Someday, a computer will solve chess. Here's what we can do.

Someday, a computer will solve chess. Here's what we can do.

Jun 13, 2010, 7:00 AM |

Mathematically speaking, chess is an extraordinarily complex game. The numbers of possible positions has been estimated to be at least 10 to the 43rd power. Because of that, in the short run we need not fear that chess will be "solved" by computers (as has happened to similar games, like checkers and very small games of Go). However, computers are getting more and more powerful every year. One might guess that someday in the very distant future (hopefully not in any of our lifetimes!) the game of chess will be solved, and a computer program will be devised which can get, say, at least a draw in every game. What should chess players do if this happens?

The most obvious step is not to panic. That a computer has become perfect at chess has little to do with the game's enjoyability for an average player (after all, people still play checkers); what's more, the immense complexity of chess means that we can be almost certain that no human will ever be able to mentally "solve" chess as a computer might. Human tournaments would thus be quite like those we have now (I'll go into the major difference in a moment), because even grandmasters or chess savants could only hope to learn a tiny portion of the computer's "solution" of chess. Computers could even be allowed, as they are now, to compete in some tournaments--given the immense processing power that would be required to "solve" chess, it seems unlikely that your typical tournament director could afford to buy or rent a supercomputer powerful enough to run the chess-solving program.

I did say that there would be some changes to tournament chess. Grandmasters and their seconds use computers to prepare themselves and analyze games (in fact, they even use supercomputers at times, as Topalov did before his recent match with Anand). The existence of a computer that's "solved" chess would mean that a grandmaster could settle, once and for all, the right response to some particular opening line. The solution of chess would provoke a scramble among opening enthusiasts to both examine how the computer plays its own openings and examine how the computer evaluates man-made lines. Thus, opening systems could be deepened and perfected, adding to the modern game's already immense emphasis on opening preparation. However, as has been said, humans can only absorb a certain tiny portion of the "solutions" to chess, so the strategy of preparation would not change so much. Suppose, for instance, that Spassky had access to the chess-solving computer for his championship match against Fischer. Spassky could indeed have prepared a wealth of pitch -perfect refutations to Fischer's traditional 1. e4. However, even a master could not hope to prepare with such depth for all possible openings, so it seems inevitable that Fischer's famous 1. c4 game would still have been a brilliant, game- changing surprise attack.

What's more, it's possible that the computer openings could rely on those elusive "computer moves," the characteristic ability of computers to make brilliant moves that would not be obvious to human thought-patterns. A computer opening may indeed be the path to a win that's theoretically more sure than any other, yet which requires a few "computer moves" in the early middlegame to ensure the avoidance of a total disaster; thus a computer opening might be a bad idea for a human player.

Of course, when a computer solves chess, we can also change chess, so that it is in some different form, buying us some time in which the game is again unsolved. This seems to be utterly unpleasant--after all, chess is a very traditional game, with a history stretching back thousands of years. Adding a few fairy pieces or changing the size of the board just so we can say for a few weeks that chess is still unsolved seems ridiculous. However, we can attempt, in our changes, to add to the complexity of the game without changing the substance of the game. There are numerous minor rule changes which would add to the number of calculations a computer must make, while being invoked so rarely in normal play that the rule change would have almost no effect on day-to-day chess play.

The first idea that comes to me is allowing pawns to promote to a piece of the opposite color. The number of situations in which this is advantageous is tiny. The only ones I'm aware of are those in which the promoted piece blocks an escape square for the king which the attack doesn't cover--thus a same-color promotion would result in a capture, but an opposite-color promotion (often this works with promotion to a knight) becomes a checkmate. See the diagram.




We could also allow a pawn to advance to a promotion square and then "promote" itself to a pawn. The pawn would then be utterly useless, sitting on the back rank and controlling nothing. I suppose this would only be useful when stalemate would result from any other promotion (though off the top of my head I can't think of a circumstance in which a stalemate could be avoided by promoting to a pawn that couldn't be avoided by promoting to a knight).

In this vein, we could allow a player to convert their queen to a pawn (perhaps irrevocably). Except when avoiding stalemate, this would seem to be utterly useless, yet it'd be one more move for the computer to rule out before each turn, thus making solving chess harder.

We could also allow a few useless king moves. My first idea is a 21st legal opening move, 1.Ke2, in which the king and the e-pawn switch places. The king has now lost his right to castle and is exposed in the center, while also blocking the e-pawn's advancement (thus delaying center control) as well as the development of the queen and the king's bishop. Allowing similar moves 1.Kd2 and 1.Kc2 would seem to have a somewhat lessened negative impact. Allowing the king and queen to switch positions on the first move (1.Kd1)would also seem to be useless--it would allow you to play a Scandinavian against a d-pawn opening, and make giving early check with the queen in the process of a pawn capture easier (that could indeed be a substantive advantage, allowing a second pawn to be captured), but it would effectively mean that you're playing as Black even when you're White.

A final idea would be the restoration of file castling. File castling apparently used to be allowed--one could, if the king had never moved and a pawn promoted to a rook in the e-file, castle 0-0-0-0, moving the king to e3 and the rook to e2 (in the case of white). This would have some uses in endgames, if one is trying to get the king into position to assist an advanced pawn structure, but it would otherwise do little beyond centralizing the king.




Rule changes like these would buy us a little time against the computers, turning them from chess solvers to mere extremely strong engines, but they would make the game different only in rare situations outside the world of chess composition. Thus, we could stave off the "end" of the beautiful boardgame by a few months or years. However, chess will not be over as long as humans aren't, a la science fiction, given computer-like mental abilities. Just as the invention of faster means of transit like the automobile didn't destroy interest in running races, the rise of the ultimate chess computer will not destroy human chess.