We already know that the best human players have met their match at games like chess and Go, backgammon and checkers. But, The Verge writes today, the actual applications of those AI accomplishments are hard to pinpoint because "most real-world problems aren't laid out neatly on a board where both sides know exactly how the opponent will operate and where all the important pieces are at all times." Learning how to out-bluff humans, though, could have a lot of real-world uses, and that's why researchers at Carnegie-Mellon built Libratus, a bot that for the last couple of weeks has been engaged in many tens of thousands of hands of poker against a group of pros.
As The Verge notes, Libratus is kicking butt, exhausting its flesh-and-blood opponents to the tune of a $750,000 lead by bluffing them into the ground, often befuddling the pros with unpredictable bets that make no sense to them, forcing them to fold hands, but which slowly build the AI's winnings. It's also simply wearing them down over the course of many 10-hour days. "The bot gets better and better every day. It's like a tougher version of us," Jimmy Chou, one of the pros playing Libratus told The Verge. "The first couple of days, we had high hopes. But every time we find a weakness, it learns from us and the weakness disappears the next day."