Here’s startling news: Artificial Intelligence (AI) scientists have developed a computer program that can beat the pros playing no-limit Texas hold’em. As complicated as the game is and with so many important decisions to be made, AI is proving a computer can beat the best human players.
The Jan. 24 issue of the “MIT Technology Review” provides a fascinating discussion of this amazing feat. It wasn’t easy to accomplish; it took some of the most brilliant AI scientists in the world. I am proud to say I graduated from there in 1952. Let me share the information in this special report.
Will Knight, MIT Technology Review’s senior editor for AI, explains, “playing poker involves dealing with imperfect information, which makes the game very complex, and more like many real-world situations.”
At the Rivers Casino in Pittsburgh, a computer program called Libratus is in the process of proving computers can outplay any human poker player – even the best of us.
As I write this, Libratus is playing thousands of games of heads-up, no-limit Texas hold’em against several expert professional poker players. A little more than halfway through the 20-day tournament, Libratus was ahead by almost $800,000 – a commanding lead. (Ed. Note: The final winner was Libratus with a $1.7 million lead.)
According to Knight, “a win for Libratus would be a huge achievement in artificial intelligence.”
“Poker requires both knowledge and reasoning that has proven difficult for machines to imitate,” he explained. “It is fundamentally different from checkers, chess, or Go, because an opponent’s hand remains hidden from view during play.
“In games of ‘imperfect information,’ it is enormously complicated to figure out the ideal strategy given every possible approach your opponent may be taking. And no-limit Texas hold’em is especially challenging because an opponent could essentially bet any amount.”
After 12 years of research, Libratus was created by Tuomas Sandholm, a professor in the Computer Science Department at Carnegie Mellon University in Pittsburgh, with his graduate student Noam Brown. Sandholm is an expert on game theory and AI.
According to Andrew Ng, chief scientist at Baidu, a prominent Chinese website published in English, “poker has been one of the hardest games for AI to crack. There is no single optimal move, but instead an AI player has to randomize its actions so as to make opponents uncertain when it is bluffing.”
The AI researchers used game theory – the mathematics of strategic decision making – to find the best strategy for each hand, while faced with a variety of uncertainties. Because the possibilities are so vast, this usually involves making appropriate approximations – quite a task!
It has not been easy. Sandholm’s team had developed a previous poker-playing computer program, called Claudico. Last year, it was soundly beaten in a match against several poker pros. Libratus uses several new advances to achieve a higher level of play, including a new approximation technique, Sandholm explains, as well as several new methods for analyzing possible outcomes as the cards are dealt up on the board. “This end-game analysis is computationally very challenging, and is performed during each game at the Pittsburgh Supercomputing Center, a facility operated by CMU and the University of Pittsburgh.
Knight’s report also includes a relevant quote by Vincent Conitzer, a professor at Duke University who teaches AI and game theory: “Whether a move is good or not depends on things you cannot observe. This also results in a need to be unpredictable. If you never bluff, you are not a good player. If you always bluff, you are not a good player. Game theory tells you how to randomize your play in a way that is, in a sense, optimal.”
To this I might add: “Wow! What’s next?” But, wait: To date, Libratus is limited to heads-up play.