Ten Years Later, Faker Sits Beside 李世石
Summary
- The episode places 李世石 and Faker in the same frame as 2 generations of “human representatives” separated by 10 years. 10 years ago, 李世石 said he would definitely win and would aim for a 5-0 sweep against AlphaGo; now, with T1 highly likely to face Grok in a League of Legends match, Faker still doubts whether AI can become strong enough to win in the near term. 庄明浩 defines 李世石 as “the one who went first” (先验者)—the person who went through the door for everyone else, took the punch, then emerged to show the wound.
- The real value of Move 78 lies not in “divinity,” but in the ability to find and exploit a system vulnerability under extreme pressure. In AI’s probability model, the odds of a human playing the move were roughly 1 in 10,000; it scrambled the weights of that version of AlphaGo and exposed a vulnerability. 10 years later, 李世石 called it akin to the only “dirty trick” and “cheap shot” he had ever used, and possibly the last (“阴招”和“损招”). 孙锡熙 then supplied the crucial counterpoint: recognizing that the move could confuse the opponent was itself proof of the insight that constitutes strength.
- 李世石 later personally verified the “chasm” between humans and top-tier Go AI, and ultimately left Go behind. In 2021, he spent 1.5 months winning 1 game under conditions in which AI gave him 2 stones, had only 20 seconds per move, and he had unlimited time; after removing the handicap and playing normally, he realized that “even if I spent my whole life playing it, I could never beat it.” “After that, I don’t think I ever touched Go again.”
- Whether League of Legends can delay a Go-style ending depends not only on the number of variables, but also on what constraints the match imposes on AI. 李世石 believes games are harder for AI to conquer than Go, but notes that Go was once described in the same terms—as having nearly infinite variables and relying on intuition and artistry; to AI, that is merely “finite infinity” (“有限的无限”). Faker envisions bringing AI’s reaction speed close to the human level, even banning it from transmitting data through something akin to a back-end interface, so that the result turns on macro judgment and tactical decisions.
- The host’s hypothetical AI question also forces a reexamination of how humans understand “intuition.” The host imagines AI asking Faker: “How do you notice that an opponent is setting up specifically against you? Where does that feeling come from? How do you sense and exploit the fear in your opponent’s mind?” Faker explains intuition as “the sum of all the data accumulated through past experience,” while acknowledging that if neither side had emotions, “no audience would want to watch.” 李世石 locates the value further in a narrative made up of “personality, emotion and story”: “We generally do not assign special meaning to things without a narrative.”
- AI becoming a teacher does not automatically mean human progress; it can also compress exploration into imitation of the right answer. 李世石 observes that top professional players increasingly copy AI techniques, prompting him to ask: “Can we still call this human progress?” Once professionals stop changing paradigms and creating their own styles, and merely follow AI’s answers, Go becomes difficult to sustain as a culture and an art form.
- Faker has not withdrawn the challenge, but 李世石’s experience has cracked his certainty. He accepts that a machine defeating humans “will happen sooner or later”; if he takes the match, he will give it everything, and he is prepared to accept reality if he loses. After hearing the full account, he said he needed to reassess and seriously asked himself: “Could I really lose?” 庄明浩’s conclusion is not that humans must win, but that even when AI can deliver more standardized code, images and reports, people must preserve “their narrative, their obsession, their dirty tricks and that 1-in-10,000 intuition.”
Deep dive
1. The 5-0 boast from 10 years ago became a mirror of the reality Faker has yet to face
庄明浩 calls the Korean interview the “B-side” of the 樊麾 story: after AlphaGo beat 樊麾 in all 5 games, he briefly felt he could no longer play Go, then joined DeepMind to help make it stronger. The missing piece was how 李世石, who had played on behalf of all humanity, processed that defeat.
孙锡熙 unexpectedly played a clip from 李世石’s 2016 pre-match interview. “I’m definitely going to win. I’ll try for a 5-0.” The footage then cuts back to 2026, where he watches that unfamiliar yet undeniable version of himself with a complicated smile. Faker laughed too, and in his laughter was the same certainty 李世石 had shown before reality had hit.
After losing Game 1, 李世石 still had doubts. In Game 2, he said he was “fully switched on” and still lost—something the program noted he had never expressed in those terms before. By Game 3, he said he had not felt ahead “for even a single moment.” He then said, almost through tears: “It was an AI developed by humans that defeated humans.” 樊麾’s judgment was harsher: no one won that day; only 李世石, representing all humanity, lost.
That explains 李世石’s wry smile when he heard Faker say, “We will ultimately win.” He could not simply declare that Faker would lose, but with the restraint of someone who had gone first, he warned that he had already traveled the full path—from absolute confidence, through a partial counterattack, to the final confirmation of the gap.
2. Move 78 was not a miracle, but the only gamble worth taking in a battle that was already lost
In Game 4, Move 78 had roughly a 1-in-10,000 probability of being played by a human according to AI’s calculations, yet it was the only choice 李世石’s intuition would allow. It disrupted AlphaGo’s weights and exposed a vulnerability in that version, producing the last formal-match win by a human against a top-tier Go AI to date. 樊麾 remembers trying to keep his composure as he walked into the press conference, only to break into “the happiest smile in the world.”
10 years later, 李世石 dismantled the move’s aura of divinity: “That wasn’t a conventional pattern.” He described it as something like “a dirty trick” and “a cheap shot” (“阴招”和“损招”), designed to lure AlphaGo into exposing a vulnerability. It was the 1st time—and possibly the last—that he had played that way in his career.
孙锡熙 refused to let that self-deprecation pass without a response: “Recognizing that the move could confuse the opponent—that judgment itself is proof of insight, and insight is strength.” The move was neither a miracle appearing from nowhere nor mere luck. It was a professional player, forced into a desperate position, finding the last tactic that might still work.
Game 5 featured no similar move, which showed 李世石’s composure. The overall score was already 3-1; even another such move would not have changed the outcome. The other side of the “divine move” was that he staked a style he had never used in his life on a battle he was bound to lose overall, simply to secure at least 1 win.
3. The experiment won in 1.5 months erased 李世石’s last illusion
After retiring in 2019, 李世石 still wanted to know exactly how strong AI had become. In 2021, he quickly played 4 games with AI giving him a 2-stone handicap and lost them all. His conclusion was that after AlphaGo Zero in 2017, it was no longer possible for humans to beat Go AI in real-world conditions.
Unconvinced, 李世石 designed what he called a “very fair” match: AI could think for only 20 seconds per move, he had unlimited time, and AI still gave him 2 stones. 孙锡熙 could not help asking whether that still counted as fair. Whenever 李世石 encountered a position he could not understand or found his thinking becoming confused, he stopped to sleep and resumed the next day. The game took 1.5 months, and he won.
He then removed the handicap and played a normal game, where he felt “an unprecedented sense of bewilderment” and encountered a “chasm”: “No matter what I tried, I couldn’t win,” and “even if I spent my whole life playing it, I could never beat it.” Winning the experiment under extremely favorable conditions, then proving to himself that a normal game offered no hope, ultimately meant that he “didn’t touch Go again.”
李世石 also described how he dealt with the pressure of competition: he played an amateur 1-dan player and captured every stone, trying to recover the feeling that he could still win. He also felt sorry for the friends he had crushed. 庄明浩’s reading is that when someone has been completely broken, what they need first may not be an explanation, but a victory they can still control.
4. Faker accepts that AI will eventually win, but wants it to compete “the human way” first
Faker initially questioned whether AI could evolve enough in the near term to beat him, because League of Legends has more complex logic and variables than Go. 李世石 also believes games are harder to conquer, but asks: if AI faces no restrictions, is a match against it really winnable even for Faker? Go appears nearly infinite to humans, but to AI it is only “finite infinity” (“有限的无限”).
Faker acknowledges that AI’s progress has far exceeded expectations and says he does not know the final outcome. But once a professional player accepts a challenge, he has to give everything for everyone watching. He does not find a machine defeating humans difficult to accept; it “will happen sooner or later,” and he is prepared to accept reality after a loss: “Even if I don’t go, someone else will. AI is the trend of the era. It can’t be resisted.”
李世石 points out that reaction speed itself creates a huge gap between humans and computers. Faker’s proposed match conditions are to tune AI’s reaction speed close to the human level and even prohibit it from transmitting data directly through something like a back-end interface. It should compete through human-like inputs, with the result determined by macro judgment and tactical decisions.
The host imagines asking Faker a question on AI’s behalf: “How do you notice that an opponent is setting up specifically against you? Where does that feeling come from? How do you sense and exploit the fear in your opponent’s mind?” Faker says this comes close to a philosophical interrogation, then gives a highly data-driven answer: “Intuition is actually the sum of all the data accumulated through past experience.”
5. What AI is really rewriting is not the result, but whether humans still create their own paradigms
Given the choice between having a perfect plan completely read and losing to a basic mistake, Faker finds the former harder to accept. He has also considered whether having no emotions might improve competition, but dismissed the thought: “If two players without emotions were competing, no audience would want to watch.”
李世石 puts the value more completely: “Personality, emotion and story—in other words, narrative” determine whether people can connect with something. “We generally do not assign special meaning to things without a narrative.” AI may approach or exceed optimal play, but watching, remembering and assigning meaning still revolve around specific people.
After AI defeated humans, professional players broadly began learning from it. 李世石 acknowledges that AI acting as a teacher and lowering the barrier to entry could have been positive, then pivots: “Reality is not like that.” Top players increasingly imitate AI’s logic. Once there is a correct answer, people stop changing paradigms or inspiring others: “If that’s the case, can we still call this human progress?” If Go is reduced to copying the correct answer, it can no longer be called art.
李世石 says: “We were probably the last generation to learn Go as art.” Wins and losses were merely byproducts of creating a perfect work, and their weight in the process was not particularly large. Those fully focused on creating an excellent work often won the final contest beautifully as well. What truly mattered was creating a one-of-a-kind style. Professional players were supposed to advance the technology; once Go became a matter of copying AI techniques, he could not easily find new meaning, which is why he has no regret about retiring.
6. 李世石 hands Faker his scars—and the same question to every knowledge worker
孙锡熙 laments: “We lost a Go player who was an artist too soon.” 李世石 remains measured: many professional players are still holding on, and each person may have a meaning of their own, so “there’s no need to take this too seriously.” But Go remains infinite in human eyes, which means humans still have to create their own Go.
After hearing the full account from 李世石, Faker did not withdraw from the match. He paused, admitted that he needed to reassess, and seriously asked: “Could I really lose?” The host hopes to see him recreate the moment when, at 17 in 2013, he came back from 0-2 to 3-2, then turned the deciding game against 岳伦 with a blind ko fight, prompting the world to ask once again: “Faker, what was that?”
庄明浩 ultimately turns the board toward everyone. 10 years ago, 李世石 took the punch on behalf of all humanity; 10 years later, he reopened the scar and showed where it had landed. Once AI can write more standardized code, draw more polished images and organize more orderly reports, what humans can still bring to work is unique meaning: “your narrative, your obsession, your dirty trick, your 1-in-10,000 intuition.”
The conclusion is not that humans can beat AI by turning work into art, but that doing so can keep them from losing themselves when they confront the “chasm.” 李世石 says he was part of the last generation to learn Go as art; 庄明浩 hopes this generation will become the 1st to treat work, expression and life as art again.