Although Google’s AI algorithm, AlphaGo, has already won the best-of-five series against human player Lee Sedol at the game of Go, both parties decided not to enact a mercy rule and kept playing. In the fourth game in the closely-watched series being held in South Korea, Lee finally struck a blow for humanity and beat AlphaGo.
While Google’s program has so far performed perfectly against Lee as well as Fan Hui—the European Go champion it took on in October—it seems that it does have weak spots. According to The Verge, AlphaGo made a mistake on its 79th move in Sunday’s game, and didn’t realize until its 87th move. Others suggest, however, that Lee managed to win the fourth game partially because he no longer had to worry about winning the series, and so played more freely than he had in the other games. Lee employed a different strategy than in other games, forcing AlphaGo to take him on directly, which it responded to with a series of “strange” moves, according to The Korea Times. Lee won when AlphaGo resigned from the match after 180 moves.
“Lee was too strong for AlphaGo today,” Demis Hassabis, CEO of the Google subsidiary DeepMind that created AlphaGo, told The Korea Times. “It estimated itself that it was doing well before the middle of the game but Lee’s incredible plays pressured it into [making] some mistakes.” Lee had previously predicted that he’d win the series 5-0 or 4-1.
The game of Go has long been considered the holy grail of artificial intelligence research. AlphaGo strategizes using neural networks, computer programs that can learn how to complete a task after being fed reams of data. Much like humans who have already won a game and have to keep playing, perhaps AlphaGo just took its foot off the pedal a bit for the fourth game. Or perhaps AlphaGo is just toying with Lee, and all humanity, as it starts its inevitable destruction of the human race.
The final game, and perhaps humanity’s last moments of freedom, is scheduled for Tuesday.