Libratus, the poker-playing AI that crushed four world-class pros in January, has put another group of human players to shame. This time, the upgraded variant of the AI known as "Lengpudashi" or "cold poker master" took on World Series veteran Alan Du and a team of engineers, computer scientists and investors. Instead of using pure poker skills to try and defeat Lengpudashi like the first team did, the new players applied what they know about machine learning to their game. Alas, their strategy didn't work, and the AI still won by a landslide after playing 36,000 hands against the team at a resort on China's Hainan island.
Unlike go, chess and other games AI play, you don't see your opponent's hand in poker. Plus, it has complex betting and bluffing techniques that present a completely different challenge. Libratus comes up with strategies by doing computations with the rules of the game in mind -- its creators at Carnegie Mellon University didn't feed it copious amounts of samples to learn from, which is one way to teach AIs new skills.
Libratus co-developer Noam Brown said that after playing and winning games against pros, its clear that people misunderstand what computers and humans are each good at:
"People think that bluffing is very human -- it turns out that's not true. A computer can learn from experience that if it has a weak hand and it bluffs, it can make more money."
He didn't say if they have more opponents lined up for the AI, but we have a feeling it's going to do just fine.