Intuitive Searching: An Approach to Search the Decision Policy of a Blackjack Agent

Zhenyu Pan, Jie Xue, Tingjian Ge

Abstract: Pan, Zhenyu Xue, Jie Ge, TingjianPorker is a game of imperfect information (Rubin and Waston, Artif Intell, 175(5-6), [1]), in which each player has cards hidden from other players. In previous research (Billings et al, in American Association for Artificial Intelligence Press, [2]; Popyack, in Blackjack-playing agents in an advanced AI course, [3]), Markov model serves as a powerful framework to design a Poker agent. In this paper, we propose a new approach named intuitive searching to build the Exploitative Agent AI against the Markov model Agent in the Blackjack Game. The correctness of our approach can be proved by analyzing the predictive distribution of winning rate. Our AI prevails Markov Agent in a 10000-round experiment. The performance of intuitive policy searching approach in general cases and corresponding limitations are also discussed.