American Go E-Journal » Computer Go/AI

DeepMind has just published extensive new commentaries on the historic AlphaGo-Lee Sedol match, played earlier this year in Seoul. Fan Hui 2P, who first faced AlphaGo in October 2015, has teamed up with Gu Li 9p and Zhou Ruiyang 9p to conduct “exhaustive analysis” not only of the five games between AlphaGo and Lee Sedol, but of three games AlphaGo played against itself shortly before the match. The commentaries provide both analysis of the moves as well as insight into AlphaGo — and its team — behind the scenes, including the AI’s realtime assessments and tidbits such as “it is clear from AlphaGo’s data that it prefers White.” For anyone who watched the games in March, these commentaries provide a fascinating opportunity to see them with a fresh eye.

“We found its ideas both exciting and inspiring, and it became clear to us that AlphaGo represents not only a scientific and technological advancement, but also a milestone in human understanding of Go,” says Fan. “Unconstrained by human biases, and free to experiment with radical new approaches, AlphaGo has demonstrated great open-mindedness and invigorated the game with creative new strategies…AlphaGo has created a unique and extremely powerful approach to the game of Go.

Noting that “no one strategy can guarantee a player’s success,” Fan adds that “learning from these games is sure to have a positive, enlightening impact on one’s Go strength and style.”
– Chris Garlock. With Michael Redmond 9P, Garlock co-hosted DeepMind’s English game commentaries on the AlphaGo-Lee Sedol match. 

by Andrew Feenberg

This is the story of two go matches. In March, Google’s artificial intelligence computer AlphaGo defeated world champion Lee Sedol 9P 4-1. This is considered a great triumph for artificial intelligence. Does it mean that AI is almost human? Or that humans have simply become machine-like? Another historic match in 1938 provides some useful perspective.

Games are ambiguous phenomena. On the one hand they are models of rationality. There are unambiguous rules, actions, and measures of success and failure. On the other hand games are social events, collaborative performances governed by meta-rules such as sportsmanship that contextualize the play. The players need each other in order to perform. They produce an object together through their combined efforts. The effect of this activity is to contribute to their own personal development as human beings. And we often qualify the product of their activity in aesthetic terms: a beautiful play in football for example.

Rationality as found in games is also a general attribute of the culture of modern societies. Modern societies contain many game like systems. Individuals act as players in these systems in order to get a raise, get a job, learn new skills, take a plane, avoid paying taxes, and so on. Thus games can appear as models of modernity and what happens with games can tell us something about what happens in modern societies in general.

This is why the Nobel prize-winning Japanese author Kawabata chose the game of go for one of his most famous novels. “The Master of Go” is based on a real match that Kawabata witnessed as a newspaperman in 1938. This was the last match of the old master Shusai and his challenger, called Otake in the novel. Each player represents an era. The old master represents traditional play, while his challenger represents modernized play, influenced by Western ideas. Traditional play is all about the human side of the game, deference, the notion of way or vocation as practiced in Japanese martial arts for example, and beauty, the beauty of the moves and the board as the game evolves. Modernity is about winning. Just that.

The novel recounts the game’s play as it reveals the characters and the historical background. The whole intrigue centers on a single move in the game, move 121. The reason this move is so important has to do with Western innovations introduced by the challenger and favored by the newspapers. In the old days the master would have regulated the flow of play, deciding when to start and end each day’s session. But the 1938 match was organized around time limits and sealed moves at the end of the day as in Western chess. Of course players are equal in the game play, but the traditional way of playing recognized the differences between the players in the world outside the game. Now the equalization internal to the game was being extended outward into the world of play. One can see how this might appeal to the challenger and to the newspapers, both of whom would prefer not to be subject to the whims of the old master. To the old master it seemed a lack of due deference.

As it happened, the challenger began to run out of time toward the middle of the game. To gain time to reflect on a difficult position in the center of the board he sealed a final move off in a corner. This trivial move required the master to respond away from the central struggle. The manipulation implied in this use of the time limits and sealed moves upset the master. He was offended and said that the beauty of the game was lost. As a result he made several mistakes and the challenger won. Modernity triumphs over tradition. But we are left with the clear impression that the old master was the better player of the two.

Kawabata wrote after World War II that he would only write elegies, elegies for the lost beauty of the old Japan. The novel is obviously a critique of modernization. But note that it is not about the contrast between rational modernity and the irrational tradition. There is nothing strategically irrational or inferior about the old master’s play. So the contrast of modernization and tradition is not about strategic rationality vs. irrational sentimentality. It is about the place of strategic rationality in the real world. Tradition is a set of meta-rules that contextualize and organize the rationalized sectors of social life such as games. Modernity extends the rationality of games into the real world. This has counterintuitive consequences. For example, in the case of the actual match the inferior player wins through manipulating the new meta-rules and upsetting his adversary rather than through superior play.

Here is the how Kawabata described his elegy for traditional go: “It may be said that the master was plagued in his last match by modern rationalism, to which fussy rules were everything, from which all the grace and elegance of go as art had disappeared, which quite dispensed with respect for elders and attached no importance to mutual respect as human beings. From the way of go the beauty of Japan and the Orient had fled.”

The losers of the two matches were Shusai and Lee Sedol. The winners were Otaké and AlphaGo. What do they have in common? If we understand games in all their ambiguity as both strategic exercises, rational systems, and also ways of self development and aesthetic achievement, then I would suggest that their reduction to mere winning is a disaster. The matches reveal the limits of modernity as a way of understanding and organizing human life. In reorganizing the social world around strategic rationality, modernity prepares the triumph of the machine.

This essay is based on a talk presented at the McLuhan Centre at the University of Toronto in early March. It has been condensed and updated. Feenberg has written extensively about Kawabata’s novel previously in his book Alternative Modernity.

Some years ago, Dutch computer scientist John Tromp (right) and his colleagues reported that the longest possible game of go would last longer than the universe is likely to. Now, the culmination of their work on Go Combinatorics—the science of counting—is reported on in Peter Shotwell’s latest essay, “John Tromp and the Big Numbers of Go: The Possible Positions, Games and the Longest.” One finds that the number of possible games “is more than a googolplex, which is a number that cannot physically be written down in the available space of the universe.” Also, with highly advanced computer power since that last report, Tromp and his team finally found the exact 171-digit number for 19 x 19 boards. “A 13×13 board has as many possible positions as the approximately 10⁸⁰ atoms in the universe and the ‘L19’ number is more than 10⁹⁰ times greater than this.” In other words, as one science writer quipped, “Saying that there are more go positions or games than atoms in the universe is like saying the national debt is more than a penny.”

What the AI Behind AlphaGo Can Teach Us About Being Human: AJA HUANG DIPS his hand into a wooden bowl of polished black stones and, without looking, thumbs one between his middle and index finger. Peering through wire-rim glasses, he places the black stone on the board, in a mostly empty zone, just below and to the left of a single white stone. In Go parlance it is a “shoulder hit,” in from the side, far away from most of the game’s other action. Across the table, Lee Sedol, the best Go player of the past decade, freezes. He looks at the 37 stones fanned out across the board, then stands up and leaves…Read the rest of Cade Metz’ report in Wired. photo by Geordie Wood

A glimpse inside AlphaGo? “Here’s a picture of the machine Google used in the match against Lee Sedol,” writes Steven Schmeiser. “It turns out that they were using custom designed chips that are optimized for machine learning.”
Google supercharges machine learning tasks with TPU custom chip

Go Commentary: Lee Sedol vs AlphaGo – Game 1: If there’s any recent game that needs no introduction, it’s this one. On March 9, 2016, the computer Go program ‘AlphaGo’ defeated Lee Sedol 9p in the first game of the Google DeepMind Challenge Match. Go Game Guru’s Youngil An takes a look at the game.

Lee Sedol overtaxed? In a related story, Gordan Castanza reports that “I just learned from KBS News (Korean Broadcast System) this morning that Lee Sedol has left the Korean Baduk Association over the issue of its imposing a 20% ‘tax’ on him.” Stay tuned for more details as they become available.