Artificial intelligence Libratus has learned to beat poker against people, a game whose outcome largely depends on the psychological component, not just logic and mathematical calculation. For a long time, machine intelligence learned to beat a person in chess or go, where the moves are calculated ahead using mathematics, probability theory, history of previous games. However, poker cannot be accurately calculated, because a person knows how to cheat, that is, bluff, even having a weak combination of cards on his hands. Therefore, artificial intelligence had problems reaching the final stage of the game, when the psychological component played a more important role than simple mathematical calculation . Several people could bluff, each player has 2 unknown cards in their hands, a huge number of combinations is simply difficult to calculate, and it seemed simply impossible to take into account the bluffing players. However, Libratus developers have found a simple solution. Artificial intelligence was taught not to calculate the situation, but to evaluate the totality of all the data: the amount of bets, the number of people playing and compare with the typical course of gaming events. In addition, the developers divided the game into 3 consecutive stages. At the initial stage of the game, the AI was careful, playing in a quiet mode. At the final stage of the game, the car used an aggressive style, increased rates regardless of the combination of cards “on hand”. The AI constantly studied, corrected errors, and took into account the individual playing style of the enemy. The result of using this algorithm is obvious – first Libratus won $ 2 million from a team of people, and then defeated Baby Tartanian8, the world’s strongest poker program. Thus, AI unraveled the concept of the psychological component of the game, which, in fact, takes the machine to the level that science fiction writers and the famous Ilon Mask are afraid of.