Skip to main content

Former Go champion beaten by DeepMind retires after declaring AI invincible

Former Go champion beaten by DeepMind retires after declaring AI invincible

/

‘Even if I become the number one, there is an entity that cannot be defeated’

Share this story

Professional ‘Go’ Player Lee Se-dol Plays Google’s AlphaGo - Last Day
Lee Se-dol is seen in 2016 during his matches with the AI program AlphaGo.
Photo: Google / Getty Images

The South Korean Go champion Lee Se-dol has retired from professional play, telling Yonhap news agency that his decision was motivated by the ascendancy of AI.

“With the debut of AI in Go games, I’ve realized that I’m not at the top even if I become the number one through frantic efforts,” Lee told Yonhap. “Even if I become the number one, there is an entity that cannot be defeated.”

Lee lost 4-1 to DeepMind’s AlphaGo in 2016

For years, Go was considered beyond the reach of even the most sophisticated computer programs. The ancient board game is famously complex, with more possible configurations for pieces than atoms in the observable universe.

This reputation took a knock in 2016 when the Google-owned artificial intelligence company DeepMind shocked the world by defeating Se-dol four matches to one with its AlphaGo AI system. The games had a global impact, alerting the world to a new breed of machine learning programs that promised to be smarter and more creative than AI of old.

Lee, who was the world’s number one ranked Go player in the late 2000s, initially predicted that he would beat AlphaGo in a “landslide” and was shocked by his losses, going so far as to apologize to the South Korean public. “I failed,” he said after the tournament. “I feel sorry that the match is over and it ended like this. I wanted it to end well.”

Despite the outcome, Go experts agreed that the tournament produced outstanding play. AlphaGo surprised the world with its so-called “move 37,” which human experts initially thought was a mistake, but which proved decisive in game two. Lee made his own impact with his “hand of God” play (move 78), which flummoxed the AI program and allowed Lee to win a single game. He remains the only human to ever defeat AlphaGo in tournament settings. (During training AlphaGo lost two time-capped games to Go player Fan Hui.)

Since the tournament, though, DeepMind has only improved its AI Go systems. In 2017, it created AlphaGo Zero, a version of the program which surpassed even AlphaGo.

AlphaGo has since been surpassed by its successor, AlphaGo Zero

While the original AI learned to play Go by studying a dataset of more than 100,000 human games, AlphaGo Zero developed its skills by simply playing itself, over and over. After three days of self-play using hugely powerful computer systems that let it play games at superhuman speeds, AlphaGo Zero was able to defeat its predecessor 100 games to nil. DeepMind said at the time that AlphaGo Zero was likely the strongest Go player in history.

In a statement given to The Verge, DeepMind’s CEO Demis Hassabis said Lee had demonstrated “true warrior spirit” in his games with AlphaGo. Said Hassabis: “On behalf of the whole AlphaGo team at DeepMind, I’d like to congratulate Lee Se-dol for his legendary decade at the top of the game, and wish him the very best for the future ... I know Lee will be remembered as one of the greatest Go players of his generation”

According to Yonhap, Lee isn’t completely giving up on playing AI, though. He plans to commemorate his retirement in December by playing a match against a South Korean AI program called HanDol, which has already beaten the country’s top five players. Lee will be given a two-stone advantage.

“Even with a two-stone advantage, I feel like I will lose the first game to HanDol,” Lee told Yonhap. “These days, I don’t follow Go news. I wanted to play comfortably against HanDol as I have already retired, though I will do my best.”

Update: Comment from DeepMind CEO Demis Hassabis has been added to the story as well as clarification about players who have beaten AlphaGo in different settings.