Facebook creates an AI poker program

Facebook developed an Artificial Intelligence (AI) enabled to play poker games of up to six participants, and in several tests has consistently emerged victorious, for the first time in a mechanism of the genre, against professional human players.

This was announced by the social network with regard to the collaboration with Carnegui Mellon, in order to create the first artificial intelligence bot capable of playing poker in six-player games in the Texas Hold’em variety, without betting limits. The bot is called Pluribus and has managed to win against human players of the size of two winners of the World Series of Poker. The discoveries of both creators were published in the journal Science this week.

This is one of the first examples of an AI developed to play poker, a game considered “a great challenge” for AI according to its creators for decades because it involves unknown information and strategies different from other games such as chess or go, like the use of lanterns.

Pluribus incorporates a new online search algorithm that can evaluate players  options efficiently by looking only for some forward movements instead of the final goal of the game. Pluribus also uses new faster auto-game algorithms for games with hidden information.

These innovations have important implications beyond poker, because the zero-sum interactions of two players are common in recreational games, but they are very rare in real life. Real-world scenarios, such as taking action on harmful content and dealing with the challenges of cybersecurity, as well as managing an online auction or navigating traffic, generally involve multiple actors and hidden information.

The results of the study “show that a carefully constructed AI algorithm can achieve a greater than human performance outside of zero-sum games for two players,” according to researcher from Facebook’s AI division Noam Brown.

K. Tovar

Source: FayerWayer