Chess should get an upgrade

It's okay to update rules over time

Different sports have different attitudes to rules changes. Most competitive sports get constant rules tweaks (except for baseball, which seems dead set on achieving cultural irrelevance). Track and field tries to keep consistent rules over time so performances in different years are comparable. Poker has a history of lots of experimentation but has mostly been a few main variants the last few decades. Go never changes the rules at all (except for tweaking the scoring system out of necessity but still not admitting it. That’s a story for another day). Chess hasn’t changed the rules for a long time and it’s a problem.

I’m writing this right after a new world chess champion has been crowned. While the result was better than previous championships in that it succeeded in crowning an unambiguous world champion, it failed at two bigger goals: Being exciting and selecting a winner who’s widely viewed as demonstrating that they’re the strongest among all the competitors.

The source of the lack of excitement is no secret: Most of the games were draws. Out of the 14 games, 5 were decisive. This also creates a problem for the accuracy of the result. In a less drawish game a match with 6 games of which only one was drawn would have equal statistical significance (assuming it wasn’t overly partisan). In fact it’s even worse than that appears on its face. The candidates tournament, which selected the challenger, was a 7 person double round robin where the winner scored 9/14 and three other competitors scored 8.5/14. Picking a winner based on such a close result has hardly any significance at all, and that format means that unless one of the competitors was ludicrously better than the others such a close result was expected. If the format were instead that the top four finishers from the candidates tournament played single elimination matches against each other then the eventual result would be viewed with far more authority. Partially the problem here is that this is just a badly designed tournament but some of the reasoning behind this format is because of the drawishness. Such matches would be long and arduous and not much fun, as they were in the past. Some previous FIDE title tournaments were far worse, following a very misguided idea that making the results random would lead to more excitement. That makes sense in sports like Soccer where there are few teams and it’s important that all of them have a shot, but the ethos of Chess is that the better player should consistently win, and adding randomness can easily lead to never seeing the same player win twice.

This leads to the question of how the rules of chess could be modified to not have so many draws. Most proposed variations suffer from being far too alien to chess players to be taken seriously, but there are two approaches which are, or should be, taken seriously which I’d like to highlight: Fischer random and the variants explored by Kramnik. In Fischer random a starting position with the pieces in the back rank scrambled randomly is selected. In the latter Kramnik, a former world champion, suggested game variants, and the Deepmind team trained an AI on them and measured the draw rate and partisanship of all of them based on how that engine did in self-play games. (Partisanship is the degree to which the game favors one player over the other). I love this methodology. Of the variants tried I’d like to highlight two of the best ones. One is ‘torpedo’, in which pawns can move two squares at once from any position, not just the starting one. The other is no-castle, which is exactly what it sounds like. No castle has the benefit that it gets rid of the most complex and confusing chess rules and that it’s just changing the opening position, in fact it’s a position reachable from the standard Chess opening position. (For some reason people don’t do no-castle Fischer Random tournaments, which seems ridiculous. Might as well combine the best ideas.)

Both no castle and torpedo have about the same level of partisanship as regular chess, which may be a good thing for reasons I really ought to do a separate blog post about. They also both do a good job of making the game less drawish. The reason for this is, in my opinion, basically the same, or at least two sides of the same coin. Torpedo makes the pawns stronger, so they’re more likely to promote and decide the game. No castle nerfs the king so it’s more likely to get captured. Of these two approaches torpedo feels far more alien to regular chess than no castle does. My proposal to make Chess even more non-drawish is to nerf the king even more: Make it so the king can’t move diagonally. Notably this would cause even king versus king endgames to be decisive. It would also result in a lot more exciting attacks because the king would be so poorly defended. This needs extensive play testing to be taken seriously, but repeating the Deepmind experiment is vastly easier now that AI has come so far, and it would be great if Chess or at least a very Chess-like game could have a world championship which was much more exciting and meaningful.