Robots Learning to Play Poker (for Improved Cybersecurity)

The robots are winning. IBM’s Watson has beat down competitors in Jeopardy and Deep Blue famously won a chess match against Garry Kasparov, a grandmaster. Tanya Lewis from LiveScience writes that developers may be able to add poker to the list.

It’s not entirely inconceivable. As Jon Iwata, senior VP of Marketing and Communications at IBM, explained in his interview with Big Think, the wonder of Watson is that it can learn — it takes mounds of information and organizes it.

Just this month, an AI system called Claudico faced off against some of the world’s best poker players in over 80,000 hands of heads-up no-limit Texas hold ’em at Rivers Casino in Pittsburgh. But Claudico did not pass the threshold to be considered scientifically valid.

However, Claudico’s lead creator, Tuomas Sandholm, a computer scientist at  Carnegie Mellon University, thinks we’re close to getting a poker-playing robot. He said to Lewis:

“I am guessing [a poker-playing AI] will be stronger than the best humans in the world in one to five years.”

You may be wondering how the developers factored in bluffing. Sandholm explained:

“People often think about bluffing as being a psychological phenomenon.”

However, he says, “Bluffing still emerges as a strategic phenomenon.”

See the Nash equilibrium. Sandholm and his team programmed Claudico with algorithms to find just this. It’s a concept in game theory that players in a non-cooperative game scenario will make the best decision possible based on the decisions of other players. But Sandholm and his team aren’t looking to just create a great poker-playing robot. He says, “Poker is a great benchmark” for what Claudico’s true purpose will be — as a way to solve problems with incomplete information, such as in cybersecurity.