Adaptive opponent modelling for the iterated prisoner´s dilemma

This paper describes the design of Laran, an intelligent player for the iterated prisoner´s dilemma. Laran is based on an evolutionary algorithm, but instead of using evolution as a mean to define a suitable strategy, it uses evolution to model the behavior of its adversary. In some sense, it understands its opponent, and then exploits such knowledge to devise the best possible conduct. The internal model of the opponent is continuously adapted during the game to match the actual outcome of the game, taking into consideration all played actions. Whether the model is correct, Laran is likely to gain constant advantages and eventually win. A prototype of the proposed approach was matched against twenty players implementing state-of-the art strategies. Results clearly demonstrated the claims.