The History of Go (Baduk) and Life After AlphaGo
Go's history from China to Korea and Japan, its basic rules, the 2016 AlphaGo vs Lee Sedol match, and how the game has changed in the age of AI.
Go (known in Korea as baduk) is a two-player strategy game in which players take turns placing black and white stones, and whoever claims more territory wins. It is believed to have been invented in China more than 2,500 years ago and is considered the oldest board game still played continuously to this day.
The rules are surprisingly simple, but the number of possible positions on a 19×19 board exceeds that of any other game. For that reason, Go was long called "the hardest game for a computer to beat a human at." The 2016 match in Seoul between AlphaGo and Lee Sedol 9-dan overturned that belief.
In this post, we look at Go's origins and its spread to Korea and Japan, its basic rules, key moments from the AlphaGo match, and how the Go world has changed since AI.
A Game Born in China
Go is called weiqi (圍棋) in China, go (碁) or igo (囲碁) in Japan, and baduk in Korea. The English name "Go" comes from Japanese.
The oldest written record is generally considered to be the Zuo Zhuan (左傳), a historical text from around the 4th century BC, in which Go appears in an account of events from 548 BC. The word yi (弈), referring to Go, also appears in the Analects and Mencius.
Legend has it that the ancient Chinese Emperor Yao (堯) invented the game to teach his son, but that's strictly legend. What is certain is that Go was one of the four accomplishments (qin, qi, shu, hua; 琴棋書畵) expected of cultivated scholars in ancient China.
There is archaeological evidence that it was first played on a 17×17 board, and by the Tang dynasty (618–907), today's 19×19 board had become standard.
Go in Korea: From Sunjang Baduk to Modern Go
Go reached the Korean Peninsula between the 5th and 7th centuries and was enjoyed mainly by the upper classes. The name "baduk" is a native Korean word rather than a Sino-Korean one, and the leading view is that it combines old words meaning "field" and "stone."
Korea had its own distinctive form of Go: sunjang baduk, which had taken hold by around the 16th century. In this style, stones are placed on set points before the game begins, and it remained Korea's main form of Go until the late 19th century. As modern Japanese-style Go was later reintroduced, sunjang baduk gradually disappeared.
In the late 20th century, Korean Go rose to the top of the world. Cho Hunhyun 9-dan won the first Ing Cup, an international tournament first held in 1989, and his student Lee Chang-ho 9-dan dominated international competitions for more than a decade. Lee Sedol 9-dan was also a leading player who carried on this tradition.
Go Culture Flourishes in Japan
Go reached Japan in the 7th century, became popular at the imperial court in the 8th century, and spread to the general public by the 13th century. In 1603, Tokugawa Ieyasu appointed the monk Nikkai, the strongest player of the time, to an official post overseeing Go. Nikkai founded the Hon'inbō house under the name Hon'inbō Sansa.
The four houses of Hon'inbō, Yasui, Inoue, and Hayashi competed with state support, and playing strength rose dramatically in the process. The dan and kyu ranking system still used today also took shape in this period. For most of the 20th century, Japan was the center of Go, and promising young players from Korea and China went to study there.
Basic Rules of Go
The rules of Go can be summed up in a few points.
- Black and White take turns placing one stone at a time on an empty intersection of the board. Black plays first.
- Once placed, a stone is not moved.
- When a stone or a connected group of stones is completely surrounded by the opponent's stones on all orthogonally adjacent points, it is captured and removed from the board.
- To prevent the same position from repeating endlessly, there is a "ko" rule that forbids immediately recapturing.
- When both players judge there is nowhere left to play and pass their turns, the game ends and territory is counted to decide the winner.
| Item | Details |
|---|---|
| Board size | 19×19 (361 intersections); 9×9 and 13×13 for beginners |
| Number of stones | Typically 181 black, 180 white |
| Komi | Points given to White to offset Black's first-move advantage |
| Komi size | 6.5 points under Korean and Japanese rules, 7.5 points under Chinese rules |
| Kyu and dan | Amateurs progress from kyu to dan; professionals range from 1-dan to 9-dan |
Komi was introduced in the 20th century. The extra half point is added to eliminate draws.
The number of legal positions possible on a Go board is calculated at about 2.1×10^170, far larger than the number of atoms in the observable universe (about 10^80). This is why Go was so hard for computers.
Before AlphaGo: The Wall Computers Couldn't Climb
In chess, IBM's Deep Blue defeated world champion Kasparov in 1997, but Go programs remained at amateur level for a long time after that. Only in 2008 did the program MoGo beat a professional player with a nine-stone handicap, and before 2015 the best Go programs were still at amateur dan level. Some researchers even said the day a computer beat a top player would never come.
2016: AlphaGo and Lee Sedol
What changed the situation was Google DeepMind's AlphaGo.
| Date | Event |
|---|---|
| October 2015 | Defeated European champion Fan Hui 2-dan 5–0, the first program to beat a professional on a full 19×19 board without a handicap |
| March 2016 | Five-game match against Lee Sedol 9-dan in Seoul; AlphaGo won 4–1 |
| Late 2016–early 2017 | An improved version played professionals on online Go servers under the name "Master" |
| May 2017 | Beat then world No. 1 Ke Jie 9-dan 3–0 at the Future of Go Summit, after which AlphaGo retired from competitive play |
| October 2017 | AlphaGo Zero, which learned on its own without human game records, published in Nature |
The match with Lee Sedol 9-dan was held at the Four Seasons Hotel in Seoul from March 9 to 15, 2016. AlphaGo's move 37 in Game 2 was a "shoulder hit" that most professionals would never have considered, and commentators praised it as a creative move.
Conversely, in Game 4, Lee Sedol 9-dan's move 78 exposed a weakness in AlphaGo, and that game became the only one of the five won by a human. Many professional players called it the "divine move." DeepMind later said it had fixed this weakness.
After the match, the Korea Baduk Association awarded AlphaGo an honorary 9-dan rank. The match is often compared to the 1997 showdown between Deep Blue and Kasparov. In 2017, the documentary AlphaGo, chronicling the process, was made, and in 2020 it was released for free on DeepMind's YouTube channel.
AlphaGo Zero learned by playing against itself without any human game records, surpassing the version that beat Lee Sedol 9-dan in just three days and going 100–0 against the earlier AlphaGo.
How Go Changed After AlphaGo
The Go world changed dramatically after AlphaGo.
- How players study: Professional players use AI to reanalyze their games and see where their win rate dropped. Fan Hui 2-dan said that playing AlphaGo actually broadened his perspective.
- Rethinking joseki: It has become common for moves long taken for granted to show up as bad moves in AI analysis, or for moves that looked strange to be reassessed as good ones. A typical observation is that AlphaGo's unfamiliar moves first looked questionable but were understood later.
- Players' struggles: Announcing his retirement in November 2019, Lee Sedol 9-dan explained, in essence, that AI had become an entity that cannot be defeated, so he could never be the best.
Competitive Go is now split along two axes: "competition between humans" and "studying by learning from AI."
According to a 2016 survey of 75 member countries by the International Go Federation, more than 46 million people worldwide know how to play Go, and more than 20 million currently play. Most live in East Asia, but interest in Go has also grown since the AI matches.
How to Get Started with Go
- Start on a 9×9 board. Games are short, so you can quickly learn the rules and get a feel for capturing.
- Practice atari and capturing first. Getting in the habit of counting the liberties of your opponent's stones is fundamental.
- Understand making two eyes. The principle that a group with two eyes cannot be captured is the starting point of life and death.
- Solve a few life-and-death problems every day. Consistently working through short problems is considered the surest way to improve.
- After a game, use AI analysis to see where your big mistakes were. That said, rather than memorizing the AI's moves as-is, it helps to think about why those moves are good.
Go, with its 2,500-year history, didn't come to an end when it met AI; it gained a new study tool. The joy of two people sitting across from each other and playing a game remains the same as ever.