How to apply deep leaning in NBA games development?

Developer apply "deep learning" technology into NBA 2k game development, which allows computers to play sport in a human way, to evaluate the situation in a continuous chess game, to abandon the suicide of those players, and to estimate that in the next 20 steps, where to gain advantage is more likely, and to accumulate more experience after each game. Thus, it can make its own probability estimation more accurate. After training innumerable top NBA 2k games and playing 5 million games with himself, AI's strength has reached the professional level and can be further improved. To understand this, we first need to know what mechanism AI is in the game we are playing, so that we can know what changes deep neural networks can make in AI. Today we play video games, whether StarCraft 2, War3, Laogong 4, Civilization 5, no matter how strange the game is, its AI uses the same set of behavioral logic - "finite state machine". It sounds very big, but in fact we don't need to go deep into it. Simply speaking, let AI choose a way to deal with certain situations. For example, NBA 2k series and Dota's computer AI will be set to automatically retreat or return to the city when their blood volume is less than a set of skill damage. NBA AI will automatically pass the locker codes just like this nba2kcodes.top to the player with the highest score success rate in the current situation. If all possible situations in a game are a question bank of a test paper, then the finite state machine is a reference answer bank. Once a certain "question type" appears, the corresponding "answer" is selected. Although the actual situation is much more complex, and many layers of state machine loops are used, the essence of this mechanism is that.