An AI Developed by Facebook and Carnegie Mellon Beats Champions of Poker
Katy Lederer, author of Poker Face: A Girlhood Among Gamblers, once said, "the cardinal sin in poker is ...becoming emotionally involved."
So what better player than an artificial intelligence not capable of real emotion?
An AI program developed by Carnegie Mellon University, in collaboration with Facebook AI is just that, and it has defeated some of the world's best at six-player no-limit Texas hold'em poker.
The AI, called Pluribus, defeated Darren Elias, the holder of the record for most World Poker Tour titles, and Chris "Jesus" Ferguson, winner of six World Series of Poker events.
Each pro separately played 5,000 hands of poker against five versions of Pluribus. Another experiment saw Pluribus take on 13 pros, all of whom have won more than $1 million playing poker.
Pluribus also emerged victorious after playing against five pros at a time — totaling 10,000 hands.
Pluribus was developed by Tuomas Sandholm, Angel Jordan Professor of Computer Science, and Noam Brown, who is currently finishing his doctorate in Carnegie Mellon's Computer Science Department as a research scientist at Facebook AI.
In a press release, Sandholm spoke of the milestone they have reached with this new AI:
"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."
Developing a remarkable AI
Pluribus computed its "blueprint" strategy by playing six copies of itself. A new limited-lookahead search algorithm is the main source of Pluribus's superhuman ability to win at multi-player poker.
The AI made remarkably efficient use of computation. Other AIs that have achieved significant milestones in games, such as DeepMind, have typically used large numbers of servers and farms of GPUs. Pluribus computed its blueprint strategy in eight days using just 12,400 core hours and only 28 cores during live play.
"We're elated with [Pluribus'] performance and believe some of its playing strategies might even change the way pros play the game," said Noam Brown.
A research paper describing the way Pluribus was made, has been published today by the journal Science.