Artificial Intelligence: Poker-Playing Pluribus Beats Humans In Multiplayer Texas Hold'em

First artificial intelligence was applied to checkers. Then Deep Blue played chess. Now there’s a bot that knows when to hold ‘em and knows when to fold ‘em.

Gaming sites are trying to protect their games from bots but a loss in player confidence could gut the business. As a result, Morgan Stanley Thursday trimmed price targets on The Stars Group (TSG), GVC Holdings PLC (GVC.L) and Playtech PLC (PTEC.L).

Facebook (FB) and Carnegie Mellon University earlier this summer unveiled Pluribus, a poker bot that beat five professional poker players in a six-party Texas Hold’em tournament, the first time an AI beat elite human players in a game with more than two players. The program was created relatively quickly and for much less than the cost of programming AI for chess or checkers.

Pluribus isn’t the first poker-playing bot. The team that built it first built Libratus, which had beaten professionals in two-player games.

Pluribus was tested during a 12-day tourney that saw more than 10,000 hands played and involved 15 human players.

“A lot of AI researchers didn’t think it was possible to do this using [our] techniques,” Noam Brown of Carnegie Mellon University told the science journal Nature. Brown developed Pluribus along with Tuomas Sandholm.

What’s remarkable about Pluribus is that it doesn’t relay entirely on game theory, which becomes less reliable as more players are added. Instead it relies on a method that allows the program to make good choices by looking a few moves ahead rather than looking at the game’s endpoint. It teaches itself through reinforcement learning what decisions result in greater winnings.

The breakthrough lays the foundation for future AIs to tackle more complex problem-solving.

Pluribus runs on just two central processing units. When playing against itself, it can complete a hand in 20 seconds, about twice as fast as humans can accomplish the task.

"It's a … scalable approach to dealing with [complex information] that could quickly make a very good decision even better than people," Murray Campbell, a senior researcher at IBM in Armonk, New York, and one of the creators of Deep Blue, told Science Magazine.