Why Lee Sedol Finally Beat Google’s AlphaGo at Go

Before last night’s fourth game of a legendary Go match between an AI and a human champ, it seemed that the machine was unbeatable.. In fact, the computer wasn’t as swift as human player Lee Sedol. AlphaGo is a program designed by DeepMind and acquired by Google to win at the intuitive, ancient Chinese game. And man, did it whoop European champion Fan Hui in a five game match in January. But finally, the human player Lee triumphed.

The Korean player felt that there had been some missed opportunities and pressure that caused him to lose the prior two games. This time though, it seemed that maybe AlphaGo reached its capacity for winning. It seemed that a surprise move from Lee confused the computer. Reporting from scene at the Four Seasons in Seoul, The Verge noted that some of the moves by the AI looked more like mistakes than strategies.

DeepMind founder Demis Hassabis used Twitter to convey the reason AlphaGo lost. In particular, there was one poorly calculated, fateful move. “Lee Sedol is playing brilliantly! #AlphaGo thought it was doing well, but got confused on move 87. We are in trouble now…” He later corrected himself, noting that it was actually move 79, but AlphaGo didn’t notice the screw up until later. He tweeted that the computer doesn’t actually think like humans do, but rather calculated the output of the “value net.”

Lee won, proving that the AI is not flawless. The champion was celebrated with great enthusiasm after the game. Hassabis said, “For us this loss is very valuable. We’re not sure what happened yet.”

Watch the final game on Monday, March 14, 11 p.m. EST, or Tuesday, March 15 in Korea.