Robots Are Getting Closer To Mastering Poker
For years, Carnegie Mellon has been on a mission to beat poker's best players with their AI program. In 2015, they pitted the AI program "Claudico" against four pro poker players, but it fell short by $732,713. Their study attributed the loss on the fact that poker is an imperfect information game, and that "increases the complexity and the uncertainty in calculating the best course of action—to raise, to fold, or to call." But Carnegie Mellon's Tuomas Sandholm and his PhD student Noam Brown don't take defeat well. In 2017, they set out to fix this issue with a new AI program, "Libratus."
Related: One Good Way To Research AI? Build Bots That Play Video Games
The new card-slinging robot changed its behavior each day at Sandholm's Brains Vs. AI event at Rivers Casino in Pittsburgh, honing its strategy with computer calculations. True to the strengths of artificial intelligence, the program learns from the moves of its opponents and from its own mistakes, so it gets better and better the longer it plays. The current best poker players in the world were amazed at how quickly the bot learned from their techniques. "...every time we find a weakness, it learns from us and the weakness disappears the next day," poker pro Jimmy Chou told Gizmodo. The practice certainly paid off, because on Monday, January 30, the AI finally won.
Related: AI Is Changing The Future Of Driving
from Poker
Post a Comment