It is the norm that before a match begins, seasoned poker players will scrutinize videos of their upcoming opponents - to learn their playing style and analyze previous poker hands. However, this match is totally different - it would be against a supercomputer designed to play poker.
Jason Les, Dong Kim, Jimmy Chou, and Daniel McAuley - four people categorized to be amongst the best poker players in the industry, have faced off against a supercomputer for 20 days.
For 20 days, they lost.
Libratus, an artificial intelligence that plays poker developed at the Carnegie Mellon University, ultimately crushed its human opponents over the course of the 120,000-hand Brains vs. Artificial Intelligence: Upping the Ante tournament this January at the Rivers Casino on Pittsburgh's North Shore. The grand prize if they beat the AI - a share of $200,000 and the badge of having outwitted a machine - for the second time.
Even as a team, the humans never got a chance. Libratus held a 1.56 million chip lead entering the last day of the match on Monday, January 30. The poker pros had lower than 5,000 hands to try to claw their way back into contention.
In their final day, the poker pros sunk even deeper, losing over 200,000 chips and finishing the tournament with about 1.77 million behind Libratus.
Dong Kim was the one closest to Libratus throughout the tournament, finishing less than 86,000 behind the AI and winning the biggest share of the $200,000 pot split between the four poker players.
Jason Les said, "Its strategy just seemed to improve every day. It seems like you're playing against a tougher opponent every day. You end up getting in this feeling that everything you do is losing. And it's demoralizing."
The game of choice for the poker tournament was Heads-Up No-Limit Texas Hold'em poker, which has been deemed as a last stronghold of resistance to computer gaming dominance. Computers have proven their prowess by beating the best humans at chess, checkers, the game Go, ‘Jeopardy', and other poker versions.
The humans were successful at beating Claudico, an earlier version of Libratus, in the inaugural Brains vs. Artificial Intelligence tournament back in 2015. However, CMU professor Toumas Sandholm and his graduate student Noam Brown made a comeback, and this time, they presented a more powerful and sophisticated pokerbot.
Brown said after the match, "I thought we had an edge over the humans, but I didn't think it would be this large."
Even with less than 5,000 hands left to play, Brown wasn't revealing any of Libratus' secrets. He said all would be made public when he and Sandholm finishes publishing their research. On his website on Monday, Brown wrote a paper on how Libratus made its decisions on the turn and the river cards, which he will show next week at an AI conference.
Brown stated he will not allow Libratus to play with unsuspecting online poker players - as it is powered up by a $9.65 million supercomputer, therefore it's quite expensive to run - however, they will continue to study the data gathered during the tournament in order to perfect the AI's ability to make sound decisions based on imperfect information.
The game of poker naturally presents a unique challenge for AI because opponents always keep their cards concealed, which forces the computer to make decisions without knowing all the information it needs. Using Heads-Up No-Limit Texas Hold'em even adds more uncertainty of bluffing and betting into the mix.
At the start of the tournament, the dean of CMU's School of Computer Science Andrew Moore said that an AI proficient with imperfect information scenarios can be applied in different areas: health, business, military applications and many other situations. It eventually can power a cellphone app designed to purchase a car for you that knows you are willing to pay $5,000 but hides that from the sales person as it negotiates.
Poker pro Daniel McAuley said Libratus was pretty impressive. He did not intend to make excuses but pointed out that fatigue was a factor. The poker pros sometimes played against Libratus for 12 hours each day, usually ending right after 10pm. They would then eat a quick dinner before meeting together in order to study that day's hands and to plan strategies for the next day. After a few hours of sleep, they were back in front of their computers to battle Libratus yet again.
For most of the duration of the challenge, McAuley rarely left the casino or hotel. He went out to dinner once, which was last week, to Emporio: A Meatball Joint on Penn Avenue in Downtown, because it apparently had a high Yelp rating. He hoped he could check it out before he leaves the city.
On the other hand, Libratus has no need to eat or sleep, and uses most of its supercomputer programming in order to study and strategize while it plays. It seemed to adjust to the strategies played out by the humans as a whole as well as to individual tactics of play. McAuley said, "If a player started calling every bet, Libratus stopped bluffing."
When the poker pros thought they found a weakness, Libratus changed course. At the beginning, McAuley said they underestimated Libratus, "If we knew it was this good, we wouldn't have made it as obvious as to what we were doing. I would love a rematch."