Why do programmers care only about elo and not about faster mating?

I think that it is a bad design and the scoring system in computer chess tournament should be changed.

A mate in 40 should give more points than mate in 80.

Possible score for the winner can be something like
0.5+0.5*0.99^number of moves when the score for the loser may be
0.5-0.5*0.99^ number of moves.

I believe that if people optimize NN's for this scoring system then programs based on NN could avoid all the problems they have in handicap matchs.

Uri Blass wrote:

I think that it is a bad design and the scoring system in computer chess tournament should be changed.

A mate in 40 should give more points than mate in 80.

Possible score for the winner can be something like
0.5+0.5*0.99^number of moves when the score for the loser may be
0.5-0.5*0.99^ number of moves.

I believe that if people optimize NN's for this scoring system then programs based on NN could avoid all the problems they have in handicap matchs.

My immediate thoughts (I might add more later)

* chess is a drawn game. Unless your opponent makes a mistake, you cannot win - so in the start position, such an algorithm would have nothing to say

* when a position is winnable, unless you can calculate right the way to the checkmate, there's no way of knowing which move gets to mate the most quickly

* today, chess programs, like humans, have a limited amount of knowledge (for me, the evidence strongly suggests that you could get MUCH more accurate position evaluation with a MUCH smaller program if you uncover chess's deep (complex) patterns, while today's NNs are finding a large number of shallow (simple) patterns - but that discussion is not for this thread). Expecting them to be able to find the fastest way to a win in an astronomically large number of positions might be wishing for more than is possible with today's tools

TheSelfImprover wrote:

Uri Blass wrote:

I think that it is a bad design and the scoring system in computer chess tournament should be changed.

A mate in 40 should give more points than mate in 80.

Possible score for the winner can be something like
0.5+0.5*0.99^number of moves when the score for the loser may be
0.5-0.5*0.99^ number of moves.

I believe that if people optimize NN's for this scoring system then programs based on NN could avoid all the problems they have in handicap matchs.

My immediate thoughts (I might add more later)

* chess is a drawn game. Unless your opponent makes a mistake, you cannot win - so in the start position, such an algorithm would have nothing to say

* when a position is winnable, unless you can calculate right the way to the checkmate, there's no way of knowing which move gets to mate the most quickly

* today, chess programs, like humans, have a limited amount of knowledge (for me, the evidence strongly suggests that you could get MUCH more accurate position evaluation with a MUCH smaller program if you uncover chess's deep (complex) patterns, while today's NNs are finding a large number of shallow (simple) patterns - but that discussion is not for this thread). Expecting them to be able to find the fastest way to a win in an astronomically large number of positions might be wishing for more than is possible with today's tools

I do not expect chess engines to solve chess and to always find the the fastest way to win but only to play for the target of mating faster.
All the training of NN are based on the idea that mate in 40 is the same as mate in 80 and I do not like it.

Here's a revolutionary idea (if I say so myself)!

How about... instead of evaluating a position in terms of pawn value, which is roughly what most chess engines do, the number an engine outputs is the number of moves to checkmate?

This would focus the minds of programmers on getting their engines to REALLY understand chess!

Uri Blass wrote:

I think that it is a bad design and the scoring system in computer chess tournament should be changed.

Already done by me. Look here. New kind of ratinglist. And the tool, doing it, as a free download.
https://www.sp-cc.de/eas-ratinglist.htm