Friday, December 11, 2020

"How a chess grandmaster tried to outwit the computer"

From Prospect Magazine, November 13, 2020:

When artificial intelligence began beating the world’s greatest players, a chess grandmaster devised his own way to give human ingenuity an upper hand against the machine. The result, however, was not quite what he expected

On Sunday 23rd July 1972, the American grandmaster Bobby Fischer made the first move of the sixth game in the world chess championship—shunting his pawn two squares up the board.

Nothing, in itself, was unusual about that. Pushing either of the middle “d” or “e” pawns two squares forward is the most common way to begin a game. But this move involved neither of these pawns and took Fischer’s opponent—the reigning world champion Boris Spassky—by complete surprise. Moreover, because he had not expected it, he had not prepared for it.

Fischer began with square c2 to c4—the “English Opening” (so called because it was a favourite of a 19th-century English chess champion, Howard Staunton). To those who don’t follow chess, it might sound a comically small twist—the same move, just one or two spaces along. But it shook everything up, and shook Spassky up in particular. During the months he had been in training, the indolent Russian had pooh-poohed the notion that he had to be ready to respond to all of white’s opening options. Fischer almost unfailingly played e4. Surely he would not unleash a new opening in the most important match of his lifetime? It’s not easy to think of analogies, but imagine a fast bowler in cricket suddenly bamboozling the batsman with an over of leg-spin.

With both sides in unfamiliar territory, the game itself proved the most beautiful of the championship. After resigning, Spassky joined the spectators in applause at his opponent’s brilliance. Fischer was now ahead in the match; six weeks later he would be crowned the 11th world champion.

A quarter of a century on, Fischer called a shock press conference in Argentina. Since his headline-grabbing battle with Spassky, the American genius had become a recluse. In the past he’d been described as “troubled,” “turbulent,” “mercurial,” and had engaged in crude antisemitism despite being of Jewish descent; it was now clear that he’d tipped into paranoia. He’d resurfaced from isolation in 1992 to play a rematch against Spassky in war-torn former Yugoslavia, in defiance of US sanctions. After winning, Fischer disappeared yet again, this time as a wanted criminal.

The 1996 Buenos Aires press conference was packed. In his meandering remarks, Fischer denounced the arrest warrant against him and complained that he’d been denied payments from various books and films that supposedly exploited his name. But eventually, he got to his point: the promotion of a new type of chess, Fischer Random, which built-in far twistier twists than his celebrated opener in 1972.

This game would be like ordinary chess in most respects. Each side would have eight pawns, arrayed on the second (white) and seventh (black) ranks. Each side would have two rooks, two knights, two bishops, a king and a queen. The pieces would move as before, and the object of the game would still be to checkmate the other side. But there would be one radical -departure: the pieces on the back ranks would be ordered—or maybe that should be disordered—randomly.

For what reason? Well, four months earlier the IBM computer, Deep Blue, had taken on the world champion Garry Kasparov. Deep Blue had humiliated Kasparov in the first game, and although it lost the series, it was clear that the era of man’s superiority over the machine was approaching its end. In 1997, Kasparov would be crushed by a new and improved Deep Blue.

One might have expected Fischer to take some schadenfreude from Kasparov’s struggles against the chess supercomputer. Fischer was a child of the Cold War, and despite the collapse of the Soviet Union five years earlier, he retained an enduring conviction that the Russians were cheats, frauds and schemers. During the Argentine press conference, he defamed his two successors, Kasparov and Anatoly Karpov—their games against each other were fixed, he said. If supposed Russian rigging were the problem, then Fischer Random could have helped: when you have no idea what the set up of the pieces is in advance, collusion becomes impossible.

But the tilting of the scales against humans of all nations was even more of an affront to Fischer. Computers, he grumbled, had an unfair edge. No human could memorise the millions of opening variations that programmers could simply enter into Deep Blue’s database. Without that advantage, he insisted, human creativity could still vanquish any silicon wannabe. His aim, then, was to provide an answer of sorts to the creeping digital dominance of the game.

Twenty-four years on and Fischer Random, though still a minority pursuit, grows ever-more popular: you can buy chess clocks that double-up as gadgets that shuffle the starting order of the pieces around. For ordinary fans, the appeal is simple: the variant rescues the top-level game from what had increasingly become a struggle between human databases.

With the assistance of chess software engines, today’s top players can spend hours on openings each day, endlessly analysing innovations that have been made in games by others, becoming encyclopaedias of past play. “It’s a lot to keep up with,” says Britain’s leading player, Michael Adams. If that’s exhausting for them, it’s also deadening for those who watch—it can mean it takes 15 or 20 moves before any novel position appears. Indeed, some games now conclude before one or even both of the players are out of their rote-learned preparation. When the player on each side of the board is going through a drill, there is little drama, and the upshot, far too often, is a crowd-displeasing draw.

Look only at how many people are playing chess, and it seems as popular as ever—there have not been many winners from the Covid-19 pandemic, but with millions stuck at home the online game has boomed. On the website chess.com, there were 204m games between humans in February 2020, but 323m by June, growth of over 50 per cent in those few locked-down months. Still, there is a nagging sense that there is something missing in the spirit of the game, particularly at the top, which has sparked many different ideas to revive it. The AI company Deep Mind has been analysing various radical options, assessing the permutations and whether potential new laws could create a dynamic but balanced game. One mooted idea is that “castling,” the manoeuvre that allows you to shuffle around a king and a rook in a single move, should be abolished. Another—which opens up what to chess players would seem like almost psychedelic strategies—permits players to capture their own pieces.

But among many weird variations, Fischer Random remains the front runner, because—by subjecting the starting position to the luck of the draw—it directly attacks the curse of over-preparation in the database age. The alien piece arrangement can flummox players from the very first move. The long years in which a grandmaster has deepened his (and it is usually “his”) knowledge of the Ruy Lopez, the Sicilian Najdorf, the Nimzo-Indian or any other openings for white and black suddenly count for nothing. All the cognitive sweat from memorising innumerable opening lines yields no advantage. The thousands of hours top players put into opening training and development are redundant: what matters is raw talent....

....MUCH MORE