How chess is helping science to add to the sum of human happiness
“Chess, like love, like music, has the power to make men happy.” These words, by the great German grandmaster Siegbert Tarrasch, sum up the enduring fascination of the game for humanity. But chess has also played a crucial role in the development of artificial intelligence. Some of the greatest names in the field — Babbage, Turing, Shannon and von Neumann — used chess to test the ability of computers to emulate the human mind. During the Cold War, the chessboard became the arena for the superpowers to compete in a virtual war that avoided the madness of mutually assured destruction.
Now, however, chess is once again the means by which we are extending the frontiers of AI. AlphaZero, a computer programme devised by the London-based company DeepMind, has defeated the leading commercial chess engine, Stockfish, by a large margin. What makes this remarkable is that AlphaZero achieved this superhuman level of play within 24 hours, learning from scratch just by playing chess, without any human intervention beyond the basic rules of the game.
What makes AlphaZero even more extraordinary is the fact that it did not win by sheer number-crunching: in fact it searches a thousand times fewer positions per second than its rival. Instead, the programme learns by much subtler means that more closely resemble human thought. According to the scientists at DeepMind, “AlphaZero compensates for the lower number of evaluations by using its deep neural network to focus much more selectively on the most promising variations — arguably a more ‘human-like’ approach to search.”
The raw statistics — in over 1,000 games against Stockfish, AlphaZero won 155 and lost 6 — surprised seasoned experts. But what is now emerging is not just the quantity but the quality of the chess being played by the new programme. Hitherto, computers have been notorious for playing very dull games, although ever since Garry Kasparov lost to Deep Blue in 1997, human players have gradually imitated computers. AlphaZero, by contrast, plays exciting chess that is not merely superior to anything seen before but entirely original.
According to the British grandmaster Matthew Sadler, who is writing a book about AlphaZero, “the games are superb. They are really, really superb.” What most impresses Sadler is the fact that the computer programme is not materialistic: “It’s not afraid to give up material,” he told the science correspondent of The Times, Tom Whipple. “It might sacrifice a couple of pawns just to get an open file against the king. These are long-term sacrifices. It can’t tell when it’ll work out. But it knows that it’ll pay off.”
As Whipple comments, human grandmasters will soon be copying AlphaZero, just as they have copied conventional programmes such as Stockfish. The world championship match between Magnus Carlsen and Fabiano Caruana, played in London last month, resulted in all 12 “classical” games being drawn. It had to be resolved with a playoff at rapid speed limits, which gave Carlsen the edge. But if both players had been more influenced by AlphaZero’s sacrificial style, they might have taken more risks. This would have been more entertaining for the millions of fans who followed the match online.
The real significance of this latest breakthrough, however, goes far beyond chess. We are on the verge of creating machines that will be able to learn and think like us. Complex political challenges, such as how to resolve the impasse over the Irish backstop in the Brexit negotiations, at present baffle our politicians and diplomats. Yet they might be child’s play if we could make use of artificial intelligence.
It is good news that AI need not necessarily be materialistic, but can take account of the intangibles of human existence. Like love, like music, chess can indeed make us happy — especially if it is the means by which humanity can harness the most precious form of power there is: intellectual power.
