Artificial Intelligence for the Game of Frank’s Zoo

In this report we discuss the implementation of artificial intelligence to play the modern card game ‘Frank’s Zoo.’ Frank’s Zoo is a deterministic game with a high amount of imperfect information. In the present research we compare a heuristic AI, based on the authors’ knowledge of the game, to several implementations of MonteCarlo based AIs, and to randomly playing AIs. We used two versions of the game, namely the published version, and a much simplified version, which is less interesting to play but easier to analyse. From the results of our experiments we conclude that the heuristic AI is by far superior to the Monte Carlo AI, while the Monte Carlo AI is superior to the random AI. Furthermore, we conclude that the performance of the heuristic AI is positively influenced by an increasing amount of incomplete information. We also developed and tested a move predictor. From the results of the experiment performed with the move predictor, we conclude that it performs surprisingly well, and that it may be a powerful tool in creating opponent models. Artificial Intelligence for the Game of Frank’s Zoo Loes Braun MICC, Maastricht University Pieter Spronck TiCC, Tilburg University