‘Historic’ win for AI as machine destroys top poker players
Computers have long held the upper hand in many games but, because it is an “incomplete information game”, poker has remained stubbornly immune - until now.
Unlike perfect-information games, such as chess, imperfect-information games cannot be broken down into smaller games to be solved separately.
During a 20-day game of no-limit Texas Hold ‘Em Libratus, an AI built by two Carnegie Mellon researchers thrashed four of the world’s top players.
Poker poses many difficulties for an AI because players don’t know what cards their opponents hold and all players try to trick their opponents by bluffing, slow play, and other tactics.
No limit Texas Hold ‘Em is even more of a challenge because it is a particularly complex form of poker that relies heavily on long-term betting strategies and game theory. In no-limit Hold ‘Em, the goal is to win the most money rather than each hand.
As a result good players use strategies that play out over dozens of hands - sometimes losing on purpose and betting low even though they have good cards.
“It splits its bets into three, four, five different sizes,” one of the competitors, Daniel McAulay, explained to Wired about Libratus’ strategy. “No human has the ability to do that.”
Labratus was designed by Professor Tuomas Sandholm and grad student Noam Brown. During the 20 day competition, the AI comfortably beat its four competitors by raking in more than $1.7 million worth of chips.
Andrew Ng, chief Scientist at Baidu, the Chinese internet company and one of the biggest web corporations in the world, described result as a “stunning accomplishment” which “made history.” He compared it to the iconic 1997 victory of IBM’s Deep Blue over then-reigning chess world champion, Russia’s Gary Kasparov.
CMU just made history: AI beats top humans at Texas Hold'em poker. A stunning accomplishment, comparable to Deep Blue & AlphaGo!— Andrew Ng (@AndrewYNg) January 31, 2017
The technology has far greater uses than just making professional gamblers feel bad.
The algorithms used by Libratus could also be used to create strategies for applications involving cybersecurity, business transactions, and medicine, which are all areas of research Sandholm and his colleagues work on in Carnegie Mellon.