Changing the rules

Nothing courts us so nimbly as technology. Perhaps the chess computers have already won you over — I am dazzled by the riches they have revealed. For a jaw-dropping sense of wonder, try playing over a forced mate in 549 moves.

Still, many yearn for a simpler time. A time when the mysteries of chess were plentiful, but studying the game ‘by hand’ yielded steady nourishment. A time when chess engines had not yet tempted us with their very own tree of the knowledge of good and evil. These days, that tangle of variations and verdicts is irresistible, but who can look at their own play without a sense of shame? Meanwhile, professional players must work harder than ever to harvest fresh ideas. Alas, spectators are underwhelmed when the consensus of optimal moves in a well-worn opening boils down to a sterile equality.

What if we could tweak the rules slightly? Ideally, we preserve the game’s natural tension and depth, but subvert the most familiar patterns of play, so players are thrown on their wits from the first moves. This is delicate, as the rules of chess have been tuned over many centuries; it is hard to foresee whether changing them would unbalance the game. That is where technology promises us a path back to paradise.

DeepMind, the artificial intelligence company, released a paper last week: ‘Assessing Game Balance with AlphaZero: Exploring Alternative Rule Sets in Chess’. AlphaZero is the ground-breaking chess engine which reached superhuman levels by learning from games played against itself. That is the crucial feature which makes it suited to ‘exploring alternative rule sets’. The same algorithm, learning from the ground up, can be applied to any variant of chess. In a matter of hours, the researchers could generate countless high-quality games in their chosen variant, which could then be mined for insights.