AI Heads-Up Poker Programs Now Capable of Beating the Pros

Home » Poker News » AI Heads-Up Poker Programs Now Capable of Beating the Pros

Tuesday, April 25th, 2017 | Written by Adam Brown

At the end of January, a Carnegie Mellon University (CMU) developed piece of AI software soundly thrashed a four player team of top poker pros, who ended their 20 day, 120,000 hand drumming in the red by a virtual $1,766,250. The AI program responsible for their defeat was called Libratus, and just over two months later CMU sent another one of its creations, this time called Lengpudashi, to take part in a mini-sized version of the original match in the Chinese province of Hainan. The result of the contest was much the same, though, with Team Dragon losing the Human versus Machine battle by a resounding $792,000 virtual dollars.

Humans versus Machine

Strategic Machine Inc. is a company that was created by a CMU professor of computer science named Tuomas Sandholm, who also co-created the AI program Libratus/Lengpudashi together with Noam Brown, a computer science Ph. D. student. The creators of this company offered a prize of $290,000 against a team of human players called Team Dragons, which was headed up by a Shangai winner of the 2016 WSOP $5k NLHE event.

This competition was put into action by a Carnegie Melon alumnus, Kai-Fu Lee, who is also a former staff member and the CEO of Sinovation Ventures, an early-stage venture capital company that takes up investments in Chinese and American startups. Lee is a previous Apple, Google and Microsoft executive, and is also one of the most prominent people in the internet sector of China. Hainan Resort Software Community co-hosted the event with Sinovation.

A Different Sort of Contest

The intricate betting strategies and bluffing techniques of poker make it an interesting phenomena to AI researchers, as player can decide to bluff, bet or fold without ever seeing the hand of the opponent.This represents a different sort of challenge than say such games such as Go or chess where all of the game pieces can be clearly seen on the game board.

Unlike the previous match against professional poker players, however, this time around Team Dragon included WSOP bracelet winner Alan Du, who is full-time businessman, as well as a team of computer scientists, engineers, and investors. Du and his teammates were trying to win where the US pros had lost by attempting to employ their understanding of artificial intelligence, and game theory to win against Lengpudashi.

Lengpudashi Triumphs

The artificial intelligence program created and developed by researchers at the Carnegie Mellon University took part in a 5-day heads-up No-Limit Texas Hold’em contest against six Chinese players between April 6th and 10th. The Lengpudashi program they faced was a new version of Libratus, the AI software that beat a number of top US poker players at the Rivers Casino in Pittsburgh, Pennsylvania. Even though Lengpudashi and Libratus played a much different number of hands during their unique events, however, Lengpudashi’s last margin of winning was much larger.

In the end, this turned out to be 220 milli-big-blinds for every game played, opposed to 147 milli-big-blinds for each of the games, which was what Libratus achieved. These metrics, the milli-big blind, stands as an indicator of a fraction of the bets that are needed in order to succeed at a poker game, and milli-big-blinds for each of the games is a typical way of pitting poker efficiencies against one another. Nevertheless, all the players who competed against the AI programs, whether American or Chinese, alla greed that the machine was much better than they had ever anticipated, and as Jason Les commented after losing in January:

“Libratus turned out to be way better than we imagined. It’s slightly demoralizing. If you play a human and lose, you can stop, take a break. Here we have to show up to take a beating every day for 11 hours a day. It’s a real different emotional experience when you’re not used to losing that often.”

Applications Beyond Poker

Strategic Machine has an exclusive license over Libratus alongside a variety of other technologies from the Carnegie Mellon University’s laboratory in Sandholm. Despite the novelty of defeating some of the best heads-up poker players in the world, the development highlights the ever increasing applications that AI can be applied towards. Strategic Machine, for instance, targets a large spectrum of applications, such as cyber security, physical security, poker and other games, negotiation, military applications, business strategy, strategic pricing, medical treatment planning, finance, political campaigns and auctions.