Markov Decision Process-Based Artificial Intelligence With Card-Playing Strategy and Free-Playing Right Exploration for Four-Player Card Game Big2

The popular East Asian card game Big2 involves rules that do not allow players to view each other's hand cards, making artificial intelligence face challenges in performing well in this game. Based on Markov decision processes (MDPs) that can handle partially observable and stochastic information, we design the Big2MDP framework to explore card-playing strategies that minimize losing risks while maximizing scoring opportunities for the Big2 game. According to our review of relevant research, this is the first Big2 artificial intelligence framework with the following features: first, the ability to simultaneously consider scoring and losing points to make the best winning decisions with minimal losing risk, second, the capability to predict multiple opponents' actions to optimize the decision-making, and third, the adaptability to compete for the free-playing right to change card combinations at the essential moment. We implement a system of four-player card game Big2 on the Android platform to validate the feasibility and effectiveness of Big2MDP. Experimental results show that Big2MDP outperforms existing artificial intelligence methods, achieving the highest win rate and the least number of losing points as competing against both computer and human players in Big2 games.