Digital games are various games designed and developed with digital technology and implemented on digital equipment. With the active development of the modern game industry, related technologies such as real-time graphics rendering, realistic interaction, and artificial intelligence of games are also constantly improving. Among them, the artificial intelligence technology of games is limited by the development of theoretical artificial intelligence and the calculation time of real-time systems, and the development lags behind graphics and interactive technology. In order to solve these problems, this paper proposes the application of artificial intelligence in digital games based on mathematical statistics methods, aimed at studying and improving the intelligence level of digital games. The method of this paper is to study the mathematical statistics method and the basic principle of Sarsa learning algorithm and then propose the technology of artificial intelligence applied to digital games. The role of these methods is to study different types of mathematical statistics, to study the behavioral tree of artificial intelligence in digital games and the development prospects of the game, and to update the value of artificial intelligence in each iteration according to the Sarsa algorithm formula. This paper proposes a decision-making system to improve the intelligence level of digital games by analyzing the research status of digital games and the mathematical statistics of basketball digital games. In the basketball game experiment, the difference between the operation data of the decision-making system proposed in this paper and the operation data of human players in five game items reaches 0.51, 0.83, 0.58, 0.49, and 0.78, respectively.