Go World No. 1 Shin Jin-seo Shocks AI KataGo With 4.5-Point Win, Levels Historic Man-vs-Machine Series

SEOUL — Shin Jin-seo, the world's top-ranked Go player, defeated the artificial intelligence program KataGo on Sunday, evening their best-of-three match at one win apiece in a closely watched showdown between human and machine intelligence.

The 9-dan grandmaster won Game 2 of the "Ssen Math·Hankyung Gisinjeon" series by 4.5 points, playing Black in a match that lasted roughly four hours and 50 minutes at The Korea Economic Broadcasting studio in Seoul's Jung District. The contest was played as a handicap game, with Shin given two black stones at the outset.

The victory marked Shin's first official win against KataGo, one of the most powerful open-source Go engines in the world. He had lost the opening game two days earlier after resigning in the 245th move, and he said beforehand that he had never beaten a similarly strong AI system in practice matches, even with the same two-stone advantage.

A different opening strategy

Sunday's result hinged on decisions Shin made in the opening minutes of the game. When KataGo again took the upper-left star point, as it had in Game 1, Shin responded at the lower-right star point rather than the small-knight move he used previously. When KataGo invaded the 3-3 point in the lower right shortly after, Shin steered the game into a large-scale joseki, a set sequence of established opening moves, that settled roughly a quarter of the board within about five minutes and 53 total moves.

Although the sequence could have concluded more simply, Shin chose a more complex variation. The pattern he used is known among professional players as the "AlphaGo joseki," a sequence widely studied since Google DeepMind's AlphaGo popularized it in 2016. By locking in a large area of the board early through a fixed sequence, the approach limited the number of variations KataGo could generate later in the game — a tactic that effectively turned techniques originally popularized by AI back against the machine.

Broadcast win-probability models showed Shin starting the two-stone handicap game with a 99% chance of victory and roughly an 18.5-point lead. In Game 1, that early advantage collapsed after Shin struggled to respond to an unfamiliar high approach move from KataGo, and a miscalculated move in the lower right in the middle of the game allowed the AI to erase his lead entirely.

Holding the advantage under pressure

Game 2 followed a different trajectory. After narrowing the board with the early joseki, Shin maintained his lead past the 150th move. He cut into KataGo's stones in the center of the board and launched an attack. When KataGo counterattacked near a connection point, Shin's on-screen win probability dipped to 89% at one stage, but he continued reading through the position rather than retreating from the fight.

Unlike in the first game, when an attempt to encircle KataGo's entire central group backfired, Shin instead cut off White's consolidation and blocked a series of counterattacks, preserving his advantage through the middle game. KataGo attempted a late tactical push in the lower-center portion of the board, but Shin closed off the attempt and protected his point margin heading into the endgame.

Speaking after the match, Shin said he thought he had prepared well but that KataGo's aggressive play made him briefly think he might lose. He added that he calmed himself, worked through the calculations and was glad to secure what he called a valuable first win.

About KataGo

KataGo is an open-source Go program developed by American computer scientist David Wu, built on refinements to the self-play training methods pioneered by the AlphaZero family of AI systems. Beyond calculating win probabilities, the program can estimate the expected margin of victory in points, a feature that has made it a standard training tool for professional players and national team members, as well as a common resource for broadcast analysis of Go matches.

The format of Sunday's match reflected the handicap given to the human player: Shin was allotted five hours of thinking time plus a 30-second byoyomi period for each move, while KataGo was required to move within 20 seconds on every turn.

Stakes rise for the finale

With the win, Shin secured a 50 million won ($36,000) bonus tied to Game 2. The outcome sets up a decisive Game 3, scheduled for 10 a.m. on July 21 at the same Seoul studio, with live coverage on Baduk TV. Should Shin win the final game to take the series two games to one, he stands to receive a Genesis G90 sedan as an additional prize.

The series has drawn significant attention in South Korea's Go community, both for its competitive stakes and for what it represents in the broader relationship between elite human players and Go-playing AI. Since AlphaGo's landmark 2016 victory over Lee Sedol, AI systems have become deeply embedded in how professional players study the game, with many of today's most common opening strategies, including the one Shin used to win Sunday's game, originating from AI analysis rather than human innovation.

Game 3 will determine the outright winner of the three-game series and is expected to draw a large audience among Go fans in South Korea and internationally, given the rarity of a top-ranked human player recording a confirmed win over a leading AI system in a formal, broadcast setting.