On January 11, 2017, four of the world’s top poker players will sit down at a table at the Rivers Casino in Pittsburgh for a 20-day tourney, playing 120,000 hands of Hands-Up No-Limit Texas Hold ’Em.
Also at the table: Libratus, a computer program developed by scientists at Carnegie Mellon University.
This isn’t the first time a computer has challenged a human at the table. IBM's Deep Blue famously bested chess grandmaster Garry Kasparov in 1997, Google’s Deep Mind beat Go World Champion Lee Sedol last March and an AI program developed by scientists at the University of Alberta beat a group of elite poker players at Texas Hold’em in 2008.
But chess and Go are games of mathematical strategy, for which formulas can be worked out in advance. Poker, on the other hand, is an imperfect game, and much of its skill lies not in the card strategy as much as in reading the tells of your opponents. The Texas Hold’em game in Alberta also had a limit. The fact that this game is No Limit introduces a further variable that Libratus's programmers must account for.
So how will they do it? Facial recognition to learn and recognize tells? Are we entering a Black Mirror-esque scenario where computers will read us like books? Not at all. In fact, facial recognition is still quite hard for computers; they’re still seeing faces where there aren’t faces at all.
Instead, the CMU folks are taking a page out of the Nash Equilibrium, aka game theory, whereby in a zero sum game the most rational response is the best one. In other words, Libratus will be playing a very conservative brand of poker. And also a boring one. The players and scientists testing Libratus have described the games as “a grind.”
As reported by The Verge, “playing the game safe is not the same as playing the game conventionally,” noting that Libratus limps and donks, moves that nobody does but have made playing against the machine quite laborious.
But seeing as machines don’t fatigue like humans, that strategy may prove a strong play.
You can watch a livestream of the game on Twitch.