UPDATE: Humans Overcome AI in Poker Competition
Kagan Pittman posted on May 11, 2015 |

The hands have been settled, the chips have been counted and the results are scientifically unreliable.

Three out of four of the world’s best poker players have defeated the AI program Claudico in 80,000 hands of Heads-Up No-Limit Texas Hold’em.

Each two-player game consisted of 20,000 hands with $170 million dollars “bet” over the course of two weeks. No wagers actually took place.

Despite Loss, Claudico holds Statistical Tie

Human player’s left the tables with a $732,713 dollar collective lead over the AI program. This unfortunately wasn’t large enough a winning to be statistical significant.

Bjorn Li led the pack in the final chip tally with $529,033 dollars over Claudico. Doug Polk followed with $213,671 dollars, with Dong Kim’s winnings at $70,491 dollars.

Poker pro Jason Les didn’t share in the others luck, falling to Claudico’s winnings of $80,482 dollars.

"We knew Claudico was the strongest computer poker program in the world, but we had no idea before this competition how it would fare against four Top 10 poker players," says Tuomas Sandholm, computer science professor at Carnegie Mellon University (CMU), who directed development of Claudico.

"It would have been no shame for Claudico to lose to a set of such talented pros, so even pulling off a statistical tie with them is a tremendous achievement."

As players don't know what cards their opponents hold and mislead each other by bluffing, slow play or other tactics, poker is an incomplete information game. This makes poker an effective testing tool for AI, explained Sandholm.

Carnegie Mellon University Team have New Ideas for Improved AI

"Beating humans isn't really our goal; it's just a milestone along the way," Sandholm said. "What we want to do is create an artificial intelligence that can help humans negotiate or make decisions in situations where they can't know all of the facts."

There’s still some way to go to reach that goal as players noted some odd strategies that no human would logically make.

Where someone might place a bet worth half or three-quarters of the pot, Claudico would sometimes bet only 10 percent or go wild betting 1,000 percent. "Betting $19,000 dollars to win a $700 dollar pot just isn't something that a person would do," Polk observed.

Claudico sets its own strategy, occupying about two terabytes of data – far more than the CMU team could analyze.

The Blacklight supercomputer was used throughout the event to compute a better approximation of game-theory-optimal strategy. The work with Blacklight was supported in part by an allocation from XSEDE, the National Science Foundation's network of supercomputing resources. 

Believing AI will soon be able to outplay top professionals, Sandholm and his team already have ideas for improving the algorithms at the heart of the program. With 80,000 hands of data, scientists can train, test and perfect successors to Claudico.

The Brains vs AI competition took place at Pittsburgh’s Rivers Casino, coming to a close May 11.

Are you surprised by the results? Do you agree with Sandholm, thinking AI will one day outplay the world’s best in poker?

Let us know your thoughts in the comments below.

If you want to learn more about the Brains vs AI competition and those involved, visit RiversCasino.com.

May 4, 2014

The Claudico AI program isn’t living up to expectations as it falls behind three of four poker champions after 40,000 hands of Heads-Up No-Limit Texas Hold’em. What weapons are the humans using to keep an edge over the machine? A pen and paper.

"Playing Claudico has been unlike any other game I've ever played," said Doug Polk, one of the four poker pros in the Brains vs. AI competition.

"I've been taking notes along the way when it makes a move that I wasn't expecting so that I can continue to improve my strategy and make sure I walk away with a win."

You’d think this low tech answer might not be enough to work against a computer hooked up to the Blacklight supercomputer, described below. Neither did Tuomas Sandholm, professor at Carnegie Mellon who led development of Claudico.

"Claudico performs real-time reasoning while playing a hand and improves its strategy during the match by continuously computing,” said Sandholm.

“I have been extremely impressed by the pros' ability to improve and adjust their game no matter what curve balls Claudico has thrown at them.”

Polk and fellow human players Dong Kim, Bjorn Li and Jason Les play 750-800 hands throughout each day. Sunday, May 3 is set aside so the players can rest.

There are still another 40,000 hands to play before the competition ends and winners are announced, scheduled for May 8.

For more information read below, or visit the Brains vs. AI homepage.

To see the leaderboard, latest scores, or watch the live stream, visit Riverscasino.com.

---

April 24, 2015

The battle of man versus machine continues as computer program, Claudico, faces off against four of the world’s best poker players.

Claudico will play 20,000 hands of Heads-Up No-Limit Texas Hold’em in settings designed to remove luck and ensure the most scientifically significant outcomes.

Players will be paired with Claudico individually, playing duplicate matches.

Player A will have the same cards as the computer facing Player B, and vice versa. This same scenario is in effect for players C and D.

How Claudico works

Developed by Carnegie Mellon University (CMU), Claudico had an earlier version named Tartanian7. This first version won a similar competition in July 2014.

Despite its victory Tuomas Sandholm, a professor of computer science at CMU and lead developer of Claudico, believed the program didn’t prove AI had any edge over humans. 

Above: Tuomas Sandholm, a professor of computer science at Carnegie Mellon who has led development of Claudico.
Above: Tuomas Sandholm, a professor of computer science at Carnegie Mellon who has led development of Claudico.

Computers have demonstrated they can outplay humans at the simpler game of Heads-Up Limit Texas Hold'em, Sandholm noted, but not the far more complicated no-limit version.

Two-player no-limit Hold'em has 10161 (one followed by 161 zeroes) possible scenarios. That’s more than there are atoms in the universe.

By contrast, the easier game of Limit Hold'em, in which bets and raises are limited to a pre-determined amount, has only 1013 scenarios.

To cope with processing all of this information, Sandholm and his Ph.D. students, Sam Ganzfried and Noam Brown, use the Pittsburgh Supercomputing Center's Blacklight supercomputer to process Claudico's strategy.
Blacklight: The SGI Altix UV1000 system
Blacklight: The SGI Altix UV1000 system

Blacklight has a random access memory of 16 trillion bytes, roughly 8,000 times more than the most powerful tablet computers.

Claudico will run on a Carnegie Mellon computer during the competition, but will continuously improve its strategy while connected to Blacklight.

What Claudico means for engineers

Could future AI technology developed from Claudico and Blacklight be able to assess a nearly infinite number of scenarios? For engineers in every field could this AI analyse every conceivable problem, proposing every possible solution with the press of a few buttons?

As farfetched as this possibility sounds, this could be the beginning of the rise in true artificial intelligence, a true rise of the machines.

And it may all start with a poker game.

Claudico will be facing pro poker players Doug Polk, Dong Kim, Bjorn Li and Jason Les at the "Brains Vs. Artificial Intelligence" competition beginning April 24 at Rivers Casino, Pittsburgh, PA.

Let us know your thoughts on AI in the comments.

Brains Vs. Artificial Intelligence Live Streams

For more information, visit https://www.cs.cmu.edu/brains-vs-ai.

Recommended For You