Pluribus, an artificial intelligence program developed by Carnegie Mellon University and Facebook, has become the first computer to have reached superhuman performance, beating professional Texas Hold’em players.
Pluribus came out triumphant after a twelve-day long tournament against five masterful players, who were selected from a pool and agreed to challenge the computer.
During the contest, Pluribus won $48,000 in virtual currency beating the five players who previously have each collected at least $1 million, playing the card game.
The AI program took to using a different approach to the game to that of the human participants, using varying bet amounts and embracing “donk betting” – a move that sees the first round of betting ends with a call and the next round open with a bet. Both are strategies uncommonly used by human players. Tuomas Sandholm, the developer of Pluribus explains, “I don’t understand it, but Pluribus understands it.”
The algorithm‘s use is not only limited to winning a card game but has many practical uses in different environments from negotiation strategies to investment banking and even calculating how much U.S politicians should spend on marketing during their campaign in different states.
Sandholm explains why the AI win is such a great achievement, “It’s the first time AI has achieved superhuman performance in a multiplayer game.”
Chris Ferguson, the holder of six World Series of Poker titles, took on the AI in a separate challenge, where he was defeated by the computer. He went onto say, “Pluribus is a very hard opponent to play against, it’s really hard to pin him down on any kind of hand.”
The findings were published in Science journal on July 11.