Digital Conqueror - Logo
Close this search box.

Artificial Intelligence In The Lead!

When the tournament began at River Casino in Pittsburgh on January 11, the impending results could not have been imagined. The game, No Limit Texas Hold’em, is a rematch poker tournament for Carnegie Mellon University’s Libratus, an artificial intelligence poker bot, against four world-champion poker players. The four defending a huge win against another CMU AI program, Claudio, in 2015.

Libratus is an artificial intelligence program designed by Carnegie Mellon’s Tuomas Sandholm, professor of computer science, and his PhD student Noam Brown. For Sandholm this creation began as a study in abstract algorithms for sequential imperfect information games, or an easier way to picture it, games in which not all information is available when a players turn comes up, as in poker where players do not get to see what other players have in their hand. A perfect information game would be something like chess, and all have the same information to base their play on.

On the last day of tournament play, the results have been astounding. So far, Libratus has been able to win$ 1.5 million in chips from the four champions, and there are only hours to go. This is not bad given recent winnings of just over $400,00 for Christian Harder in the Pokerstars Championship Bahamas tournament earlier this month. It is also quite a different outcome than the 2015 tournament where Claudio lost $732,713 in chips after making irrational bets a human wouldn’t make.

As each player started with a bank of zero, the $1.5M is all winnings…or loss from the other four. It’s an amazing amount, and luckily for the four competing against Libratus, not real money. The prize amount is $200,000 to be divided at the end of the tournament. Regardless, a Libratus win against human reasoning would be a huge achievement in artificial intelligence, and a first in poker.

Proven success with the algorithms that fuel Libratus, which are not poker specific, could mean great breakthroughs on any front requiring strategic reasoning with imperfect information. This means advances could be made not only in recreational games, but in business, military and cybersecurity, and medical technology, as well.

At the same time, there could be concerns raised. Is the world ready to pit human intelligence against a machine that has proven skills in reasoning? The answer is obviously not far from being known. At any rate, the undoubted improvement over 2015’s result is quite impressive.