Flesh-and-bone poker players in the US are currently in the midst of a tournament of Heads-Up, No-Limit Texas Hold’em against a game-playing artificial intelligence (AI). Much to the surprise of the humans, however, the assumed underdog AI system is winning.
The poker-playing computer is known as Libratus and was developed in the computer science lab at Carnegie Mellon University in Pittsburgh, Pennsylvania. The test began on January 11, when it was pitted against four professional poker players in a “Brain Vs AI” poker tournament at River Casino, Pittsburgh. Over a 20-day period, they will play 120,000 hands in a series of no-limit Texas Hold ‘em games via computer screens.
“We’re so good at poker that not even a supercomputer can beat us. If we lose, we will also lose that prestige,” professional player Jason Les confidently told New Scientist before the kick-off.
However, so far, it looks like Libratus is crushing the humans. The most recent update at the time of publishing this article shows that the computer is up $471,600, with 40,340 hands dealt.
Check the video below for the very latest update. You can watch a live stream of the games on Twitch each day between 11am until around 7pm local time.
Although the AI is storming the competition at the moment, it hasn’t been completely clear sailing. According to Business Insider, day one saw Libratus up by $82,000, day two up by $150,000, and day three up by $193,000. But then after day four, the humans made a comeback and the AI’s wins fell to $151,000. Day six saw the machine’s coffers fall further to just $50,500 before it made its own comeback.
As testament to how far artificial intelligence has come, a team of top poker plays managed to swiftly defeat the machines in 2015’s competition, eventually winning with a combined total of $732,713.
Playing humans at their own games is a long-established tradition of computer science, regularly used to mark the milestones of how far the technology has come. The most recent of these landmark battles saw Google’s AlphaGo playing humans at the ancient Chinese game of Go, which the AI won.
Poker is now being called the “last frontier” for game-solving AI, as it is a game of bluffing and imperfect information, something computers aren't typically known for.
“There’s been a lot of success in these perfect information games [such as Go and chess]. Real world situations deal with hidden information, deal with uncertainty. That’s the kind of situation that poker really represents,” Noam Brown, one of the researchers on the project, explains in the video below.
"If you try and take AlphaGo, and make it play poker it wouldn't know how to bluff and it would be very easy to beat. And that is because a game like Poker and a game like Go are fundamentally different. When you introduce the element of uncertainty it changes the dynamic completely.”