What is it? A self-taught chess program
Innovator: Google’s DeepMind division
What were they thinking? Computers have been spanking chess champions since 1997, when IBM’s Deep Blue bested Garry Kasparov. But the leading programs have had a lot of human help developing their approaches to the game. Google’s DeepMind tried something different, giving its AlphaGo Zero artificial intelligence program the basic rules of chess and letting it learn on its own.
Did it work? Just four hours after getting the rules, AlphaGo Zero thumped the globe’s best specialized chess software, Stockfish 8, in a 100-game match up — winning or drawing every game. “I always wondered how it would be if a superior species landed on earth and showed us how they played chess,” chess grandmaster Peter Heine Nielsen told the BBC. “Now I know.” No word, yet, on when AlphaGo Zero begins colonization.