AlphaGo versus Ke Jie

The AlphaGo versus Ke Jie match was a historic three-game Go encounter between AlphaGo Master , an artificial intelligence program developed by DeepMind , and Ke Jie , the world’s top-ranked human Go player at the time. The match took place from 23 to 27 May 2017 as part of the Future of Go Summit in Wuzhen , China. AlphaGo emerged victorious in all three games, marking a definitive milestone in the intersection of artificial intelligence and traditional board games.

Match Overview

Final Result: AlphaGo 3–0 Ke Jie

Background

Ke Jie: The Reigning Human Champion

At the time of the match, Ke Jie was universally recognized as the strongest human Go player in the world. He held the No. 1 ranking under multiple systems, including:

• Rémi Coulom ’s ranking system (since late 2014)
• Korea Baduk Association ’s rankings
• Japan Go Association ’s rankings
• Chinese Weiqi Association ’s rankings

Ke Jie had already faced AlphaGo Master in online games between late 2016 and early 2017, losing all three encounters. This set the stage for the high-stakes rematch in Wuzhen.

AlphaGo Master: The AI Prodigy

The version of AlphaGo used in this match was AlphaGo Master , the same iteration that had dominated top professional players in 60 consecutive online games under the pseudonym “Master.” Unlike its predecessor, which relied on human game data, AlphaGo Master was trained primarily through self-play, using four Tensor Processing Units (TPUs) on a single machine.

DeepMind estimated that AlphaGo Master was three stones stronger than the version that defeated Lee Sedol in 2016. However, what was not publicly known at the time was that DeepMind had already developed AlphaGo Zero , a far more advanced version that learned entirely from self-play without human input. The existence of AlphaGo Zero was only revealed in October 2017 when the research was published in Nature.

The Games

Game 1 (23 May 2017)

Result: AlphaGo (White) wins by 0.5 points after 289 moves.

This game was a nail-biter, with AlphaGo securing the narrowest possible margin of victory in Go—a half-point win. Ke Jie, playing as Black, employed aggressive tactics, but AlphaGo’s precision in the endgame proved decisive.

• Key Moments:
  • Ke Jie attempted to exploit perceived weaknesses in AlphaGo’s opening.
  • AlphaGo responded with unconventional moves, including a controversial shoulder hit that baffled commentators.
  • The game stretched to 289 moves, one of the longest in professional Go history.

Game 2 (25 May 2017)

Result: AlphaGo (Black) wins by resignation after 155 moves.

About one hour into the game, Demis Hassabis , CEO of DeepMind, tweeted that AlphaGo’s internal evaluations showed Ke Jie was playing “perfectly.” However, Ke Jie later lost ground in the lower part of the board, and after four hours, AlphaGo simplified the position, making it clear that Ke Jie was at a disadvantage.

• Key Moments:
  • Ke Jie attempted a complex invasion, but AlphaGo’s counterplay was flawless.
  • AlphaGo’s efficiency in reducing complexity forced Ke Jie to resign.

Game 3 (27 May 2017)

Result: AlphaGo (Black) wins by resignation after 209 moves.

The final game saw Ke Jie, playing as White, resigning after AlphaGo built an insurmountable lead. At the time of resignation, AlphaGo had 90 minutes of its time remaining, while Ke Jie had 32 minutes left.

• Key Moments:
  • Ke Jie experimented with unorthodox openings, but AlphaGo adapted seamlessly.
  • The game concluded with AlphaGo demonstrating superior positional judgment.

Aftermath and Legacy

Prize and Recognition

• Prize Money: Google DeepMind offered $1.5 million to the winner, with the losing side receiving $300,000 for participation.
• Honorary Title: Following the match, the Chinese Weiqi Association awarded AlphaGo an honorary 9-dan professional title, the highest rank in Go.

AlphaGo’s Retirement

After the match, DeepMind announced that AlphaGo would retire from competitive play. The team shifted focus to other AI research areas, including AlphaZero (a generalized version capable of mastering multiple games) and later MuZero (which learns without prior knowledge of game rules).

Cultural and Technological Impact

• China’s Coverage: Despite initial restrictions on live-streaming, the match was broadcast on Zhejiang TV and widely discussed in Chinese media.
• AI Milestone: The match reinforced the idea that AI had surpassed human expertise in Go, a game long considered the pinnacle of strategic complexity.

See Also

• AlphaGo versus Lee Sedol (2016)
• Future of Go Summit
• AlphaGo Zero