A second poker playing machine has demonstrated the ability to beat skilled human players in heads-up no-limit hold’em.
A poker bot developed by computer scientists at the University of Alberta, Czech Technical University and Charles University in Prague soundly defeated 10 out of 11 poker pros, including Phil Laak, who each played 3,000 hands against it over a weeks-long contest, according to reports. The match began late last year.
News of the results from the computer program dubbed “DeepStack” follow in the footsteps of Carnegie Mellon Unversity’s bot “Libratus,” which crushed four elite online poker pros in a 120,000-hand match that ended in late January. DeepStack’s match actually ended just a few weeks before Libratus started its battle, as the two projects basically coincided with one another.
An earlier version of CMU’s bot lost to its human opponents in 2015.
DeepStack beat its opponents by close to 50 big blinds per 100 hands, while Libratus won by nearly 15. The discrepancy could be due to the strength of the human players and other conditions of the contests. So it’s not possible to say which machine is better.
According to reports, there’s so far no plan to have the machines play against each other. Poker AI is a field full of new and innovative research, but for now Libratus and DeepStack are widely considered to be the best.
What’s potentially next for Libratus is learning how to play six-max games.
The achievements of both bots are remarkable considering that limit hold’em, a far simpler variant of poker for computers, was solved just a few years ago.
The game of poker has long been viewed as quite the challenge for games-playing bots. Unlike chess, for example, poker involves incomplete information. No-limit hold’em also includes far more possible “moves” than there are in chess.