AI Plays a 'Perfect Game' of Texas Hold'em Poker

Artificial Intelligence (AI) is growing by leaps and bounds as technology improves. In fact, computer systems have been built that can beat human players in games. Yet defeating top human players isn't the same as actually solving a game. Now, though, scientists have solved heads-up limit hold'em poker.

"Poker has been a challenge problem for artificial intelligence going back over 40 years, and until now, heads-up limit Texas hole'em poker was unsolved," said Michael Bowling, lead author of the new study, in a news release.

Poker is part of a family of games that exhibit imperfect information, where players don't have full knowledge of past events. The most popular of these games is Texas hold'em. When it's played with just two players, called heads-up, and when it is fixed with bet-sizes and number of raises (limit), it's called heads-up limit hold'em. Yet the imperfect information of this game makes it a far more challenging game for computers to solve.

"We define a game to be essentially solved if a lifetime of play is unable to statistically differentiate it from being solved at 95 percent confidence," said Bowling. "Imagine someone playing 200 hands of poker an hour for 12 hours a day without missing a day for 70 years. Furthermore, imagine them employing the worst-case, maximally exploitive, opponent strategy, and never making a mistake."

While many perfect information games, such as Connect Four, have been solved, no nontrivial imperfect information game played competitively by humans has been solved. This is partly because these games are more challenging with theory, computer algorithms and instances of solved games lagging behind results in the perfect information setting.

Now, though, it seems that researchers have succeeded. With the help of general algorithmic advances, scientists have managed to essential solve heads-up limit hold'em poker.

This doesn't just have implications for games, either. With real-life decision-making settings almost always involving uncertainty and missing information, this could also have applications for future AI endeavors.

The findings are published in the journal Science.

For more great science stories and general news, please visit our sister site, Headlines and Global News (HNGN).