From: https://onezero.medium.com/how-a-i-put-the-humanity-back-in-chess-401a3f38d207
How DeepMind Restored the Beauty to Chess
Chess masters feared that computer programs would suck the artistry out of the game, but the company’s AlphaZero plays with soul
Follow
Jan 10 · 7 min read [An easy, fun read.]
Photo: Anton Novoderezhkin/Getty Images
In the world of professional snooker — a game similar to pool, played throughout the U.K. — there is confusion about the word “tactic.”
Loosely put, there are two kinds of shots in snooker. There are attempts to pocket balls and score points, and there are safety shots where the goal is to make things more difficult for your opponent. A frame of snooker can descend into a protracted exchange of safety play — not particularly compelling to the casual observer — with neither player attempting to pocket balls. Such frames are described as “tactical.” This is wrong. The proper word would be “strategic.”
The world of professional chess has it right, and the difference between tactics and strategy in chess was once elegantly described by Austrian-born grandmaster Savielly Tartakower: “Tactics is what you do when there is something to do; strategy is what you do when there is nothing to do.”
This difference is crucial to understanding — or rather, appreciating — a recent eruption of enthusiasm in the chess world. Last year a computer program called AlphaZero was pitted against another program called Stockfish 8, and in technical terms, AlphaZero ate Stockfish 8’s lunch. The story of the match has a lot to say about computers, people, intelligence, the future — and perhaps surprisingly, beauty, and art.
What is beauty in chess? Let me try to explain, at least to the extent that I came to understand it while writing my first book, The Chess Artist. Most simply put, there are the laws of chess — the rules of the game — and there are the principles that dictate good play. Beauty tends to result when a player violates a principle of good play and wins anyway.
Consider the chess knight. With its L-shaped move, a knight that sits in the middle of the chess board controls or attacks eight other squares. By contrast, a knight that sits in a corner of the board attacks only two other squares. Hence, objectively speaking, a knight in the middle of the board is four times more influential than a knight in a corner.
Thus, Polish grandmaster Siegbert Tarrasch’s principle of good play: A knight on the rim is dim. But if you do move your knight to a corner of the board, and you win because of that move — well, this is the kind of thing that can inspire in chess players the sudden flush of awe and joy that is typically associated with the appreciation of beauty and art.
A few years ago, the A.I. company DeepMind set out to use games to hack human intelligence. They thought that if they could invent a better computing system, one that taught itself how to do things without human bias or preconceptions, they could “solve intelligence and then use it to solve a lot of other things.” First the Asian game Go, and then chess, became laboratories for experiments aimed at broader applications.
DeepMind devised a program that bested the world’s top Go player, South Korean Lee Sedol, in 2016. Next up was chess — though not against human players. Rather, the goal now was to beat the world’s top computer programs or “chess engines,” which had been evolving for decades.
To understand what came next, we need to back up to history — and to strategy and tactics. It’s not a hard and fast rule, but it’s generally held that tactical chess play — the exchanges of pieces, “gambits” or sacrifices, mating attacks and “king hunts” — is more fun, both for players and observers. Aesthetically pleasing, is how chess players put it. Strategic play, by contrast, in which theoretical advantages become decisive over a period of 30 or 40 moves, can look like a whole lotta nothing. Historically, players of this style — Cuban world champion José Raúl Capablanca, and Russian world champion Mikhail Botvinnik — have been described, in a not laudatory way, as cold, mechanical, machine-like.
Here’s where it gets weird. Because it turned out that strategic play was more human than tactical play, after all.
Or that’s how it seemed. When chess engines began to evolve, it was assumed that tactics were creative and intuitive. A cold, mechanical machine might be able to strategize a decent game, but it could never compete with a human player who could bring to bear a more profound level of tactical creativity and ingenuity. When chess engines finally achieved enough brute force computing power, however, it was revealed that tactics were actually pretty simple. If you could calculate deep enough into a position, you could easily “see” whatever nifty tactical opportunities might be hiding there. Tactics were hardly creative at all, in fact. For a time — not very long, it turned out — human players managed to remain competitive with chess engines because, thinking strategically, they could sometimes intuit their way deeper into a position than the engine could calculate.
In essence, chess engines were perfect versions of human players. Every human player. [chess engine, ?] Like Deep Blue, which beat the Russian world champion, Garry Kasparov, in 1997, [was a ?] brute force chess engines combine[ing?] computing power, a comprehensive library of previously played chess games, and an encyclopedia of canonical chess wisdom (a knight on the rim is dim). The one weak spot in their armor was that a chess engine was, ultimately, a program. That is, if there were flaws or mistaken assumptions in the understanding of chess that came from the humans who wrote its code, then the engine shared those flaws and assumptions.
In any event, chess engines soon had little trouble beating the world’s top grandmasters. Not only that, the games that the engines played — against people or other engines — struck most observers as cold and mechanical. Tactics were easy to refute, so the games turned strategic and dull. The chess world worried that their centuries-old game had been solved. “Not human-playable” became a term used to describe positions so far beyond understanding, all you could do was shrug. It looked like the game was dead.
When DeepMind created AlphaZero, it was given absolutely no advance knowledge of chess. No history of known opening moves. No encyclopedia of chess wisdom accrued over a thousand years of human play, as Deep Blue possessed. Just the rules of the game.
In nine hours, AlphaZero played 44 million games against itself, more than a thousand per second, to learn how to play. Then it was pitted against Stockfish 8. Ten games were released to the public at the end of 2017, and dozens more early the next year. The world of chess has not been the same since.
The internet changed chess in countless ways, not the least of which has been a rise in chess players making videos of themselves discussing chess games, talking through the moves of famous games of the past, or games played in recent tournaments. British YouTuber kingscrusher has uploaded more than 9,000 videos for his 104,000 followers; Croatian YouTuber agadmator has 500,000 followers who have viewed his videos nearly a billion times.
These and many other commentators made a host of videos analyzing the AlphaZero-Stockfish 8 games. What is most interesting about the videos is the unbridled enthusiasm that analysts have for AlphaZero’s style of play. At times, AlphaZero makes moves that leave the analysts baffled and mystified, but enthralled as well. They react as if they have stepped into the presence of the work of an artist who represents a breakthrough, a leap forward.
The most exciting thing was that AlphaZero played tactically. It sacrificed pawns and pieces. It moved bishops and queens to the corner of the board. It risked its king in ways no human player would ever consider. This meant a couple things. First, the chess world had been wrong again — this time about having been wrong the first time around. Tactical play really was the best way to approach chess. Second, the principles of good play that had been worked out over a millennium were far from complete — or, put another way, AlphaZero demonstrated that principles regarded as rock-solid could be beautifully defied.
Chess was not dead. It was not dead for the same reason that periodic claims in the literary world about the novel being dead are always wrong. Even though human beings had been unable to calculate all the intricacies of chess, they had somehow known the best way to play. They just needed a chess engine to prove that their intuition, their gut — their muse — had been right all along.
Without doubt, the greatest chess player among artists was Marcel Duchamp. Duchamp famously gave up art for a career in chess. An art exhibition he organized in 1944, The Imagery of Chess — featuring work by Alexander Calder, Man Ray, Robert Motherwell, Yves Tanguy, Dorothea Tanning, and many others — ensured that there would forever be a link between the worlds of chess and art.
This isn’t what was supposed to happen. When DeepMind set out on its quest to crack intelligence with games, it wasn’t thinking about art and beauty — it was thinking about value. Value in health care, energy, macroeconomics, education, robotics. By contrast, Duchamp gave up art for chess precisely because chess didn’t have any value. “Chess has no social purpose,” he said. “That, above all, is important.”
What’s being missed right now in the revolution of machine learning is that the AlphaZero-Stockfish 8 match represents the first time in history that a machine has created something that even experts appreciated not for its value, but for its beauty. Maybe that’s a cause for hope. Maybe we don’t need to fear the Skynet Armageddon or the death of treasured media. Most beauty in the universe is not human-made, and a striking sunset is no less striking for the fact that it appeared without the help of any intelligence at all.
The undercurrents of the future. A Medium publication about tech and science.
Following
16
· Gaming
· Chess
· Deepmind
16 claps
Applause from you and 6 others
WRITTEN BY