Update: Last time AI took on humans at poker, the flesh-and-blood brigade won comfortably. This time, not so much. Libratus has won $1,766,250 worth of chips at no-limit Texas Hold’em, knocking the stuffing out of its four human opponents in the process. “Yeah, this was a beat-down,” was all human opponent Jimmy Chou could say after finishing his 30,000th hand on Monday.
Libratus’ co-creator, PhD student Noam Brown, was delighted with the AI’s performance, telling The Guardian, “When I see the bot bluff the humans, I’m like, ‘I didn’t tell it to do that. I had no idea it was even capable of doing that.’ It’s satisfying to know I created something that can do that.”
For the human players, a demoralising defeat. Jason Les, who played against Libratus’ predecessor two years ago, described the experience as “demoralising.”
“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.”
Libratus’ aggressive poker technique also ended up changing the way the humans played, as like its predecessor, it would occasionally make huge bets on relatively small pots. “It’s just not something a human would normally do, but it forces you to be on your toes for each game,” said Les. “It’s almost like we’ve been shellshocked into being much stronger players. Nothing anyone does will seem that crazy any more.”
The lesson may be never to gamble against a machine, but you may end up wishing you had bet on it. “The international betting sites put us as 4-1 underdog and the humans expected to win,” commented Carnegie Mellon University’s Tuomas Sandholm.
The original article continues below.
Can an AI beat the world’s best players at poker? In 2015, the answer was no, but computer scientists from Carnegie Mellon University (CMU) have a new AI that’s ready to step up to the table in a 20 day “Brains vs Artificial Intelligence” match at Rivers Casino in Pittsburgh.
The AI – known as Libratus – is the successor to Claudico, the AI that lost in a similar tournament two years ago. Since then, AIs have increasingly been getting the upper hand over humans, with Google DeepMind managing to beat the human Go champion 4-1 back in March.
Libratus has learned the rules of the game with over 15 million hours of computation. That’s significantly more than the 2-3 million absorbed by Claudico, which finished fourth out of five last time around. While Claudico played well in parts, competitor Doug Polk was clear it still had a thing or two to learn: “Where a human might place a bet worth half or three-quarters of the pot, ‘Claudico’ would sometimes bet a miserly 10% or an over-the-top 1,000%,” he said at the time. “Betting $19,000 to win a $700 pot just isn’t something that a person would do.”
Two of the competitors that played Claudico – Jason Les and Dong Kim – return to face Libratus, with the competitors all vying for shares of a $200,000 prize. You can watch the entire event on Twitch with a stream provided for each of the human player, but early signs are that Libratus is going to be a tougher opponent than Claudico proved.
After the first day in Pittsburgh, Libratus is in front, but with just 2,840 hands played out of a total of 120,000 there’s still all to play for.
“I expected it to play better, and it is playing better,” Les told the Pittsburgh Post Gazette during the first day. “It’s playing a more reasonable game, but it’s a small sample. It’s just too early to tell.”
It was also playing a lot slower than reckless Claudico, sometimes taking over ten seconds to make the kind of choice that the human players were making instantly.
We’ll know if AI has conquered another game in 20 days… or perhaps a bit longer, if Libratus continues to drag its virtual feet.
Lead image: Morgan used under Creative Commons
Disclaimer: Some pages on this site may include an affiliate link. This does not effect our editorial in any way.
