After three consecutive losses, one of the world’s best Go players has finally beaten Google’s DeepMind AI at the ancient Chinese boardgame.
Lee Sedol is in the middle of a five-game series of Go games with AlphaGo – the AI program written by Google’s UK-based DeepMind AI team.
The 8-time Go world champion from South Korea had expected to sweep home to victory in comprehensive fashion, as no computer program had been able to stand up to a top-level Go player before. He lost the first three games to Google’s extraordinary AI.
Lee may have lost the series, but he has retained some pride by winning the fourth game.
AlphaGo’s first loss seemed to stem from a major error of judgement, as DeepMind founder and CEO Demis Hassabis explained.
This error eventually resulted in the AI resigning after five hours of play, to the human victor’s evident delight.
“I’ve never been congratulated so much just because I won one game,” Lee told press after his win. “This one win is so valuable and I will not trade this for anything in the world.”
As for the defeated DeepMind team, this loss is far from a negative thing. “For us this loss is very valuable. We’re not sure what happened yet,” said Hassabis (via The Verge).
You can watch the fourth Go match here:
Lee believes that he has figured out a couple of AlphaGo’s weaknesses, one of which could be the inability to handle surprises. He said that the AI’s error came about when he himself played an unorthodox move.
Related: What is DeepMind?
The final match in the series will take place on Tuesday, March 15, when we’ll get to see if Lee Sedol really has figured out an effective strategy against AlphaGo.
DeepMind’s apparent Go mastery had once been seen as out of reach for a computer program, owing to the huge number of potential moves and scenarios in each game.