Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

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

Moderators: hgm, chrisw, Rebel

mwyoung
Posts: 2727
Joined: Wed May 12, 2010 10:00 pm

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by mwyoung »

BrendanJNorman wrote: Thu Nov 12, 2020 5:18 pm
mwyoung wrote: Thu Nov 12, 2020 5:10 pm
Alayan wrote: Thu Nov 12, 2020 5:06 pm
mwyoung wrote: Thu Nov 12, 2020 4:37 pm
"current Stockfish with contempt may be slightly higher rated than 32 men TB in Round Robin rating pool. Prettty crazy right?"
:lol: :lol: :lol:
32-men TB by itself never loses but sucks at winning. Something you're too thick to understand even when it's spelled out slowly. It completely breaks the assumptions of the elo model. That's why considering a "God" entity that knows how to play challenging move is much more interesting than a raw TB.
I noticed, and maybe you should learn to read. You did not answer the question.

Where in the hell does he say "would not lose as white from the starting position." :lol: :lol: :lol:
He is apparently implying it with:
"If Stockfish is playing against 32 men TB, the extreme condition would be 50 draw 50 losses , which is within 200 elo."
The 50 draws would be with white, he imagines. :wink:

He is assuming that chess is a forced draw with perfect play and that SF plays perfectly with white.
I guess, but still does not make sense. If chess is a draw color does not matter. If you are playing near perfect chess. :lol:

If black has the forced win, then black would be the case in your example.

You ASSUME too much, but not as much as claiming draws prove we are playing perfect chess! :lol:



Yes, It was clear as mud!
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
User avatar
towforce
Posts: 12019
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK
Full name: Graham Laight

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by towforce »

BrendanJNorman wrote: Thu Nov 12, 2020 5:16 pmThere is no proof AT ALL that chess is a draw with perfect play. And this is the basis for OP's entire hypothesis. In fact, there is more evidence pointing toward perfect chess being a win for white. The overwhelming white wins vs black ratio at ALL levels of chess is not a coincidence.

Maybe we should have a poll about this. I agree that there is no proof that chess is a draw, but my opinion is that it is:

1. There is no known way for either side to force the win of material from the opening

2. The higher the standard of the players (and the players being of roughly equal standard), the more likely the game is to be drawn

So in this thread:

Win for white: 1

Draw: 1

Win for black: 0
Want to attract exceptional people? Be exceptional.
mwyoung
Posts: 2727
Joined: Wed May 12, 2010 10:00 pm

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by mwyoung »

towforce wrote: Thu Nov 12, 2020 5:58 pm
BrendanJNorman wrote: Thu Nov 12, 2020 5:16 pmThere is no proof AT ALL that chess is a draw with perfect play. And this is the basis for OP's entire hypothesis. In fact, there is more evidence pointing toward perfect chess being a win for white. The overwhelming white wins vs black ratio at ALL levels of chess is not a coincidence.

Maybe we should have a poll about this. I agree that there is no proof that chess is a draw, but my opinion is that it is:

1. There is no known way for either side to force the win of material from the opening

2. The higher the standard of the players (and the players being of roughly equal standard), the more likely the game is to be drawn

So in this thread:

Win for white: 1

Draw: 1

Win for black: 0
Again false assumption.

You do not need to win material to win a game of chess. Many chess games are won with both sides having equal material. So you win by checkmating the King! And this can also be done by sacking your material to deliver checkmate to the King!
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Nay Lin Tun
Posts: 710
Joined: Mon Jan 16, 2012 6:34 am

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by Nay Lin Tun »

For people who are still wondering white or black forced win, show me a single game where one side forced win without a single blunder from the opponent.

There are billions of computers games and human games played already to proove/ disprove your assumption.
mwyoung
Posts: 2727
Joined: Wed May 12, 2010 10:00 pm

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by mwyoung »

Nay Lin Tun wrote: Thu Nov 12, 2020 6:32 pm For people who are still wondering white or black forced win, show me a single game where one side forced win without a single blunder from the opponent.

There are billions of computers games and human games played already to proove/ disprove your assumption.
Strawman! Only God and a 32 man table base knows if chess is a draw or a win for white or black in chess.
If oneside is a win, a 32 man tablebase would lose playing the losing side without a single blunder against another 32 man TB from the winning side.

And chess engines and humans are not good enough to know and show this, if one side has a forced win.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
User avatar
towforce
Posts: 12019
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK
Full name: Graham Laight

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by towforce »

mwyoung wrote: Thu Nov 12, 2020 6:53 pm
Nay Lin Tun wrote: Thu Nov 12, 2020 6:32 pm For people who are still wondering white or black forced win, show me a single game where one side forced win without a single blunder from the opponent.

There are billions of computers games and human games played already to proove/ disprove your assumption.
Strawman! Only God and a 32 man table base knows if chess is a draw or a win for white or black in chess.
If oneside is a win, a 32 man tablebase would lose playing the losing side without a single blunder against another 32 man TB from the winning side.

And chess engines and humans are not good enough to know and show this, if one side has a forced win.

I don't want a re-run of the "Is Chess Solved" thread until I've made some good progress, but I do think that it is likely that chess can be solved with today's technology, and that means without having a 32 piece tablebase.

One of the ideas discussed in that thread follows on from what Nay Lin Tun said in the quoted text above. Checkmate in chess is closely analogous to winning material, so if you can force checkmate, it's very likely that it must be possible to force the win of material. So the following route to resolution might be possible:

1. What conditions must exist in a position in order to win material? Condition Set 1 (CS1)

2. What conditions must exist in order to get CS1 (CS2)

3. What conditions must exist in order to get CS2 (CS3)

etc.

Initially, this looks daunting, but remember we have automated mathematical proving applications (based on SAT solvers) available, so it might be doable (it's not the approach I am personally using - but it might well get you to the solution of chess faster than the one I have chosen).
Want to attract exceptional people? Be exceptional.
Uri Blass
Posts: 10609
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by Uri Blass »

BrendanJNorman wrote: Thu Nov 12, 2020 5:16 pm
Alayan wrote: Thu Nov 12, 2020 3:16 pm
I said I'm out, because there is an astonishing lack of reading comprehension. Your stupid meme made me write this... You'd be ashamed if only you were aware of how wrong you've got what is discussed.
hahaha Yeah, the meme was good for a laugh. I knew some emotional types would probably get triggered. :lol:

You must be a young guy like Andrew, right? Were you sobbing when NNUE first came as well? Pretty sure you were.

Shouldn't let things get to you so easily. Come out of the basement once in a while, you'll feel better. ;)
Alayan wrote: Thu Nov 12, 2020 3:16 pm
The OP's claim was the following:
If there were no opening book, Stockfish will be minimally able to draw another Stockfish or Leela in 95% of games. If Stockfish is playing against 32 men TB, the extreme condition would be 50 draw 50 losses , which is within 200 elo.

There is no proof AT ALL that chess is a draw with perfect play. And this is the basis for OP's entire hypothesis. In fact, there is more evidence pointing toward perfect chess being a win for white. The overwhelming white wins vs black ratio at ALL levels of chess is not a coincidence.

At the beginning of a chess game, black is trying to stop the wave of white initiative (the advantage of the first move) from engulfing him.

For all we know, a draw occurs when white makes an imprecise move and allows this "wave of initiative" to be stopped - us chessplayers call this "equalizing".

Perfect chess is very likely a constant wave of white initiative ending in a white checkmate.
Alayan wrote: Thu Nov 12, 2020 3:16 pm Hint : losing a blitz game as black on a weak laptop is irrelevant.
If Stockfish played "perfect chess" or was even close, it wouldn't lose such games, regardless of hardware or time control. Most importantly, if Stockfish were close to "perfect chess" you would not be adding Elo at such a fast rate in dev versions.

Forget your math and graphs and emotional triggers and excuses about my "weak laptop" and whatever...perfect chess is impossible to improve upon and "nearly" perfect chess is VERY VERY VERY hard to improve upon - you cannot improve "nearly perfect chess" by 2-3 Elo every day or two, regardless of how you try to rationalize it.

And this rapid improvement is still happening with "perfect" Stockfish.

1)I see no reason to believe perect chess is not a draw considering the fact that usually the final result of games between top players is a draw.

2)We do not know that stockfish continue to add elo at fast rate at long time control.
Tests are only at fast time control.

3)For results of stockfish against the perfect 32 piece tablabases or against future programs the fact is that we do not know.
Results against 32 tablebases is not defined because we do not know how the 32 piece tablebases choose moves.

The fact that there are many positions when stockfish cannot find the best move is known and basically prove nothing


4)I see no reason to believe that going for the longest draw is the correct way.
The target is to find the moves that make it harder for the weaker side to make a draw.
Harder is not the same as longer.

You can have in one line 200 moves that are easy to find and in another line 40 moves that are harder to find.
OneTrickPony
Posts: 157
Joined: Tue Apr 30, 2013 1:29 am

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by OneTrickPony »

Strawman! Only God and a 32 man table base knows if chess is a draw or a win for white or black in chess.
If oneside is a win, a 32 man tablebase would lose playing the losing side without a single blunder against another 32 man TB from the winning side.

And chess engines and humans are not good enough to know and show this, if one side has a forced win.
There is more to forming reasonable beliefs than:

NOT MATHEMATICALLY PROVEN -> I CAN BELIEVE WHATEVER WITH ANY CONFIDENCE

We have overwhelming evidence that perfect chess ends in a draw. You are arguing an equivalent of Goldbach conjecture being false. Sure, it might be the case but no mathematician believes it the same way no reasonable person who knows anything about chess believes the game is not drawn with perfect play.
User avatar
towforce
Posts: 12019
Joined: Thu Mar 09, 2006 12:57 am
Location: Birmingham UK
Full name: Graham Laight

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by towforce »

From an article called "God's Chess Rating" (link):

This chart shows that the 100% draw rate will be achieved at elo rating 5237:

Image

This chart shows that, at the current rate of progress, this rating will be achieved in the year 2059:

Image

Of course, someone might work out a solution to chess before that time...
Want to attract exceptional people? Be exceptional.
Alayan
Posts: 550
Joined: Tue Nov 19, 2019 8:48 pm
Full name: Alayan Feh

Re: Perfect chess engine elo ( 32 men TB) can be within 200 of Stocfish in Tcec LTC conditions

Post by Alayan »

This chart looks nice, but it's also very bad quality data.

No data source is given, I'll assume it's something like the CCRL DB.

The opponent mix that each engine face is different, and has a decisive impact on the draw rate. An engine that only faces engine within +-50 elo would not exhibit the same draw rate as an engine facing a 300 elo weaker or stronger entity. Stockfish 12 8CPU at CCRL blitz scored >89% against 5 of its 13 opponents.

If you take FastGM bullet list, newer versions of Stockfish have a decreasing draw rate because they dominate weak engines ever more decisively.

Therefore, the draw rate of an engine at CCRL is only a very poor and unreliable lowerbound of the draw rate between similarly strong engine of this level.

Another major issue is comparing human classical elo with engine elo in unclear conditions. A 2500 GM classical game and a 2500 CCRL blitz game are very different in nature. The use of forced openings for variety in CCRL data also alters the draw rates.

The 5200 elo lowerbound extrapolation is a case of garbage in, garbage out.

Also, the A0 "datapoint" is a joke...