Gathering and utilising domain knowledge in commercial computer games

In practice, adaptive game AI in commercial computer games is seldom implemented, because machine learning techniques require numerous trials to learn effective behaviour. To allow fast adaptation in games, in this paper we describe a means of learning that is inspired by the human capability to solve problems by generalising over limited experiences with the problem domain. We compare three approaches, namely straightforward case-based reasoning, situated case-based reasoning, and a k-nearest neighbour classifier. From the experimental results we conclude that the situated approach performs best, both in representing knowledge and generalising to similar situations.