Go (game)

• 1. Introduction

  ›

  Go is an abstract strategy board game that has been played continuously for more than 2,500 years, making it the oldest board game still in regular play anywhere on earth. Two players. Black stones and white stones. A grid of intersecting lines. The rules fit on a single page. Yet the number of legal board positions has been calculated at approximately 2.1 times ten to the power of 170, a figure that dwarfs the estimated number of atoms in the observable universe. What kind of game hides such staggering depth behind such apparent simplicity? How did a pastime invented in ancient China become a subject of intense scientific study for artificial intelligence researchers in the 21st century? And what does it mean that, as recently as 2016, the world's strongest human player lost four out of five games to a computer program that had taught itself by playing hundreds of millions of games against itself? Those questions sit at the heart of Go's story.

• 2. A Game of Territory

  ›

  A standard Go board carries a 19-by-19 grid of lines, producing 361 intersections where stones can be placed. Players alternate turns, setting one stone at a time onto any vacant point. Once placed, a stone never moves. It can only be removed if it is captured. Capture happens when an opponent's stones surround a single stone or a connected group on all directly adjacent points, leaving it without any open neighbor. Those open neighbors are called liberties, and a group's survival depends entirely on keeping at least one. The ultimate goal is not to eliminate the opponent's pieces but to fence off more empty territory than the opponent by the time both players pass consecutively. Counting at the end involves tallying each player's surrounded empty points, adding captured stones, and adjusting for komi, the compensation points awarded to the white player for moving second. The concept of komi arose during the 20th century precisely because playing first carries a meaningful advantage. Under Japanese rules white receives 6.5 points; under Chinese rules the figure is 7.5. That fractional half-point matters because it prevents a tie. The minimum condition for a group of stones to stay alive permanently is possessing two enclosed empty points called eyes. A formation with two eyes cannot be captured because an opponent would need to fill both eyes simultaneously, which the rules forbid. A group unable to form two eyes is dead and will eventually be removed. Much of every game is a calculation, often conducted many moves in advance, about which groups will live and which will die.

• 3. Origins and Cultural Prestige

  ›

  The earliest written reference to Go appears in the historical annal Zuo Zhuan, dated to around the 4th century BCE, which itself mentions an event from 548 BCE. The Analects of Confucius and two books by Mencius, writing around the 3rd century BCE, also name the game, calling it yi. Chinese legend credits its invention to the mythical emperor Yao, said to have lived between approximately 2337 and 2258 BCE, who allegedly commissioned his counselor Shun to design the game to discipline his unruly son Danzhu. Other theories link the game to military commanders who arranged stones on maps to represent troop positions. Whatever its exact origin, Go achieved remarkable cultural standing in China. It was counted among the four cultivated arts expected of a Chinese scholar-gentleman, alongside calligraphy, painting, and playing the guqin, a stringed musical instrument. The board originally carried a 17-by-17 grid. By the time of the Tang dynasty, which ran from 618 to 907 CE, the 19-by-19 grid had become standard. Go reached Korea sometime between the 5th and 7th centuries CE and was embraced by the upper classes. A Korean variant called Sunjang baduk had developed by the 16th century and remained the dominant form there until the late 19th century. Japan received the game in the 7th century CE. It caught on at the imperial court in the 8th century and had spread to the general public by the 13th century.

• 4. Japan's Formalized Schools

  ›

  In 1603, Tokugawa Ieyasu re-established Japan's unified national government. That same year he appointed the strongest player in the country, a Buddhist monk named Nikkai, born Kano Yosaburo in 1559, to the post of Godokoro, or Minister of Go. Nikkai renamed himself Hon'inbo Sansa and founded the Hon'inbo school. Several rival schools appeared shortly afterward. The four houses, Hon'inbo, Yasui, Inoue, and Hayashi, received official recognition and state subsidies, allowing players to devote themselves entirely to studying the game. Each year, players from the four schools competed in castle games held in the presence of the shogun. The dan and kyu ranking system, which martial arts would later adopt, grew directly from this period of institutional competition. The most coveted honor was the title Kisei, or Go Sage. Only three players across the entire era were judged worthy of it: Dosaku, Jowa, and Shusaku, all of the Hon'inbo house. After the Tokugawa shogunate ended and the Meiji Restoration period followed, the Go houses gradually dissolved. In 1924, the Nihon Ki-in, the Japanese Go Association, was established to carry the game forward. The player Go Seigen, born in China as Wu Qingyuan, won roughly 80 percent of the newspaper-sponsored matches that were prominent in the early 20th century, beating most opponents down to inferior handicaps. He and Minoru Kitani are jointly recognized for developing new opening theory known as Shinfuseki.

• 5. The Spread Westward

  ›

  For most of its long history, Go remained almost entirely within East Asia. Although the game appears in Western writing from the 16th century onward, it did not begin to gain a foothold in Europe until German scientist Oskar Korschelt wrote a treatise on it at the end of the 19th century. By the early 20th century Go had spread across the German and Austro-Hungarian empires. In 1905, Edward Lasker encountered the game in Berlin. When he later moved to New York, he co-founded the New York Go Club with Arthur Smith, who had learned the game in Japan and published a book on it in 1908. Lasker's own book, Go and Go-moku, appeared in 1934 and helped carry the game to a wider American audience. The American Go Association was formed the following year, in 1935, and the German Go Association followed in 1937. World War II interrupted most Go activity outside Japan, but the postwar decades saw steady expansion. The Japan Go Association published an English-language magazine called Go Review through the 1960s and established Go centers across the United States, Europe, and South America. In 1978, Austrian Manfred Wimmer became the first person from outside East Asia to receive a professional player's certificate from an East Asian Go association. In 1996, NASA astronaut Daniel Barry and Japanese astronaut Koichi Wakata played a game aboard a spacecraft, becoming the first people to play Go in space; both were awarded honorary dan ranks by the Nihon Ki-in. By 2000, American Michael Redmond had achieved the rank of 9 dan, the first Westerner to do so.

• 6. Tactical Depth

  ›

  Three classic capturing tactics illustrate how much complexity lives inside Go's simple rules. The first is the ladder, sometimes called a running attack, in which one player uses a relentless series of threats to force an opponent's stones into a predictable zigzag across the board; unless the fleeing stones connect with friendly pieces further along, they cannot escape. The second is the net, known in Japanese as geta, a looser encirclement that prevents escape in every direction without the tight sequential dependency of a ladder; a net does not fail if opposing stones lie in the path. The third is the snapback: a player allows a stone to be taken, then immediately recaptures the point, swallowing the larger group that moved in to take the bait. Beyond these specific techniques lies the broader discipline of reading ahead. The strongest players can calculate sequences up to 40 moves deep even in complicated positions. Practice material known as tsumego, or life-and-death problems, trains exactly this skill by presenting a group under threat and asking the student to find the sequence that either saves it or kills the opponent's group. Ko fights add another layer of tension. A ko is a position where a single stone has just been captured and the opponent's stone now sits vulnerable to being taken back in exactly the same way, which would recreate the previous position and potentially loop forever. The rules forbid immediate recapture, so the player who wants the ko back must first threaten something important enough to force a response elsewhere on the board, freeing the ban. The evaluation of how many ko threats each player holds and how large each one is ranks among the most demanding calculations in the game.

• 7. AlphaGo and the Computer Era

  ›

  Go resisted computer mastery far longer than chess. Before 2015, even the most advanced programs could reach only amateur dan level on a full-size board, though they performed better on smaller 9-by-9 and 13-by-13 grids. The core problem was scale: on most turns in Go there are roughly 150 to 250 legal moves, compared to an average of 37 in chess. For a program to consider all combinations through just eight moves ahead, it would face more than 512 quintillion possibilities. Even the world's most powerful supercomputers could not search that space in practical time. The difficulty of evaluating a position made heuristic shortcuts unreliable; a 6-kyu amateur can glance at a board and gauge who leads, but translating that judgment into an algorithm proved elusive. It was not until August 2008 that a computer first beat a professional player in a game where the human started with a 9-stone handicap, the largest normally given. The program was called Mogo, and it achieved that result at the US Go Congress. By 2013 the required handicap had shrunk to four stones. Then in October 2015, Google DeepMind's AlphaGo defeated Fan Hui, the European champion and a 2-dan professional, five times out of five on a full 19-by-19 board with no handicap. AlphaGo learned almost entirely through deep learning, playing hundreds of millions of games against itself rather than relying on rules handed down by programmers. In March 2016, AlphaGo faced Lee Sedol, a 9-dan player widely considered the strongest in the world at the time. Lee and other top professionals expected him to win. AlphaGo defeated him in four of the five games. Lee's one victory in game five remains the last time a professional defeated AlphaGo in a formal match.