A new paper published by Science today details how Libratus, an AI developed by researchers at Carnegie Mellon’s computer science department, managed to take on and defeat top industry professionals at one of the most challenging forms of poker: No-limit Texas Hold’em. Yes, the same variant of poker that swept the nation during the heady days of the early to mid-aughts.
This is significant because no-limit Texas Hold’em is what’s called an “imperfect-information game,” which means that not all information about all elements in play is available to all players at all times. That’s in contrast to games like Go and Chess, both of which feature a board which contains all the pieces in play, plainly visible to both competitors.
CMU’s team detailed Libratus and some of its early successes back in January, but with the publication of the full scientific paper today, we can see how its made progress and understand moe fully its successes in taking on humans in this particularly human game.
Libratus is most interesting because it’s managed to master a game where bluffing is a core, necessary component. Determining when and how to bluff separates adequate players from the truly transcendental, and bluffing is all about imperfect-information gaming, since it involves predicting or guessing at the unpredictable behaviors of an opponent who has a potentially completely different set of information from your own.
Also, poker is a game that spans many individual hands, which means that in strategizing for overall victory a player has to be willing to take individual, strategic losses and look at the bigger picture. This is another element of complexity that’s not typically something computers are really great at managing.
Over the course of a 20 day competition, with 120,000 poker hands played in total and a prize pool of $200,000, Libratus defeated top human pros – all using techniques that the researchers say aren’t uniquely applicable to poker, but that could apply to a broad range of imperfect-information games in general.