Artificial intelligence has wiped the floor with humans in games such as checkers, chess, Jeopardy! and, in perhaps its finest hour so far, beating flesh, blood and brain cells at the ancient Chinese board game of Go.

But strategy games such as those cannot be compared to poker, a game that while asking players to employ logical strategies based on winning formulas also requires a certain amount of cunning and bluffing. AI has gone up against human poker players in the past and gone home empty-handed.

The reason is that the computer has to deal with incomplete information, and so it must adapt, assess a situation given its competitors’ sometimes random behavior and the mathematical odds of the game. It must know when to fold ’em, while never knowing what the others are holding. At the same time, a human player’s betting strategy is never easy to understand. Sometimes it must be random in order to perplex the other players. All that is why Andrew Ng, chief scientist at Baidu, has called poker “one of the hardest games for AI to crack.”

Yet a machine is doing just that right now at the 20-day Heads-Up, No-Limit Texas Hold’em tournament at Rivers Casino in Pittsburgh. The AI, named Libratus by its creators at Carnegie Mellon University’s School of Computer Science, is no ordinary player.

Its creators explain that “Libratus’s strategy is not based on the experience of expert human players, so its game play could differ markedly from the pros.” Its algorithms take the rules of poker and then it sets its own strategy based on many millions of hours of computation. It should also be mentioned that Libratus has been programmed only for two-player or “head-on” poker.

And it’s winning.

So far it’s reported that Libratus is sitting pretty with almost a million bucks in the bank. One the pros playing against Libratus, Jimmy Chou, said he underestimated the AI, and now says it’s getting “better and better every day. It’s like a tougher version of us.”

There has been some speculation that Libratus is having its behavior altered on occasions by those that developed it. Still, Chou explained that playing the AI has been frustrating. “Every time we find a weakness,” he said, “it learns from us and the weakness disappears the next day.”

Another competitor, American-Korean,  Dong Kim, has said it feels like the AI is cheating, as if it could see his cards, as it was “just that good.” Libratus is totally unpredictable, said Kim, bluffing liberally, changing its strategy daily. This has led some of its foes in Pittsburgh to ask if Libratus’ creators might be tinkering with it when the table is closed for the night.

Either way, when an AI is beating humans in such a complex game, it’s clear the range of what makes humans unique is narrowing.

Photo credit: Arnaud Fraioli via Flickr