Grandmaster Lee Se-dol reveals how he beat AlphaGo: 'Every move was a trick'

Home > Business > Tech

Grandmaster Lee Se-dol reveals how he beat AlphaGo: 'Every move was a trick'

Published: 28 Aug. 2025, 07:00

Audio report: written by reporters, read by AI

Go grandmaster Lee Se-dol, who recently released a new memoir detailing his historic 2016 match against Google DeepMind's AlphaGo, poses for a photo during an interview with the JoongAng Ilbo at the JoongAng Ilbo's studio in Mapo District, western Seoul, on Aug. 14. [KWEN HYEK-JAE]

[INTERVIEW]

In 2016, when Google DeepMind’s AlphaGo stunned the world by defeating Korea’s Go champion Lee Se-dol 4–1, the ancient board game seemed to collapse overnight under the weight of AI. Humanity took consolation in Lee’s lone victory — “the divine move” in Game 4 — but the game’s foundations, built over 5,000 years, were suddenly shaken.

Novelist Chang Kang-myoung captured much of that upheaval in his nonfiction book published this June, “The Future That Arrived First: People Who Experienced the Age of AI,” for which he interviewed 29 professional players and six experts to chronicle the Go world after AlphaGo.

In the book, Chang described the Go world after the collapse of a philosophy and value system that had endured for 5,000 years as “the future that arrived first,” the titular phrase of his book.

But Chang’s work had a limitation. The voice of the person at the center of the upheaval was missing. Lee — “the only human to defeat artificial intelligence” — appeared only in indirect quotes.

Now, nine years after the legendary match against AlphaGo, Lee has finally broken his silence in a new memoir, “The Art of Reading Moves in Life: Breaking Through Life's Uncertainty with the Strategies of Go.”

But first, more on the match itself: From March 9 to 15, 2016, Lee played a five-game match against AlphaGo, in which the AI defeated the top-ranked human Go player by four games to one. Although the series marked the first defeat of a human champion by AI, most people cheered Lee's single victory for preserving humanity’s pride.

Following the match, Google upgraded AlphaGo. The version that played Lee, known as AlphaGo Lee, evolved into AlphaGo Master and then AlphaGo Zero. AlphaGo Zero did not study any human games; it was fed only the rules of Go and developed its own operating system.

Korean Go grandmaster Lee Se-dol competes against Google DeepMind's AlphaGo at Four Seasons Hotel Seoul in Jung District, central Seoul, on March 12, 2016. [GOOGLE]

After releasing AlphaGo Zero, Google exited the Go world. It repurposed the neural network technology used to challenge Go (and humanity) for protein research. Demis Hassabis, CEO of Google DeepMind and architect of the AlphaGo-Lee match, received the 2024 Nobel Prize in Chemistry for his contributions to protein structure research.

Lee, the opponent of AlphaGo in the historic match, was born in 1983 on Bigeum Island in Sinan County, South Jeolla. He learned Go from his father at age five, dropped out of middle school and trained under master Kweon Kab-yong. He became a professional in 1995 at age 12.

Lee’s career record stands at 1,324 wins and 576 losses, with a win rate just under 70 percent at 69.7 percent. He won 14 major world titles and 50 championships in total.

On Nov. 19, 2019, Lee announced his retirement from professional Go, saying, “I can no longer enjoy the game after AlphaGo.” He is currently a special appointed professor at Ulsan National Institute of Science and Technology.

Lee left no official written account of the historic match against AlphaGo. Occasionally, he gave interviews, but he never recorded a written review of the games.

After nine years, that long-delayed review has finally arrived, in “The Art of Reading Moves in Life.” In this new memoir published by Woongjin ThinkBig, Lee reveals, for the first time, his complex mindset during the match and corrects widely misunderstood points. As soon as the book was released, the JoongAng Ilbo sat down with Lee face-to-face to talk about “the future that came first” — one that he experienced ahead of everyone else.

The following interview has been edited for length and clarity.

The cover of Go grandmaster Lee Se-dol's new memoir, "The Art of Reading Moves in Life: Breaking Through Life's Uncertainty with the Strategies of Go" [WOONGJIN THINKBIG]

Q. It's been six years since you retired. How have you been?

A. I’ve met a lot of people. For 25 years, all I knew was Go, so I started meeting people in other fields. Meeting people led to lecturing at universities.

Who is Lee Se-dol now? “Former professional player” sounds like the past. Professor? Writer? You make Go content — how about “Go creator” in today’s terms?

Good question. What am I? I’m 42 this year. I’ll just say I’m meeting people and experiencing new things as I prepare for the second act of my life.

In your book, you define Go as an “abstract strategy game.” What do you mean?

Go is unlike other games. Other games have fixed pieces on the board, but a Go board starts empty. It begins in a state of abstraction. The possibilities are infinite. Within that abstract space, you devise strategies to fight your opponent. Go is unpopular because it’s difficult. It takes too long to learn. But it’s the greatest game humanity has ever created.

Go grandmaster Lee Se-dol poses for a photo during an interview with the JoongAng Ilbo at a venue in Mapo District, western Seoul on June 19. [JANG JIN-YOUNG]

Maybe that’s why AlphaGo chose Go. But why did it choose you?

I heard Google reviewed game records from the past 10 years and picked me. At the time, Ke Jie had better results, but he was Chinese, and they may have been reluctant to choose a Chinese player first.

Let’s revisit the AlphaGo match. At first, you underestimated it. Then what happened?

I reviewed game records from five months earlier, when AlphaGo played Fan Hui, a 2-dan [level] professional in Europe. It was amateur-level. I was complacent. I didn’t think AI could improve that much in five months.

You really didn’t suspect anything?

The day before the first game, Google CEO Eric Schmidt said something that stuck with me: “The development of this technology will not threaten humanity but help it greatly.” That’s something you say when you’ve won. I understood his confidence, but it wasn’t something to say before a match. It gave me a cold feeling.

You played five games with AlphaGo. Your strategy changed each time?

In Game 1, I tested AlphaGo. Within 10 moves, I realized it was strong. After that, I was crushed. In Game 2, I tried a calm response, but it didn’t work. In Game 3, I made a poor strategic choice and lost in the opening.

A poor strategic choice?

AlphaGo had a 50-second time limit per move. I thought AI would make mistakes early on since there are more possibilities when the board is empty. So I tried to force a win early. That was a mistake. AI was strongest in the opening. Humans rely on intuition at the beginning, but human intuition can’t compete with AI’s calculations.

Are you saying it wasn’t impossible to win?

People don’t realize this: AlphaGo made small and large errors in all five games. I just didn’t punish them correctly — except in Game 4.

The most surprising part of the book is your review of Game 4. The world called move 78 “the divine move.” Google calculated the odds of that move as 0.007 percent. But you pinpointed move 68 as the decisive move. Why was that?

After losing three times, I had figured out AlphaGo’s strengths and weaknesses. I couldn’t compete in the opening or in the endgame, which is purely calculation. In Game 4, I tried to force an error as the game transitioned to the midgame. That attempt was move 68. If I were playing a human, I would never have played that move. It captured four black stones on the right side, but to build territory I should’ve played wider. I intentionally pressured AlphaGo. The follow-up was move 78, which caused a critical bug. Every move from 68 to 78 was a trick.

You said the Game 5 loss was the most regrettable. Could you tell us why?

I felt bad that I had used a trick to win in Game 4. So I went for a straight-up fight in Game 5 and lost. There was a bug in Game 5 too, but it came too early and I couldn’t take control of the board.

Do you regret playing AlphaGo?

I don’t regret the match itself. But I do regret one thing. After Game 4, Hassabis asked me what I thought about releasing AlphaGo’s source code. I agreed without much thought, but later I realized that was a mistake. Of course, Google would have released it anyway.

Go grandmaster Lee Se-dol speaks during an interview with JoongAng Ilbo in Mapo District, western Seoul, on June 19. [JANG JIN-YOUNG]

Why was that a mistake?

What good came from open-sourcing it? The value of Go was damaged. Go is a game where humans find the best move in an abstract space. But after AlphaGo, it became a game with correct answers. In a game with fixed answers, what is a professional Go player? A pro gamer? Even pro gamers don’t just follow answers given by a computer.

Some say AlphaGo raised the overall level of play. What is your take?

Those people will say the skill level rose across the board, but that’s not true. The gap actually widened. Shin Jin-seo was already strong before AI. After studying AI, he became an unbeatable player. Your results now depend on how well you can study and utilize AI. This isn’t unique to Go.

Still, weren’t there valuable lessons?

AlphaGo broke human preconceptions. Early 3-3 invasions are a prime example. Playing on the fifth line or shoulder hits — all techniques AlphaGo introduced. Now we see they make sense, but humans avoided them for centuries due to tradition. Today, everyone starts with 3-3 invasions.

Is the future of Go a bleak one?

I made the mistake of not realizing that AlphaGo could improve so much in five months. When ChatGPT first came out, people mocked it. Now look how fast it’s advanced. We’re in the age of AI, but everyone is making the same mistake I did. We need to expand our thinking. We must collaborate with AI to create something entirely new. In Go, that may fall to the next generation — one that learns the game from the ground up with AI.

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.
BY SON MIN-HO [[email protected]]