The rating of a perfect player

Very interesting thread!! 

Off the subject a little but relevant in maybe a longshot is my favorite chess game for PC - Chessmaster, I loved the seemingly different personalities of the computer players and the different styles. 

Of course it is all programming in the end and just thinking out loud it would be neat to have a sophisticated endough machine to not only be a database of information but a self learning and delevoping personality devoted only to chess. A Skynet of sorts :)

Well, that settles that!

That was more simple than I would have liked XD

[EDIT: just realised what I originally wrote here was covered in loomis' post #8]

Simple for you perhaps, Ilmago.

I would imagine it would be possible to generate an evaluation function to see which lines might pose more problems. I recall doing a complete computer analysis of variants of solitaire about 10 years ago, and coming up with a method to identify which positions were more difficult to solve. It was based on the number of best lines available compared to all the lines available. A more sophisticated version of this might serve the purpose.

Yes, and another possibility for such an algorithm may just be to let one of the normal chess engines calculate without tablebase support and use its evaluations to choose among the moves that are correct according to the tablebase.

But as soon as a pure tablebase engine starts to swap off its extra material ("because it is the fastest way to achieving the result; a draw ;-) "), then defending against it becomes disappointingly simple for almost every human player :-)

And yet, such play can be seen from a pure tablebase engine, this is why a pure tablebase engine, without making any mistakes that change the outcome of the game at best play, is playing very poorly just because it does not try to win what may well be winnable against non-perfect players.

It seems that together in this thread, we already have collected quite some knowledge about such cases and tries for possible remedies here, not only in chess, but also in solitaire, checkers, ...

Yes, ilmago. It could be said that practical chess is essentially about trying to give the opponent chances to lose positions that are widely believed to be theoretically drawn (eg the initial one). And, on the other hand, trying to avoid losing theoretically drawn positions.

For reasons discussed elsewhere, I am not as sure as many other people that the theoretical result of chess is a draw. In brief, our evidence that black can draw is based on the average score in chess games between imperfect players of similar rating being nearer 0.5 for white than to 1. One hint of doubt is that, at the highest levels (2600 and higher), white's score against black increases slightly with rating, rather than decreasing. As I believe the current standard of top human and computer players is a very long way from perfection, it would be hazardous to extrapolate to the realm of players that might consistently crush all current human and computer players. And to another level that might crush them. Only time will tell whether such players exist.

Yes, that is an interesting trend, that with increasing rating, white's score increases.

Maybe part of the reasons behind the widespread assumption that "solved chess will be a draw" is the analysis work done by the top players and by engine-driven correspondence chess play. My best guess is that those people up there spending so much time analysing may have started to develop a feeling about it and may turn out to be right with it -  in case chess is ever "solved", hopefully a long time away in the future :-) .

I didn't know that white's score increases at the higher levels, but there may be another explanation for it (rather than white having an objective advantage).

I remember reading an interview of Topalov who said that in his preparation he was looking for moves that give an advantage AND moves that creates some specific problems to the opponent, even if the objective evaluation of the move was =. If other top players do the same (and why wouldn't they), it might be that it's easier to find such difficult-to-navigate variations as white, than as black, the extra development tempo giving more opportunities for safety nets ?! 

I remember an example of such moves, not from Topalov's but from his second Cheparinov (I guess they share the same approach to preparation) :

http://www.chessgames.com/perl/chessgame?gid=1554811

21.Nc7+ was the novelty - as Aronian put it : "it's sharp, but nothing special, it's computer sharpness..."

Perhaps not the best example of causing problems for black, given that it was a draw by repetition after a few natural moves.

1.e4 might be.

Can you elaborate, Ziryab, I have no idea what your comment means.

1.e4 might possibly be a non-perfect move. It is aggressive, but entails more risk than the others you listed. Against flawed human play, it scores well, but may be refuted as computers deepen their positional understanding.

It's just a hunch, of course.

As a devout  1e4  player I must say : 

                 H O G W A S H ! 

Correct me if I'm wrong, but doesn't the mainline Italian offer a draw for white? 

I mean, very few people play for a draw as white (and if we're playing to draw against a  superior human, he'll gladly sacrifice a minute positional advantage to complicate the situation), but a perfect engine will not, by definition, make bad moves for a practical advantage.

This reminds me of a very tongue-in-cheek statement attributed to Breyer in the 1920s "After 1 P-K4 White’s game is in its last throes". [It is possible he never actually said this]. Of course, no serious chess player would take such a statement literally, even if they preferred other moves.

It is just within the bounds of plausibility (for me, a habitual 1. e4 player), that (if you had access to the hypothetical 32-piece tablebase) some other move might win by force for white but 1. e4 only draw. But for practical purposes 1. e4 achieves a plus score like 1. Nf6, 1. g3, 1. d4 and 1. Nf3 (in order of increasing success for white). It would be speculative to read anything into the slight differences between them - it is unclear whether one of these moves is better than another in absolute terms (most chess players would say that it is likely that all these moves draw with perfect play).

All but the most overanalysed tail-ends of opening lines have plenty of branches whose results are not known for sure. This includes the mainline Italian, to any reasonable depth.

Not even close. Modern engines are already at that level.

I think a good approximation would be the white-only rating of houdini + 400 hundred.