The First Man-Machine
The First Man-Versus-Machine Poker Championship
The future masters of technology will have to be light-hearted and intelligent. The machine easily masters the grim and the dumb." -- Marshall McLuhan
August 1, 2007
By Jake Gosselin
Bodog Nation Contributing Writer
Science-fiction authors have been predicting the demise of the human race at the hands (or robotic claws) of sentient robots for decades now. And each time a computer beats the world champion at one of our own games, we all feel that potential destiny take one methodical, calculated step forward.
It started with checkers in 1994 when a piece of software by the name of "Chinook" beat the reigning champion Dr. Marion Tinsley. Since then Chinook has been perfected by the scientists who created it at the University of Alberta to the point where in early August it was announced that Chinook was now officially unbeatable and had essentially "solved" checkers.
This must have been foreboding news to poker pros Phil Laak and Ali Eslami who were slated to play against Polaris in late August at the annual meeting of the Association for the Advancement of Artificial Intelligence in Vancouver, Canada. Polaris is a poker program that saw its first incarnation begin development in 1991 by the now infamous artificial intelligence team at the University of Alberta.
Laak played poker against Polaris previously and barely beat it, so this recent rematch was highly anticipated by both poker players and anyone with even a passing interest in artificial intelligence.
With computers now beating the best human chess and checkers players in the world, poker has held out as one of the last games where top human players have continued to beat the top machines.
The competition took place at an artificial intelligence conference in Vancouver starting at noon on Aug. 23 and concluded the evening of Aug. 24. It was broken down into four sessions of 500 hands of $10/$20 Limit Holdem in each session with $50,000 going to the winning team.
To balance the contest Eslami started by playing in public in a conference room while Laak played isolated in a separate room. To maintain balance, the cards dealt to Eslami were the exact same ones that the computer received while playing Laak. After 500 hands the players would switch positions.
With Eslami in the public hot seat and Laak tucked away in a separate room it was looking very good for the human team as spectators were treated to watching Eslami put a beating to Polaris.
The session ended with Eslami up by $395. But when Laak returned to share his score of -$465 it left the balance at -$70. But according to the rules of the tournament the margin of victory had to be greater than $250 to declare a winner and consequently Session One was deemed a draw.
Once again it was looking good for the human team as Phil Laak made some solid gains against Polaris. Laak spoke to the audience while he played hands and throughout the match often stated that Polaris had made a “sick” play, an indication that Laak was shocked at how well Polaris was playing.
But, despite Polaris' "sick" skills, the session ended with Laak well in the lead with $1,570. Eslami, however, returned from his game with some bad news -- he had finished at -$2,495. With a balance of -$925 the human team suffered its first defeat.
The University of Alberta programmers changed things up for this session. In Sessions One and Two only one version of Polaris was playing. Sessions Three and Four used 10 different "bots," or versions of Polaris, that each played with a different style. Governed by a "coach" program that would alternate the bots into the player's seat, Polaris became a competitor that was constantly changing gears and learning which tactics worked best against its opponents.
Faced with a learning, gear-changing robot from hell Eslami took a beating and ended the session at -$635. Laak, however, was able to exploit some weaknesses in some of the bots and ended the session with $1,455 giving the human team a balance of $820.
This final session proved a dramatic one. Eslami was on the hot seat playing in front of the audience and after 175 hands was down to -$680. But at the end of the full 500 hands Eslami had ground his way up to $460.
So it was to much rejoicing that Laak returned from his seclusion to say that he had finished with $110.
And so it seemed a solid victory for team humanity. Was this more proof that the complexity of poker was still currently too much, even for a program that had been in the works for 16 years?
Not according to Eslami and Laak.
As the applause died down Eslami spoke to the crowd, "This was not a win for us. First of all there are a few things you need to know. One of the bots completely clobbered us. Another one had kind of a glitch in the second match that we won."
Both players also agreed that they had played their absolute best poker and if there had been a time limit on the hands, they would not have been able to beat Polaris.
Undoubtedly the tenacious team of A.I. scientists at the University of Alberta will be back at it for a rematch in the near future as they get under the hood of Polaris and try to teach it to get under the hood of Phil "The Unabomber" Laak.