An AI poker gamer created by research scientists from Facebook AI and Carnegie Mellon University has defeated two of the World’s best poker players.
Unlike other poker-playing AI models today that can only take on as many as two players, Facebook and CMU’s AI program called Pluribus was designed to play with multiple human players.
During the experiment, detailed in a paper published in the journal Science, Pluribus was able to take on 15 professional poker players in a six-player no-limit Texas Hold ‘Em game. That includes four-time World Poker Tour title holder Darren Elias and six-time winner of World Series of Poker events Chris “Jesus” Ferguson.
In an interview, Elias who helped train the AI program said:
“It’s just me and then five versions of this AI poker bot, which I would play against every day, thousands of hands.
It was improving very rapidly, where it went from being a mediocre player to basically a world-class-level poker player in a matter of days and weeks. Which was pretty scary.”
Bet on the AI Poker Gamer!
Both Ferguson and Elias were pitted against five versions of the AI poker gamer, playing over 5,000 hands each time during the research team’s first experiment. A second experiment saw the AI bot playing five pr0 players at a time for 10,000 hands.
Tuomas Sandholm, a co-author of the study, said in a press release:
“Pluribus achieved superhuman performance at multi-player poker, which is a recognized milestone in artificial intelligence and in game theory that has been open for decades.
Thus far, superhuman AI milestones in strategic reasoning have been limited to two-party competition. The ability to beat five other players in such a complicated game opens up new opportunities to use AI to solve a wide variety of real-world problems.”
Pluribus works by computing its so-called “blueprint strategy” playing six copies of itself. The AI poker gamer’s superhuman ability comes from the “limited-lookahead” search algorithm it is equipped with.
Instead of predicting the way its opponents would play their hands, the AI gamer was trained to only look two or three moves ahead. Pluribus also became a good bluffer, with its opponents praising its relentless consistency.
Noam Brown, a co-author of the study from Facebook AI, said proudly:
“It’s safe to say we’re at a superhuman level and that’s not going to change.”
Pluribus don’t need numerous servers and GPU farms to make its calculations, unlike DeepMind‘s AI gamers. In fact, the researchers were able to develop Pluribus in just eight days using a 64-core server equipped with a 512 GB RAM.
Comments (0)
Most Recent