Artificial intelligence is at it again. In a poker tournament that ended on 30 January, Libratus-a poker playing bot- beat 4 of the best poker players in the world. The 20-day tournament dubbed as “Brain vs. Artificial Intelligence featured 4 human players -Dong Kim, Jason Les, Jimmy Chou and Daniel McAulay.
Each of the players spent 11 hours every day behind computer screens at the Rivers Casino in Pittsburgh playing against the Libratus software at a no-limit Texas Hold’em; a two player unlimited form of poker. At the end of the tournament, Libratus had beaten all the 4 players and had winnings of $1.7 million in chips. Luckily the players were not using real chips.
Libratus marks yet another milestone for artificial intelligence after bots beat human beings in Chess and Go. AI beating humans in poker looked like a pipe dream with betting sites giving the human players a 4-1 odd over the machine.
Poker poses a greater challenge than both chess and Go due to the amount of imperfect information involved. Chess and Go players can see the entire board while poker players don’t get to see each other’s hands. The artificial intelligence in Libratus is also forced to bluff and correctly interpret incorrect information.
“This challenge is so huge and complicated that it’s been elusive to AI researchers until now,” said Tuomas Sandholm, a Carnegie Mellon University Professor of computer science who build Libratus along Noam Brown-his Ph.D. student. Sandholm himself did not have confidence in Libratus beating human players.
Libratus is an improvement on Sandholm’s and Browns previous poker-playing AI called Claudio which lost the tournament in 2015. The new AI system is an overhaul of Claudico with more computing power and a different algorithmic approach to the way the system deals with determining and handling misinformation.
“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 refined its approach to the game and built a winning strategy by playing trillions of hands randomly.
Every day after the poker game ended, Libratus was connected to Pittsburgh Supercomputer center’s Bridges computer to run algorithms and sharpen its winning strategy overnight. Brown would then spend 2 hours in the morning to bring it back up and running.
“Libratus turned out to be way better than we imagined. It’s slightly demoralizing,” said Jason Les, one of the poker pros who also played against Claudico in 2015. The 4 players get to split a $200,000 prize depending on how well they did against Libratus.
“When I see the bot bluff the humans, I’m like, I didn’t tell it to do that. I had no idea it was even capable of doing that.’ It’s satisfying to know I created something that can do that.” Said Brown.
The algorithm on Libratus is not specific to poker and can be used in other games, negotiating business deals, in setting up cyber security and military strategy and in planning for medical equipment. The algorithm is useful in any situation requiring human decisions with imperfect information.
“Poker is the least of our concerns here, you have a machine that can kick your ass in business and military applications. I’m worried about how humanity as a whole will deal with that,” said Roman V Yampolskiy, a professor of computer science at the University of Louisville.