Incrementally Learned Subjectivist Probabilities in Games

In this paper, we show how our AI opponents learn internal representations of probabilities. We use a Bayesian interpretation of such subjectivist probabilities but do not implement full Bayesian methods of parameter estimation since we wish the AIs to be as human-like as possible. Thus the parameters of the subjectivist probabilities are learned incrementally.

• Type: Conference
• Year: 2005
• Where: CIG
• Authors: Colin Fyfe
• Keywords: AI Opponents, Bayesian Interpretation, CIG 2005, Subjectivist Probabilities