According to country star Kenny Rogers, playing poker is a matter of knowing when to fold, hold, walk away or, in seedier circumstances, run. Researchers from Carnegie Mellon University tested that theory recently with Libratus, an artificial intelligence (AI) program developed to play a mean game of cards.
Libratus just made history by defeating four of the world’s best professional poker players during the Brains Vs. Artificial Intelligence: Upping the Ante challenge at Rivers Casino. The event was a rematch of a 2015 tournament where another CMU AI, Claudico, failed to best its opponents.
According to SpinPalace Casino, the event was historic. On January 30, after a marathon 20-day competition and a total of 120,000 hands of Heads-up No-Limit Texas Hold’em, Libratus led by a collective $1,766,250 in chips.
The event presented a prime opportunity for CMU innovators to show off their skills, but what does it mean for the future of AI?
“Poker has been a benchmark in the field of AI for a long time,” says Noam Brown, a computer science PhD student at CMU who developed the code for Libratus.
As Frank Pfenning of CMU’s School of Computer Science points out in an official statement, Libratus adds to other man-versus-machine victories the school has played a role in, including Deep Blue beating chess master Garry Kasparov and Watson dominating Jeopardy!. Unlike those previous demonstrations, however, the Libratus win opens up new possibilities for AI.
“There have been similar milestones in AI in the past, but those were games of perfect information. Those were games where both players could see everything that was going on,” says Brown. “Poker is a fundamentally different kind of game because each player only has access to limited information. They can’t see what cards the other player is holding, so all the techniques that work for perfect information games like chess don’t apply.”
To prepare Libratus for the big event, Brown and his partner, CMU computer science professor Tuomas Sandholm, took a more hands-off approach.
“Everything it learned it learned on its own,” says Brown, adding that it had never seen a hand of poker being played by humans before the start of the competition.
Equipped with the basic rules and description of poker, Libratus played trillions of hands. What began as a random process became more refined over time as it gradually started to recognize what actions would result in winning the most money, including when to bet or fold.
According to recent posts about this on www.dreamjackpot.com, their approach was based on the Nash equilibrium, a gaming strategy that guarantees a player’s likelihood of winning in the long run. Named after celebrated mathematician John Forbes Nash Jr., the theory posits that, in a game like poker, no player has anything to gain by deviating from his or her own strategy.
In other words, while poker may seem like a game of chance compared to the likes of chess, the Libratus victory was no mere stroke of luck.
While there are obvious advantages for the gambling industry to adopt an AI like Libratus, Brown and his colleagues stress more practical applications. They believe areas like business negotiation, military strategy, cybersecurity and medical treatment planning could benefit from the technology because it better represents decision-making in the real world.
“The computer can’t win at poker if it can’t bluff,” said Pfenning. “Developing an AI that can do that successfully is a tremendous step forward scientifically and has numerous applications. Imagine that your smartphone will someday be able to negotiate the best price on a new car for you. That’s just the beginning.”
Statistics sourced: https://www.easyslots.com