An artificial intelligence software called Libratus beat four of the world's best poker players, who lost $1.5 million, in a three-week-long no-limit Texas Hold'em competition held in a Pittsburgh casino.
The four human players - Jason Les, Dong Kim, Jimmy Chou and Daniel McAulay - spent 11 hours each day at the Rivers Casino in Pittsburgh battling at the two-player unlimited form of poker.
In the past, machines have beat humans at other games such as chess and Go, but this is the first time that it has happened with poker, which is much more difficult because it is a game with imperfect information. With chess and Go, each player can see the entire board whereas with poker, players do not get to see each other's hands and on the top of that, the AI is required to bluff and correctly interpret misleading information to win, according to the Guardian.
Tuomas Sandholm, Carnegie Mellon University professor of computer science, built Libratus alongwith his Ph.D. student Noam Brown. "This challenge is so huge and complicated that it's been elusive to AI researchers until now," said Sandholm.
Sandholm said that he "wasn't confident at all" about Libratus beating the poker pros. However, after the match, Brown said that the human players "put up the best fight they could."
Libratus is an improvement over Sandholm and Brown's previous poker-playing AI called Claudico. Claudico had competed in the same tournament in 2015 but lost against four poker pros.
Libratus not only has more computing power but an enhanced algorithmic approach to the game, particularly the way it deals with imperfect or hidden information, Sandholm said. Libratus was not told how to play poker, but instead was given the rules of poker and told to "learn on your own".
"We didn't tell Libratus how to play poker. We gave it the rules of poker and said 'learn on your own,'" said Brown. The bot started playing randomly. Over the course of playing trillions of hands, it was able to refine its approach and arrive at a winning strategy.