Cepheus has already played millions of hands of hold 'em poker
It learned from its mistakes and now can beat any human
Scientists say it's a breakthrough in artificial intelligence that might have applications in other areas
Poker players like to read another player’s tell.
That’s the thing an opponent does that gives away the quality of cards he is holding.
Well, you can’t read Cepheus, and you can’t beat it.
You wanna take on the card king? Have at it. You may win a few hands, but in the end, you better know when to walk away.
Cepheus has already played millions of hands of heads-up limit Texas hold ‘em, a poker game between two players involving two held cards and three shared.
That would take a lifetime or two for you or me, but Cepheus is a computer program, set up by researchers at the University of Alberta.
Michael Bowling’s team taught Cepheus by having it play itself, and the program learned from its mistakes, according to the team’s website.
“With each hand it improved its play, refining itself closer and closer to the perfect solution,” the website says.
All it took was 4,000 computers doing the math on 6 billion hands every second for several months.
And voila, game solved.
“We define a game to be essentially solved if a lifetime of play is unable to statistically differentiate it from being solved at 95% confidence,” Bowling said in a news release announcing the findings, which were published in the journal Science. “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 exploitative opponent strategy, and never making a mistake.”
In other words, Cepheus, over the long run, will always win, at least against humans.
The researchers say it’s a breakthrough in artificial intelligence and game theory, which can have applications in other areas. Programmers potentially could use it in computer systems for airport security checkpoints or Coast Guard patrols.
“With real-life decision-making settings almost always involving uncertainty and missing information, algorithmic advances, such as those needed to solve poker, are needed to drive future applications,” the news release said.
CNN’s Chandler Friedman and Matthew Mochow contributed to this report.