The Carnegie Mellon University’s artificial intelligence (AI) program has beaten a team of human poker professionals in a heads-up no limit hold’em match.
The AI beat a team of four elite poker pros over a 120,000-hand heads-up match, played over 20 days at the Rivers Casino in Pittsburgh. The AI, Libratus, won more than $1.7m in chips over the course of the substantial game.
“It is a huge deal,” said the lead designer of the AI, Tuomas Sandholm. “Of all the games that have been seriously worked on in the AI community, this is the only game where AI has not surpassed the best humans.”
Can AI conquer poker?
Artificial Intelligence programs have beaten humans at games of skill in the past. However, poker has always presented a challenge, especially no limit hold’em. The variable sizing of bets means that the AI must deal with a large decision tree.
On top of that, poker is not a game of complete information like chess. In the game of chess, the world’s best humans cannot beat the world’s best computers. Because bluffing is a key element of poker and players cannot see their opponents’ hole cards, it is impossible to create a perfect decision tree.
The Guardian newspaper notes that this “crushing defeat for humanity” is a huge milestone for AI,” as “the algorithms that power Libratus aren’t specific to poker.” This means that the CMU AI could have many strategic applications outside of gaming.
How AI beat humans in other games
Humans have been testing AI in strategy games for decades, most famously in the game of chess.
In 1996, IBM’s computer Deep Blue became the first chess-playing computer to win both a chess game and a chess match against a reigning world champion, defeating Gary Kasparov first in two games of a six-game match and later winning a six-game rematch.
While chess remains an unsolved game, current chess computer systems are able to defeat the strongest human players relatively easily under regular game conditions. Other games present problems to strong AIs, but they continue to improve.
The ancient Asian game of Go, which is more complex than chess, recently saw 18-time world champion Lee Sodol defeated 4-1 in a best-of-five match against Google DeepMind’s AlphaGo program. This victory was a major milestone for AI, as Go had previously been regarded to be a game that AI would be unable to conquer for some time.