Chess is a Draw

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
hgm
Posts: 27930
Joined: Fri Mar 10, 2006 10:06 am
Location: Amsterdam
Full name: H G Muller

Re: Chess is a Draw

Post by hgm »

lkaufman wrote: Wed May 15, 2024 12:45 amChanging the draw rules will not "delay the inevitable". With the rules I proposes, there would be no draws, every game would be decisive. The risk is that one side may have a clear advantage, which at top engine level might translate to one color always winning. But if we get the rules just right, the practical chances should be so close to equal that this won't be an issue until the entire game tree is mapped out, which presumably won't happen before the end of life on earth, though you never know. Basically, we have this situation in shogi now, where it isn't yet clear whether the first player wins or the result is a draw with perfect play, though recently evidence has shifted towards the "first player win" side of the argument. With the right rules, chess could be right on that narrow edge where no one will know the correct result even in a thousand years (probably).
In mini-Shogi draws are not possible: on repetition (other than perpetual checking) sente (the player that had the first move) loses. This appears to create a disadvantage much larger than the first-move advantage, though, so the game is lost for white, and I believe that current top engines play good enough to always convert that win.

There are several historic Shogi variants where repetitions are forbidden, (e.g. Tori Shogi,, Tenjiku Shogi), but it is often not clear which player has the burden to deviate. (I.e. whether repeating a position is forbidden, or making the same move from a repeated position), and how perpetual checks are treated.

In general it is considered very unsatisfactory if you can force a win through perpetual check, so most games with unambiguous rules that forbid repetition make some provision that makes the checker lose in that case. No matter how other repetitions are treated.