Korean Go master Lee Se-dol has proved human intuition can still surpass the ever-evolving artificial intelligence technology, with his first victory against Google’s AI program AlphaGo.
“This single win is so valuable and I would never exchange this with anything in the world,” Lee said after the victory on Sunday in Seoul.
Demis Hassabis, CEO of DeepMind, the AlphaGo developer, also offered his respect to Lee’s victory. He said, “We are very happy about Lee’s victory. This is why we came here to test AlphaGo’s limit and improve the program.”
Lee Se-dol (Yonhap)
Strictly speaking, Lee is not the best Go player today. China’s teenage prodigy Ke Jie, 19, is the highest-ranked player currently, followed by Korea’s 23-year-old Park Jeong-hwan and Japan’s Yuta Iyama, 27.
Lee, the No. 4, is also more than 10 years older than the top players. Like in other sports games, younger players in their 10s or 20s show their best performance in Go as well.
Then, why did Google choose Lee to play with AlphaGo?
Data should be one of the key factors. Lee who turned pro at the age of 12 has more than 20 years of experience, which means he has more old matches for AlphaGo to study and learn from.
Another factor is his personality. He is known for his aggressive and unconventional style of play. And unlike other Go players who were skeptical about playing against an AI, he showed willingness to try.
“I accepted Google’s offer in three minutes. I always had curiosity about AI’s Go skills so I thought playing with it was the best way to solve my curiosity,” he said.
He also never hid his confidence in the upcoming five-game match in Seoul and said, “I do feel pressure, a pressure that I should not lose even one game. Computers may exceed humans someday but that is not this time.”
But after his first defeat on Wednesday, he seemed quite confused. Even though it was a tug-of-war overall, he made mistakes at the last minute and AlphaGo seemed never disturbed.
“I was really surprised. I didn’t think AlphaGo would play in such a perfect manner,” he said.
And the second match was no difference.
But in the third game, he changed strategy. Even though he was defeated again, he tried diverse experiments in a humble manner to learn more about his opponent.
Experts say this third match paved the way for his first victory in the fourth game.
Amid a close game on Sunday, Lee attacked AlphaGo’s territory on the center zone of the board, but AlphaGo didn’t make defending moves, putting some stones in inexplicable places in other parts of the board.
Lee didn’t miss the opportunity to take advantage of the opponent’s mistakes. He continued attacking the center zone and secured more territory.
Even after that, AlphaGo made some clumsy moves until it finally admitted its first defeat to Lee.
“AlphaGo seems less trained in losing situations. Lee has won the match with his own aggressive and unconventional style of play,” said a commentator at the end of the match.
With the latest victory, Lee became the first professional Go player to win against AlphaGo. In October, the AI took a 5-0 victory against Fan Hui, the first professional Go player to play with it.
Apart from the records of the five matches, Google has not revealed any information about AlphaGo’s old matches. The AI has honed itself by studying old matches and training through simulated games.
Lee and AlphaGo have only one more match planned on Tuesday. Even if Lee wins again, the $1 million prize will be given to AlphaGo, which has already won three matches.
By Lee Ji-yoon (
jylee@heraldcorp.com)