificial Intelligence (AI) need to be trained before they can be used. For example, when applied to a computer game, its AI framework has to “learn” how the game should be played. However, such trainings may not be trivial due to the often complex game world environments. This paper presents a novel training approach for game AI frameworks where, instead of manually a priori defining game playing rules at design time, training of the AI system takes place interactively while sample games are being played. Additionally this paper provides an example of modeling the game world so that game objects, such as Non-Playing Characters (NPCs), can be trained interactively. We also give an outlook to our ongoing research on the incorporation of the presented interactive training approach to a real-world game, namely an autonomous controller for the arcade Ms. Pac Man video game using Case-Based Planning.