Men vs Poker Bot Round 2

Last summer saw some students from the University of Alberta in Edmonton, Alberta, Canada take a computer trained to play Head’s Up Limit Hold’em and play against professional poker players. Last year the Polaris program took on Phil Laak & Ali Eslami, in Vancouver, British Columbia, Canada. While the Pro’s won last year this year the format was changed up a little. Instead of just two competitors the Polaris program had to face 7 different human challengers from the Stoxpoker poker coaching website. The humans were Nick Grudzien, Kyle Hendon, Rich McRoberts, Victor Acosta, Mark Newhouse, IJay Palansky, and Matt Hawrilenko.

In a duplicate poker format in which two matches occurred simultaneously, a match was declared a draw if either the machine or the human won less then 25 small bets, or 25 small blinds. The way it would work is Polaris would be dealt the same hole cards as Human B in one match and the same as Human A in the other. In duplicate poker all players see the same community cards thus removing the variance/luck from the match and giving a clear indication of talent. Each match will consist of 500 hands. With blinds of 500/100 and betting limits of 1,000/2,000 to score a win either the Human or Polaris would have to amass a profit, or limit a loss depending on how you want to look at it of 25,000.

In the first match Polaris A vs. Human B (Kyle Hendon) & Human A (Nick Grudzien) vs. Polaris B. After the 500 hands had concluded, both Kyle Hendon and Polaris B came out on top. Kyle Hendon profited 37,000 while Polaris B profited 42,000. Therefore, while Polaris did better by profiting a net of 5,000 that is less then the 25k required and match #1 was scored a draw.

The second match pitted Polaris A vs. Victor Acosta (Human B) and Rich McRoberts (Human A) vs. Polaris B. With 500 hands in the books the humans scored the first win of the series, as Rich profited 89,500 while Acosta lost only 39,500. Thus giving the Humans a net difference of 50,000 and leaves the score sitting at 1-0-1 for the humans!

However, not to be counted out Polaris made a strong come back by winning the third match in which it was pitted against Mark Newhouse (Human B) and IJay Palansky (Human A). After all the hands were dealt Mark had amassed a very formidable 251,500. Unfortunately, for the humans though Polaris B took IJay for 307,500 giving the Polaris program a profit line of 56,000. In this match the cards were definitely tilted in favor of Polaris A and Mark Newhouse! Consequently, after three matches the record of 1-1-1 leaves us in a dead heat.

In match #4 Polaris dominated by winning both of it matches in a duplicate format which is very impressive. In the match between Polaris A and Mark Hawrilenko (Human B), Polaris came away with a profit line of 60,500. In the match between Polaris B and IJay Palansky (Human A), Polaris B recorded a profit line again this time 29,000. As a result, Mark lost 60,500 while Polaris B with the same cards was able to win 29,000. This gave Polaris a record of 2-1-1 at the Gaming Life Expo, at the Rio All-Suite Hotel and Casino July 3rd – 6th 2008.

In two earlier matches conducted else where against the Stoxpoker team, that had been decided in advance to count in combination with the Gaming Life Expo results. The Humans won their second match but ended up splitting the two matches with Polaris, as the computer won the second. The final tally ended up being 3-2-1 giving Polaris revenge from the competition in Vancouver.

In discussions with Professor Michael Bowling of the University of Alberta, who is the supervising supervisor for the graduate students, which were in command of programming Polaris 2.0 and improving it over the previous program. The biggest difference was they were able to add an element of learning, in which the program could identify the poker strategy being used against it and react by changing its own strategy accordingly. This also prevented the humans from talking with each other and trying to adapt to the Polaris program. The other major difference was that Polaris 2.0 could adapt by learning from the mistakes it was making in certain situations and applying what it has learned to future similar situations.

All in all it was an interesting day for the Man vs. Machine debate. As in the world of poker which is very non-linear, for a machine to win shows that computer’s have come a long way. And brings up the debate where will it end and should it continue to evolve? The Students of the Canadian Poker Research Group at the University of Alberta say they will continue to work on the program to not only improve it, but to also teach it play and compete in other forms of poker.