The latest artificial intelligence (AI) vs. human poker showdown has seen the free-thinking computer open up an early lead at Rivers Casino in Pittsburgh, PA.
Following on the heels of the 2015 battle between a computer program known as Claudico and four pros (Dong Kim, Jason Les, Bjorn Li and Doug Polk), Carnegie Mellon University is at it again with another AI creation.
A Bright Start for the Computer
Although Tuomas Sandholm’s Claudico wasn’t able to defeat the quartet of pros in a series of heads-up No Limit Hold’em matches, the result was close enough to suggest that it was possible.
After taking the best bits from Claudico and developing some new ones with his team at Carnegie Mellon University’s AI department, Sandholm returned at the start of 2017 with a new challenge. This time, four poker pros will take on a new machine known as Libratus over 120,000 No Limit Hold’em hands.
The series of heads-up matches is due to run for 20 days, but Libratus is already starting to show signs that it may defeat the pros.
Although it’s still too early to tell whether or not the computer brain has evolved enough to beat the humans, there is a chance the quartet won’t be winning a share of the $200,000 top prize.
According to the stats after the first day of play on January 11, Libratus was ahead after 2,840 hands thanks to the following results:
Libratus vs. Jimmy Chou – Chou led by $2,290
Libratus vs. Daniel McAulay – McAulay led by $4,938
Libratus vs. Dong Kim – Libratus led by $60,305
Libratus vs. Jason Les – Libratus led by $21,411
More than Claudico 2.0
With thousands more hands to play, the sample size is currently too small to draw any meaningful conclusions. But, returning pro Les has said that he’s seen a definite improvement in the software. Having been victorious against Claudico in 2015, Les told the Pittsburgh Post Gazette that Libratus is playing a more “reasonable” game and making decisions more like a human pro.
In addition to more solid fundamentals, it’s also been noted that Libratus is varying its range a lot better and, interestingly, taking more time with its decisions.
This behavior, according to Sandholm, is Libratus thinking through the hand. But, it’s also having the benefit of frustrating the players according to the Post Gazette.
Should Libratus win the match, it would not only be a breakthrough for gaming technology, but for AI as an industry. If computers can solve a game such as No Limit Hold’em by actively learning and improving, the applications are potentially endless.
From healthcare and cyber security to self-driving cars, the result of Libratus vs. the pros could have a major impact on the tech world in 2017.