Google's artificial intelligence technology beat a top Go player last year. Now it's taking on a whole team.
DeepMind, an AI research group within Google's parent company Alphabet, is bringing out its AlphaGo software to play the ancient board game Go in China this week. Unlike its previous matches against individuals, AlphaGo will go up against a team consisting of five of the world's top-ranked Go players, according to a Google events page.
Prior to taking on the team, AlphaGo will play one on one against Go's current world champion, Ke Jie. AlphaGo hasn't had any practice in public since March 2016, when it succeeded in beating another one of the world's best Go players, Korea's Lee Sedol. AlphaGo did play a few games a few months ago without unveiling its identity until afterward. In one of those games, it happened to beat Jie.
Last year's victory over Sedol was considered a major achievement in AI. Go is harder than chess. There are more potential moves for a player to make at any given turn, so predicting the best next move is more complicated.
In 1997 IBM made headlines when its Deep Blue supercomputer beat world champion Garry Kasparov at chess, the first time a computer defeated a champion. But AlphaGo is more powerful than Deep Blue. Rather than being developed with carefully scripted rules, AlphaGo relies in part on reinforcement learning, a technique that makes systems smarter over time through trial and error.