Humans Strike Back: Shin Jinseo Defeats KataGo 2–1 in the Kishin Match
Published on July 21, 2026 by StoneBase Team
Ten years after Lee Sedol’s historic win against AlphaGo, another chapter in human-versus-AI Go history was written in Seoul. Shin Jinseo 9p — the world’s top-ranked player, nicknamed “Shin-telligence” for his AI-like precision — faced KataGo, the strongest open-source Go AI, in a three-game match with a two-stone handicap. And he won the series 2–1.
This isn’t a simple replay of 2016. The handicap itself tells the story: AI has advanced so far that even the best human on earth needs a two-stone head start. The question wasn’t “can a human beat AI at even odds?” but rather “with two stones, can a human still win?” Shin’s answer was a decisive yes.
Here’s everything we know about the match — the conditions, the games, the strategies, and what Shin himself learned about the AI’s weaknesses. Game records (SGF) for all three games are available for download below.
The Match: SSEN Math & Hankyung Kishin Match
| Detail | Info |
|---|---|
| Full name | 쎈수학·한경 기신전 (SSEN Math & Hankyung Kishin Match) |
| Organizer | Hankyung Media Group / Korea Baduk Association |
| Dates | July 17 (Game 1), July 19 (Game 2), July 21 (Game 3), 2026 |
| Venue | Korea Economic TV Studio, Jung-gu, Seoul |
| Format | 3 games (all played regardless of score) |
| Handicap | Shin (Black) with 2 handicap stones; Black gives 0.5 points komi |
| Rules | Japanese rules |
| Time (Shin) | 5 hours main time + one 30-second byo-yomi |
| Time (KataGo) | 20 seconds per move, no main time limit |
| Hardware | Dedicated Tygem rig: 4× RTX 3090 GPUs, 96 GB VRAM total |
| KataGo operator | Lee Dan-bi 1p (placed KataGo’s moves on the board) |
| Prize | ₩50M per game + ₩50M per win; Genesis G90 for 2+ wins (Shin earned ₩250M total, ~$170K) |
The hardware deserves attention. KataGo creator David J. Wu built the engine to run on consumer hardware, not supercomputers. The 4× RTX 3090 setup was custom-built for this match, giving KataGo enormous parallel compute power within its 20-second-per-move thinking time.
Why Two Stones?
The two-stone handicap (~18 points advantage) wasn’t arbitrary. Shin revealed before the match that in practice games, he had never beaten KataGo at two stones, and never lost at three. The handicap sat exactly on the boundary of human capability — making the challenge both realistic and meaningful.
This also highlights how far AI has come. In 2016, Lee Sedol played AlphaGo at even (no handicap). A decade later, the world’s strongest human needs two stones to make it a fair fight.
Expert Predictions Were Split
Prognostications divided the Go community:
- Cho Seung-ah 7p: Shin 2–1
- Kim Myung-hoon 9p: KataGo 2–1
- Lee Sedol (the man who beat AlphaGo): Shin wins
“No matter how well AI plays Go, it has no conviction, no philosophy, no narrative. I believe Shin will win.” — Lee Sedol, pre-match interview
Lee Sedol’s prediction turned out to be correct.
Game 1 (July 17): The AI’s Shock Opening
KataGo 1–0 Shin · 245 moves · White wins by resignation · Download SGF
KataGo wasted no time establishing its alien style. On White’s second move (move 3 of the game), with a corner still open, KataGo played high on the right side — a move virtually unseen in human professional Go. The message was clear: this AI plays by its own rules.
Shin later admitted:
“The moment I saw that move, my heart almost broke. I honestly wanted to request a rematch.” — Shin Jinseo, after Game 1
Despite the two-stone advantage, Shin’s 99% opening winrate began eroding around move 70, and a mistake near move 90 dropped it into the 80s. The decisive blow came at move 102: Shin staked the game on surrounding White’s large central group — but KataGo escaped cleanly, and the evaluation flipped for good. By move 103 KataGo had taken the lead, and by move 117 Black trailed by over 7 points. Guest commentator Hong Min-pyo, coach of the Korean national team, lamented the heavy losses from the lower right to the bottom edge. Shin resigned after about 4 hours 20 minutes.
His pre-match concern had materialized: if it becomes a fight, even two stones aren’t enough.
Game 2 (July 19): “Shin-telligence” Strikes Back
Shin 1–1 KataGo · 299 moves · Black wins by 4.5 points · Download SGF
Shin changed everything. His new strategy: eliminate complexity.
In the lower right, he entered a 53-move ultra-complex joseki — the kind of “flying knife” AI joseki where one wrong move is fatal. KataGo’s miscalculation: its opponent is the human most like an AI. Shin executed the entire sequence flawlessly, settling a quarter of the board and removing KataGo’s favorite weapon — complicated fighting.
But avoiding fights entirely wasn’t an option. KataGo invaded the top side and called Shin’s bluff, taking the territory there — and in the midgame, a central battle finally erupted. This time Shin chose to fight: he allowed the cross-cut and backed his reading, aiming to fill the center with battle, survive it, and enter the endgame ahead. The broadcast winrate briefly dipped to 89% when KataGo counterattacked, but Shin answered everything and emerged with his lead intact.
After nearly five hours, Shin won by 4.5 points — becoming the first professional to officially defeat KataGo at a two-stone handicap in a sanctioned match. Korean media reported he drank 11 bottles of mineral water during the game.
“There were countless moments I wanted to push back or attack. Each time, I held back and endured it. When the fight broke out in the center, I checked the clock and saw I had time to spare. I decided to fight while minimizing my losses.” — Shin Jinseo, after Game 2
Hong Min-pyo praised the performance: “From a difficult shape he entered the endgame and lost absolutely nothing.”
The full game was broadcast live by Korea Economic TV (Korean commentary).
Game 3 (July 21): The Blueprint for Beating AI
Shin 2–1 KataGo · 221 moves · Black wins by 11.5 points · Download SGF
The decider was Shin’s masterpiece — and it almost started with a farce. Shin went to the Korea Baduk Association building by mistake instead of the TV studio, arriving late. Off the board: flustered. On the board: flawless.
KataGo changed its approach, opening with a komoku (small-point) in the upper left instead of its usual moves. Shin responded calmly and executed his declared strategy: avoid fighting, settle with territory.
Around move 80, while pressing White’s stones, Shin built a massive Black framework spanning the top side into the center. Unlike Games 1 and 2, the 99% winrate never cracked. No collapse. No drama. After just over three hours, Shin closed out the series.
Post-match, Shin received:
- ₩250 million total prize (~$170,000)
- A Genesis G90 luxury sedan
- The title of first to beat KataGo at two stones in an official match
What Shin Learned: The AI’s Perfection Is Its Weakness
The most fascinating part came in the post-series press conference. Shin shared his insights after three intense games against the machine:
“KataGo’s ‘perfection’ is itself a weakness — it tends to shy away from lines that could become difficult. Next time, I’d like to try under even less favorable conditions.” — Shin Jinseo, post-series press conference
He also reflected on the broader meaning:
“Rather than imitating the AI’s style, I felt it’s important to play calmly in my own way. Before the match I thought it was a very disadvantageous challenge, but through this series I showed how far humans can compete against AI.” — Shin Jinseo
The winning formula evolved visibly across the three games:
| Game | Strategy | Result |
|---|---|---|
| 1 | Tried to fight the AI on its own terms | Loss |
| 2 | Simplified the board with a long joseki, fought only when necessary | Win |
| 3 | Denied AI complexity entirely, steered into a quiet territorial game | Win |
The lesson: don’t try to out-calculate a calculator. Channel the game where creativity and judgment matter more than raw reading.
Game Records & Videos
- Download SGF files (all three games): Game 1 · Game 2 · Game 3 — cleaned and standardized for use with any SGF viewer, including StoneBase
- Game 2 full broadcast: Korea Economic TV on YouTube (Korean commentary, ~5 hours)
- English commentary: “The Painting of God” (Game 1) and “The Man of Steel” (Game 2) by Go Games Series — search on YouTube
What This Means for Go
Lee Sedol’s “divine move” (move 78 in Game 4 against AlphaGo) was an even-game win. Shin’s series win came with a handicap. The comparison isn’t direct — but that’s the point. It measures a decade of AI progress.
In 2016, a human beating AI was a miracle. The AI was new, alien, and seemingly unbeatable.
In 2026, a human beating AI is a strategy. Shin designed a game plan specifically to neutralize the AI’s strengths — and executed it twice in a row. This is what happens when a generation of players grows up training with AI: they learn not just AI moves, but AI thinking, and eventually, how to beat AI at its own game.
The Kishin Match proved that humans who trained on AI have grown stronger too. The gap between human and machine in Go isn’t just about technology pulling ahead — it’s also about humans catching up by learning from the machines themselves.
Want to analyze these games yourself? Download the SGF files above and load them in StoneBase to run KataGo analysis on every move. Get StoneBase for free.
Sources
- Korea JoongAng Daily — ‘Shin-telligence’ beats an actual AI
- KBS World — S. Korea’s Go Master Shin Jin-seo Beats AI KataGo
- Asia Today — Game 3 report
- Korea Times — Humans strike back
- Kyunghyang Shinmun — Game 3
- Korea Baduk Association — official announcement
- gokifuai.com — Full 3-game summary (Japanese)
Ready to analyze with AI?
If you haven't set up KataGo yet, check out our complete KataGo setup guide — it covers Windows, Mac, and Linux installation in minutes.