A decade after AlphaGo, Shin Jin-seo proves that humans still have game
The world’s top Go player defeated AI program KataGo 2-1 in a two-stone handicap series, showing humans can still pose a challenge if conditions are right.
Shin Jin-seo, the world's top Go player, waits for the start of his match against KataGo, one of the world's strongest Go AI programs, during the three-game Ssen Math-Hankyung Kishin Match at Korea Economic TV's studio in Jung District, central Seoul, on July 21.YONHAP
A decade after AlphaGo convinced the world that AI had surpassed human capabilities in Go, Shin Jin-seo, the world's top-ranked player, showed that the gap may not be entirely unbridgeable. By defeating KataGo, one of the strongest Go AI programs ever created, in a three-game handicap match, Shin demonstrated that humans can still compete and win — under the right conditions.
After losing his first match against the AI last week, the 26-year-old Korean professional won Tuesday's deciding game by 11.5 points as Black after 221 moves, taking the three-game series 2-1.
But the result cannot be understood without one crucial detail: Shin played with a two-stone handicap — which means he was allowed to put two stones on the board before the AI made its first move, giving him roughly an 18-point starting advantage — granted to compensate for the vast gap between human and AI.
Even with that advantage, the victory was remarkable.
Shin dominated the final game from the opening. He built a solid position, neutralized KataGo's attacks before they could develop and patiently resisted the AI's repeated attempts to lure him into complications.
Rather than forcing the game, he reduced uncertainty and steadily preserved his handicap advantage to the end. There was never a genuine moment of danger.
The AI's win probability, which began below 1 percent because of the handicap, never once climbed above that level. Shin's calm, methodical play recalled legendary Go master Lee Chang-ho at the height of his career.
“I thought this was an extremely difficult challenge before the match, which is why I asked for favorable conditions,” Shin said afterward. “Still, I think it was meaningful because it showed that humans can hold their own against AI to a certain extent.”
The game was organized to mark the 10th anniversary of the historic Google DeepMind Challenge Match between former Go professional Lee Se-dol and AlphaGo, a five-game contest that transformed public perceptions of AI.
Shin Jin-seo, the world's top Go player, plays against KataGo, one of the world's strongest Go AI programs, during the three-game Ssen Math-Hankyung Kishin Match at Korea Economic TV's studio in Jung District, central Seoul, on July 21.NEWS1
Dubbed the Ssen Math-Hankyung Kishin Match, this year's event was hosted by the Hankyung Media Group and organized by the Korea Baduk Association. The match awarded Shin 250 million won ($170,000) and a Genesis G90 sedan for winning.
A decade ago, humans entered the match against AlphaGo without truly understanding how strong AI had become. Many quietly believed that humanity would prevail, only to watch as AI triumphed.
Today, however, AI is now overwhelmingly superior. Instead of teaching AI how to play Go, humans now study how AI plays Go. The world's best players memorize AI variations, imitate AI strategies and compete to play the style closest to machine perfection.
The world's elite players still lose about 0.3 points of value on every move against today's strongest AI, despite making no obvious mistakes, according to Hong Min-pyo, coach of Korea's national Go team.
After just 10 moves, that difference amounts to roughly three points. After 100 moves, it grows to about 30 points. Shin alone has narrowed that gap to about 0.2 points per move, earning him the nickname “Shin-telligence.”
Korean professional Go player Lee Se-dol, right, puts down the first stone against Google's AI program, AlphaGo, as Google DeepMind's lead programmer Aja Huang sits during the Google DeepMind Challenge Match in Seoul on March 9, 2016.AP/YONHAP
The challenge was made even tougher because KataGo represents the cutting edge of Go AI. It analyzes roughly 26,000 possible variations within 20 seconds.
Compared to AlphaGo, the version Lee faced in 2016, KataGo is vastly more powerful.
“If AlphaGo played KataGo 10,000 times, KataGo would win all 10,000 games,” professional player Park Jeong-sang once explained.
Against an opponent of that caliber, the two-stone handicap and five-hour time limit were less an advantage than a necessity. Shin himself said he had never previously defeated KataGo with a two-stone handicap before this event.
Since AlphaGo's arrival 10 years ago, humans have done little but lose to increasingly powerful AI. As artificial intelligence has improved, the gap has seemed only to widen.
Yet Shin's victory suggests that the gap, while enormous, is not completely beyond human reach.
Much as how humans can't move as quickly as cars but people still keep running, perhaps the same is true of humanity's quest to keep pace with AI.
However, if another human-versus-AI match is held in the future, the conditions will almost certainly be less favorable to the human player.
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.
