A computer can now beat anyone at Texas hold'em poker

Poker face will only get you so far.

A computer program can now beat anybody at the game of two-player limit Texas hold'em poker, researchers in Canada reported Thursday in the US journal Science.

Lead author Michael Bowling of the University of Alberta told AFP the team's goal since 2003 has been to "produce a perfect player. One that doesn't lose to the current top humans. But also doesn't lose to any strategy."

In 2008 the team created the first poker-playing program, called Polaris, that could defeat top poker pros at heads-up limit hold'em poker.

"At that point, computers established themselves as stronger at this form of poker than all humans," Bowling said.

After discovering that Polaris could beat people at poker, the "next logical step was to consider if (the game) could be solved," he added.

A cluster of 4,800 computer central processing units (CPUs) began the computations to solve the game about eight months ago.

"It took just over two months of computing to reach the goal of essentially solving the game," Bowling said.

The result is that the most popular version of poker played around the world is now "essentially weakly solved," meaning a lifetime of human play could not beat it by any measure of statistical significance.

"While we sought to compute a perfect strategy that solves the game, our strategy only essentially solves the game, which means it is so close to perfect that the amount that it can be made to lose is so small that it would be indistinguishable from luck even after playing 60 million hands of poker (nearly a lifetime of a human playing the game)," Bowling explained in an email.

- Computer vs man -

Other computers that have beat people at games include the checkers-playing computer Chinook which in 1994 became the first program to win a human world championship and Deep Blue which beat Garry Kasparov at chess in 1997.

The IBM computer Watson also won on Jeopardy in 2011.

But Texas hold'em poker proved a particular challenge because with only two players, there is plenty of unknown information, like which cards have already been dealt to the opponent.

As such, poker is one of many so-called "imperfect information" games, and is the biggest of its kind to be conquered by artificial intelligence, wrote computer scientist Tuomas Sandholm of Carnegie Mellon University, in an accompanying editorial in Science.

"This is, to my knowledge, the largest imperfect-information game essentially solved to date, and the first one competitively played by humans that has now been essentially solved," he said.

Sandholm explained the general approach for solving these so-called "imperfect-information games" as first abstracting the game "to generate a smaller but strategically similar game, reducing it to a size that can be tackled with an equilibrium finding algorithm."

The abstract games are solved for equilibrium or near-equilibrium, and then the strategies are mapped back onto the original game.

The challenge of solving the game may also help lead to advances for modern society.

Game theory can help boost security systems at airport checkpoints, improve coast guard patrols, and make better medical decisions, said Bowling.

Algorithmic advances like the one used for poker may help advance solutions in real-life decision-making settings that involve uncertainty and missing information, concluded the paper.

"However, we also echo a response attributed to (pioneering British computer scientist Alan) Turing in defense of his own work in games: 'It would be disingenuous of us to disguise the fact that the principal motive which prompted the work was the sheer fun of the thing.'"

ksh/jm