According to Technology Review, the algorithm was so successful that the team decided not to release the code, fearing it could be used to empty the coffers of online poker companies. Cited by the site, Noam Brown, Facebook investigator in the area of AI, guarantees that could be "quite dangerous for the poker community."
For decades, poker has been a huge challenge for AI, which in recent years has only been able to beat the best poker players in a two-player game mode. But now the challenge has been overcome and the details were published this Thursday in a paper.
Pluribus defeated the best professional players in the world's most popular poker game, Texas Hold'em, in No-Limit Hold'em format, which has no limit on betting. That is, the player can bet what he wants, including all his chips at any time. Facebook has, however, shared the interface used during the game of the bot with professional players.
In this game the goal is to assemble the best possible combination of cards or convince opponents that they have the best combination.
Pluribus defeated professional players in two formats. Some games took place between five bots and one human player, while in another only one Pluribus played with five professional players.
What is Pluribus' strategy to beat professional players?
The bot is the result of several innovations of the Libratus, the artificial intelligence robot that in 2017 defeated professional players in the same game, but only in the version of two players.
In particular, Pluribus integrates a new online search algorithm that allows you to effectively evaluate the options on the table, searching only a few moves ahead and not until the end of the game.
As Noam Brown explains in a blog post on Facebook about AI, this bot has been trained to play against itself, not needing to resort to specific techniques of the game.