<
>

Representing mankind

When the news came that Phil Laak and Ali Eslami would be the two players representing mankind in the match against Polaris, probably the best poker-playing computer program developed at the University of Alberta and supported by the Poker Academy platform, I questioned the selections. Hell, the poker world questioned the selections.

Laak, the same guy known as the "Unabomber," runs around the floor screaming incoherent babble, doing push-ups and sit-ups. Eslami, meanwhile, is unknown to 99.7 percent of the poker community. Thing is, poker being the intensely private game it is, there's a lot we don't see on television. It's that stuff, discovered in my interviews with Laak, Eslami and Alberta professors Darse Billings and Jonathan Schaeffer, that may have made them the two most-qualified men on the planet for the job.

The university needed players who not only showed excellence in the game, but also were capable of speaking through their thought process to a live audience, while at the same time understanding the technology in a way that would optimize play and feedback. Laak played Polaris the last time the university set up a match in late 2005. When the plan was hatched to have two pros play both sides of the same match, he recruited Eslami, a successful cash-game pro for years, who'd shown an intense interest in the project. In Eslami, there was the added bonus of a man whose aggressive game ran counter to Laak's more conservative style.

"Phil and Ali played incredible poker," Schaeffer said. "You have to give it to them; they didn't do it for the money and there was a lot of pressure on them. These guys didn't bat an eye. They were great sports and perfect gentlemen."

Billings was the mastermind behind the project.

"Darse walked in here 15 years ago," Schaeffer recalled of the project's beginnings, "and said, 'Chess is easy. Let's try poker.' The whole program really owes itself to Darse for his computer and poker expertise."

Billings understood the root of the poker problem: You need to learn. However, at some points, whoever guesses better is going to win.

Thus, the problem: How do you make a machine … guess?

The answer is still being worked on, using a combination of "equilibrium bots" and "learning bots." Equilibrium bots play defensively by nature. They use a series of predetermined moves based on formulaic mathematics that essentially ignore the opponent. Learning bots, meanwhile, use the information presented to them to overcome and adapt. As of yet, however, poker-playing learning bots can't adjust as quickly as their human counterparts.

Laak and Eslami made their way to Alberta for "The First Man-Machine Poker Championship" almost immediately after the end of the World Series. The 2,000-hand match between humans and computer was played over four sessions on consecutive days. That's 500 hands per session of live poker, going through the thought process behind each hand for an audience that may or may not have understood the game it was watching. That's a draining process for even the most willful minds in the world.

The match began with Eslami sitting on a raised platform with a microphone nearby, as the audience of mostly scholars and a few poker aficionados watched. Every move was projected onto a massive screen that sat directly behind Eslami. Laak, meanwhile was playing the exact same 500 hands -- just at different times, so the two players never played the same hand simultaneously. By having it set up this way, a good portion of the luck factor was nullified.

Each move the computer made happened in a nanosecond.

"It was incredibly intimidating," Laak said. "It made its moves faster than your moves registered on screen, so by the time the display showed what you'd done, the computer had already made its play. It felt like it knew what you were going to do before you did it."

Eslami won his first round to the tune of $395, but Laak was losing $465 in the same hands. The rules of the contest considered that net loss of $70 a draw, but the pros knew better. In the second session, Polaris made sure there wouldn't be any technicalities: It lost $1,570 to Laak, but at the same time decimated Eslami to the tune of $2,495, for a $925 net victory. More important, it held the upper hand emotionally. As good as the cards had been to Laak, they'd been bad to Ali. His posture suggested he was a defeated man.

After those first 1,000 hands, the humans regrouped.

"They took the hand logs and discussed them for hours," Billings said. He acknowledged it was a major advantage they wouldn't have against other human opponents, but it was part of the process by which the programmers would learn.

The second round of 1,000 hands took on a distinctly different flavor. Armed with a new understanding of how the computer was playing and adjusting, the Laak-Ali team took its first victory in the third session: Phil won by $1,455 while Eslami lost just $635.

This was Eslami's shining moment. The day before, he'd been frustrated into submission. Now he slowed down and deliberated every move carefully. With no clock and no opponent to pressure him, he agonized over the minutia. The careful nature of his play, tempered by a greater understanding of the technology than any lesser-prepared opponent could be expected to have, was the deciding factor. The humans won again in the fourth and final session, this time by $570. They finished with a 2-1-1 record in the sessions and $395 in overall winnings.

As any experienced player on a bad run will tell you, 2,000 hands isn't enough to eliminate the variance factor from the results, but even that small sample told us Polaris could compete. Despite the victory for flesh and blood, the professors were pleased with the results.

"I think the results of this match were going to be a win-win scenario," Schaeffer said. "We showed the world that the project is advancing and it was quite clear the program was competitive."

That's despite a flaw that may have cost the professors the final leg. An algorithm chose the three most optimal computer "player" programs to mix and match against Phil and Ali in the finale, but when Phil's play early on determined which of the three was least efficient, that "player" was excluded from the remainder of the match. That made the program's play much easier for Laak to decipher. The flaw wasn't the only thing working in the humans' favor. Phil and Ali were able to confer with one another on their mutual opponent's play. The computer played alone.

In all of the excitement over who won and who lost and how, Eslami seemed to hold the clearest vision of what was important here.

"Here's what's more exciting," he said. "If you can figure out a human being's next move, instead of opposing it, you can cooperate with it. I'm not talking about poker here, I'm talking about life. Anticipating a person's needs can help a computer augment them. This research could lead to computers acting as a personal concierge … a computerized personal assistant. Forget online poker, what we're doing here could help society at large."

"It was exhausting and exhilarating," Laak concluded. "Like winning $500,000 … like winning a World Series bracelet. I think right now, it's like humans 56 percent, Polaris 44 percent, but it could definitely compete with mediocre players. Give it a year."

Asked if he'll be back, Laak answered, "They're going to keep inviting me back until they beat me. Then we're bringing in the uber-geniuses."

Who are they? Phil says "14-year-olds" who've been doing the math since prepuberty. I'm guessing Chris Ferguson, Michael Binger, Brian Townsend … maybe even another Phil Ivey.

Laak and Eslami showed charisma, expertise, humility and guts throughout the process. They put their reputations on the line in the name of something bigger than themselves and performed the tasks put in front of them and then some. That, not the bottom line, is how one represents humanity.

"We owe them a lot," Schaeffer said. "Enormous respect for two guys who put themselves on the line."

Billings added: "We're looking forward to the rematch."

So are we.