唯一贏過 AlphaGo 的棋士 Lee Sedol：AI 讓高手與其他人的差距更大 【AI 大人說】

Summary

The discussion examines why AI has not made all Go players equally stronger—and why the gap between top players and others may have widened.

• AI’s moves can be difficult even for experts to understand. Lee Sedol contrasts the earlier AlphaGo, which learned from human game records, with AlphaGo Zero, whose play could seem unfamiliar and hard to explain. He estimates that top players initially understood only about three of every ten important moves; players below the top level understood even fewer.
• AI is not a good starting teacher for beginners. Go is too difficult to learn from AI alone without a foundation. Players need enough experience to interpret why a move works before they can learn from it.
• Understanding developed gradually. For years, players often copied AI’s moves without fully grasping them. After roughly nine years of study, top players have begun to understand more of AI’s choices—progressing, as described in the subtitles, from about three moves to around seven—and to use the underlying ideas rather than simply imitate the moves.
• AI can lead to new human strategies. Once players understand AI’s principles, they can combine them with their own judgment and develop distinctive styles of play. Lee sees this as an encouraging change.
• The strongest players benefit most. Players who can interpret and apply AI insights gain more than those who cannot, helping explain why the skill gap can grow. Lee cites Shin Jin-seo, a leading Korean player, as an example of how far modern play has advanced compared with earlier generations.
• The broader lesson: Using AI is not enough to reach the top. People need to understand it and create something of their own with it.

Speakers and Sources

• Lee Sedol — Go professional and interview guest.
• Kim Deok-jin — Interviewer, identified in the video metadata.
• AI 101 — Channel presenting the video; any brief narration or framing in the subtitles is not clearly attributable to a named speaker.