Ten years after AlphaGo defeated Lee Sedol, artificial intelligence is no longer a shocking outsider in Go. It has become the training room, the coach, the reference point and, for many professionals, the standard they are trying to approach.
At the Korea Baduk Association in eastern Seoul, the sound of stones on boards has partly given way to players working at monitors. They review games inside AI software, compare choices against machine recommendations and watch programs play each other. The old game remains, but the way elite players learn it has changed sharply.
AI has become the professional baseline
Go is an ancient strategy game invented in China more than 2,500 years ago. Two players place black and white stones on a 19x19 grid, trying to control territory by surrounding opposing stones. Its possible board positions are so vast that players traditionally relied on principles, intuition and inherited patterns rather than direct calculation alone.
That learning culture was disrupted by AlphaGo, Google DeepMind’s program, when it defeated Lee Sedol. AlphaGo was trained on 30 million Go moves and improved by playing millions of games against itself. In 2017, AlphaGo Zero took a different path: it learned only from the rules of the game and self-play, without studying human games. After three days of training, it beat AlphaGo Lee 100 games to zero.
Google DeepMind retired AlphaGo that same year, but open-source systems inspired by AlphaGo Zero followed. In South Korea, KataGo is now the program most widely used by professional Go players. It is described as faster and sharper than AlphaGo, and it can estimate not only the likely winner but also ownership of each point on the board at a given moment.
For Shin Jin-seo, the top-ranked Go player in the world, KataGo is a daily partner. He studies its suggested moves and tries to understand why the program points to a particular place on the board. A study in 2022 by the Korean Baduk League found that Shin’s moves matched AI’s 37.5% of the time, compared with a 28.5% average among all players.
The opening game is less personal now
AI has not merely made players stronger. It has changed what good play looks like. Park Jeong-sang, a South Korean Go commentator, says that moves once treated as common sense are now rarely used, while techniques that did not exist before have become popular.
The clearest change is in the early part of a match. The first 50 moves once gave players room to express personality, taste and philosophy. Lee Sedol was known for provocative play that welcomed disorder. Ke Jie, who lost to AlphaGo Master in 2017, was associated with agile and imaginative choices.
Now many professionals begin games with efficient patterns recommended by AI. The source article describes a shift in emphasis: less of the contest is decided by original opening ideas, while the middle moves carry more weight. At that stage, memorized openings are less useful because the board has branched into too many possibilities.
This has created a visible tension in Go culture. A study in 2023 found that over a third of moves by top Go players matched AI recommendations. Many players say the first 50 moves are often identical to what AI suggests. Ke Jie told a Chinese news outlet in 2021, “I feel the exact same way as the fans watching. It’s very tiring and painful to watch.”
Players are learning from a black box
AI can identify strong moves, but it does not always explain them in a way humans can convert into clean principles. That leaves professionals in a difficult position. They can imitate the output, but they may not fully understand the reasoning behind it.
Kim Chae-young, one of the top female Go players in the world, described the transition as a kind of reset. She said, “I needed time to abandon everything I had learned before.” Looking at KataGo’s suggested moves, she added, “The intuition I had built up over the years turned out to be wrong.”
The software can display winning probabilities, but those numbers do not amount to a full lesson. Kim said, “It seems like it’s thinking in a higher dimension.” Her process has become less about working through every move through explicit logic and more about developing a new kind of intuition shaped by AI.
Researchers are also trying to extract concepts from game-playing AI. In 2024, Google DeepMind researchers extracted new chess concepts from AlphaZero and taught them to chess grandmasters using chess puzzles. Nicholas Tomlin, a computer scientist at Toyota Technological Institute at Chicago, coauthored a study probing Go concepts encoded in AlphaGo Zero and said the ideas players have learned so far are “probably only a small portion of what you could potentially learn.”
Nam Chi-hyung, a Go professor at Myongji University, says top players have not yet found the general principles behind AI moves. In her view, the machine can be copied before it can be fully understood.
AI is widening access to elite training
The same technology that has narrowed some styles of play has opened doors elsewhere. Nam says female Go players long faced a training disadvantage because the strongest experience was concentrated around top male players and competitive male circles that were hard for women to enter.
AI changes that training environment. It gives players access to a powerful opponent and analyst without relying on those older networks. Nam says, “Female players never had access to that experience,” adding, “But now they can study with AI, which has made their training environment much more favorable.”
The ranks have begun to reflect that shift. In 2022, Choi Jeong, then the top female player in the world, became the first woman to reach the finals of a major international Go tournament. She played Shin and lost, but the match marked a major step for women in Go. In 2024, Kim Chae-young made headlines by winning the Korean Go League’s postseason playoffs as the only female player in the tournament.
For Kim, AI has also changed the psychology of competition. By analyzing male players’ moves against AI, she saw that they were not beyond error. “AI broke the psychological barrier,” she says.
Human Go still has an audience
Even with AI’s strength, fans still prefer human matches. Park says games between AI programs are not very fun for fans to watch because they are too complex and too flawless. Human games include uncertainty, mistakes, recoveries and personal choices.
That is where professionals still find meaning. Shin says, “I can play a kind of Go that tells a story that only a human can.” Lee Sedol, who retired three years after losing to AlphaGo, has also become more hopeful that AI may help players approach the dream of a masterpiece game.
The result is not a simple story of replacement. AI has made Go more precise, more demanding and in some ways more uniform. It has also made world-class analysis more accessible and given players new reasons to improve. The board is the same 19x19 grid, but the mind across from every professional now includes a machine.