AI has claimed some sizable scalps in recent years, with IBM's Watson toppling humans at Jeopardy and Google's AlphaGo program taking down a professional Go player last year. And researchers are hailing its latest triumph as yet another landmark moment for computer science, with Carnegie Mellon University's Libratus program cleaning up at a 20-day poker tournament against professional human counterparts.

The Chinese game of Go was considered by many to be the Mt. Everest of artificial intelligence. Where computers mastered simpler games like chess and checkers long ago, Go was a different ball game, with the massive amount of possible positions offered by the grid-patterned board amounting to more atoms than there are in the universe.

Sure, this made developing an algorithm that could consider all the possibilities a huge challenge, but one thing AlphaGo had in its favor was that it could at least see the entire board. In this sense, poker presented another kind of challenge in that it is impossible to know what cards another player is holding, so the program needed to account for and even indulge in misinformation.

"The computer can't win at poker if it can't bluff," said Frank Pfenning, head of the Computer Science Department in Carnegie Mellon's School of Computer Science said. "Developing an AI that can do that successfully is a tremendous step forward scientifically and has numerous applications. Imagine that your smartphone will someday be able to negotiate the best price on a new car for you. That's just the beginning."

Over the course of 120,000 hands of No Limit Texas Hold'em, Libratus went head to head with poker professionals Dong Kim, Jimmy Chou, Daniel McAulay and Jason Les at the same time. After 20 days of bluffing and folding, the computer found itself ahead of the pros by no small margin – a total of $1,766,250 in chips across the four games. One of the features of Libratus' play that came as a surprise to those in attendance was the ability to improve from day to day.

"After play ended each day, a meta-algorithm analyzed what holes the pros had identified and exploited in Libratus' strategy," said Tuomas Sandholm, professor of computer science and developer of the program. "It then prioritized the holes and algorithmically patched the top three using the supercomputer each night. This is very different than how learning has been used in the past in poker. Typically researchers develop algorithms that try to exploit the opponent's weaknesses. In contrast, here the daily improvement is about algorithmically fixing holes in our own strategy."

This kind of artificial intelligence isn't about to go mainstream. Libratus is powered by Pittsburgh Supercomputing Center's Bridges computer. This piece of kit has a total speed of 1.35 petaflops and 274 terabytes of memory, around 7,250 times as fast and 17,500 times as much memory as you'll get in a high-end laptop.

But still, the possibilities are exciting for areas where machines deal with incomplete information and misinformation. The smartphone-car purchase example offered above might be a ways off, but the technology could conceivably find use in business negotiations, military strategy, cybersecurity and medical treatment.

And the researchers will be openly sharing the secrets behind the Libratus' success, starting with talks at the Association for the Advancement of Artificial Intelligence between February 4 and 9 in San Francisco and then by looking to publish in peer-reviewed journals. You can hear some mid-tournament comments from the human poker players in the video below.

View gallery - 4 images