You are here

Artificial Intelligence Shows Its Hand

PITTSBURGH—Dong Kim is one of the best poker players in the world, but he has never played against an opponent quite like the one he just faced here.

The player bluffed brilliantly, never got tired or frustrated and, over the course of a nearly-three-week tournament last month, only got better.

“In the midst of it, I was really stressed out,” said Mr. Kim, 28 years old. “I’m not used to losing.”

Libratus, an artificial intelligence program developed by Carnegie Mellon University, handily beat Mr. Kim and three other top players in Heads-up, No-Limit Texas Hold’em, winning by a total of $1.7 million in chips.

Noam Brown, at center, a computer science Ph.D student, helps Daniel McAulay, a professional poker player, at the start of a tournament at Rivers Casino in Pittsburgh, on Jan. 11, 2017. Human poker players challenged an artificial- intelligence computer program developed at Carnegie Mellon University that uses algorithms to make decisions based on ‘imperfect information.’ PHOTO: ANDREW RUSH/ASSOCIATED PRESS

The creator of Libratus, Tuomas Sandholm, professor of computer science at the school, called the 20-day tournament a “once-in-a-lifetime event” for artificial intelligence, and another step toward applying AI to business negotiations, military strategy, medical-treatment planning, cyber-security and other areas.

Prof. Sandholm said Libratus’s victory was statistically significant and not a matter of luck. “I thought we had a 50-50 shot, but to have such a huge victory I would have never guessed,” Prof. Sandholm said. “We have proven that the best AI is better than the best humans.”

Texas Hold’em has been an elusive goal for AI. Unlike chess, Othello or other games in which all of the information needed to make decisions is out in the open, the face-down cards in Texas Hold’em require poker players to make decisions based on “imperfect information.” The “no limit” version of the game, in which players can bet or raise up to all their chips, is more complex than the limit version.

There are an astronomical number of potential situations in Head’s-Up No-Limit Texas Hold’em, namely 10 to the 160th power. That’s a 1 followed by 160 zeroes.

During the tournament, the four professionals each played two games simultaneously on computer screens in the poker room at the Rivers Casino. The AI program played them one-on-one over a total of 120,000 hands, or roughly 1,500 hands per player each day; winning was tracked in chips, not games. In-person attendance was often sparse, but thousands tuned in to watch from 151 countries on Twitch, a video platform for gamers.

Libratus won chips worth between $85,000 and $888,000 from each player, Mr. Sandholm said, partly because its algorithms plugged holes in strategies that the professionals were initially able to exploit. The program got smarter as play went on. That includes its ability to bluff, which Mr. Sandholm said has more to do with math than with psychology.

It also didn’t need sleep.

Mr. Kim, who divides his time between Vancouver and South Korea, said sleep was sometimes in short supply as he and the other players met into the evening trying to figure out strategies on how to beat the machine.

Mr. Kim was the top human player, and walked away with $74,500 of a $200,000 purse. The chips won by Libratus were effectively play money.

Mr. Kim said he thinks AI will ultimately help humans, though it might make some jobs obsolete, including possibly his own.

“I’m like one of the best in the world in ’Head’s-Up, No-Limit,’ ” he said. “But it’s saying that there’s a process better than me that I cannot beat.”

Prof. Sandholm is more optimistic. He said AI that he created has been running a nationwide kidney-exchange program since 2010 and saved hundreds of lives.

“When people talk about Terminator robots that have their own ideas and don’t obey human commands, there’s no accountability to have to be realistic,” he said.

“One thing I feel very strongly about,” he added, “is that AI is going to improve the lives of people and companies by way more than it’s going to hurt them.”