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

[BrendanJNorman]

Talking about "perfect chess" or 32 piece TBs is like talking about aliens or similar.

We have next to no data to analyze any of these things.

Given that Stockfish plays far from perfect chess (very strong is still far from perfect, just as 7 piece tablebases are exponentially far away from 32 piece...imo), my feeling is that it would lose 100-0 to 32 piece tablebases or a "perfect chess" entity.

Do you have an elo estimate for 32 piece. I think Kai would you just give it another 400 elo.

Maybe I'm ignorant on the technology, but to me, 32 piece tablebases=perfect play and perfect play= Elo unlimited.

It should be for example: 1.Nf3 d5 -+ (mate in 42)

Stockfish is say 3680 now (or whatever) and still far from perfect chess...far from perfect.

If Stockfish NNUE is even 3700, probably perfect chess is 5000 or more.

Totally theoretical of course. Even Capablanca thought chess was "solved" and that was in the 1930s. If Stockfish can still lose games, and it can, it is FAR from perfect...so perfect chess would crush SF 100-0 IMO.

Bit of rambling...12.18am and a couple nice German beers downed.

I see no proof that perfect chess is going to crush SF 100-0
I even see no proof that stockfish can lose games with white in TCEC conditions.
A proof can be to show me one game that stockfish lose with White from the opening position in TCEC conditions(when stockfish does not use book).

Note that I do not claim that stockfish cannot lose games and only that I do not know.
I do not understand why people post claims with no evidence for them.

Note that stockfish is not deterministic so even if it is possible to beat stockfish with black it does not mean that 100-0 is possible against it.

I didn't "claim" anything and I actually said "Maybe I'm ignorant on the technology" - this was followed by an opinion.

So calm down, sister.

My opinion was that Stockfish, despite its enormous strength, is still far from perfect chess and that given that...

1. we only have 7 man tablebases at the moment
2. this is less than 25% of theoretical "perfect chess" (with 32 man TBs)
3. the strength of play increases exponentially with each new bigger set

We have NO IDEA what perfect chess looks like...and neither does Stockfish.

If/when perfect chess arrives, we might not even recognize what the hell is going on on the board.

A complete revolution in pattern recognition.

It is like a monkey in the Amazon jungle trying to understand what is airplane flying over.

Again, just my opinion, don't get too excited.

I agree. People need to understand that Stockfish can not even play 7 man positions correctly. And we know this because of tablebases. Or perfect play with 7 man positions.

So what do you think a 32 man tablebase would do with today's Stockfish. It would crush Stockfish into fish paste.

You made a good point.

If SF can't play 7 man positions perfectly...

...it is absolutely obvious that a "perfect" 32-man TB engine will humiliate SF 100-0.

This isn't about something like a 5 centipawn loss...it is ONE minor inaccuracy and you're being mated in x moves.

SF loses 100-0 and the monkey in the Amazon still has no idea what the passing airplane is.

[Alayan]

The comparison is silly.

You don't need to be able to play every single legal 32-men position perfectly to never lose a game from the start position as white. You only need to draw or win from positions that follow sensible white moves your book/engine chooses to play. If black plays to equalize, it makes white's job of not losing even easier. To hope for a win, black must play for complications, but this makes black's position worse against a competent enough opponent.

Furthermore, the positions that are really hard to play perfectly feature a strong advantage for one side and/or big imbalances. This is not the case of the start position, which is very symmetrical and extremely balanced.

The hypothesis that perfect chess would look nothing like what Stockfish or Leela play today is baseless. Looking at how engines changed human opening theory is clear evidence. Humans play very imperfect chess, and the details of many lines were changed with the advent of powerful engines. But the main structures have remained unchanged. The Ruy Lopez, the QGD, the Sicilian, the Nimzo... were thought to be great openings in that neither side made any inaccuracy. Stockfish and Leela today support this assessment. Engines from 2030 will still support it.

[Nay Lin Tun]

As chess is extremely drawish, one side should get +5 material advantage to get a winning chance ( assuming pawns are potential +10 materials).

You can set up pgn with TCEC final like positions which are extremely sharp and one inaccuracy will lead to lost against TB. However, in reality, if you let Stockfish play freely, those sharp or cramped positions are practically impossible to reach as there are usually 1-2 inaccuracies/blunders somewhere and Stockfish would have avoided it.

In fact, strategies to get draw is x100 times easier in chess than winning. If we create a small opening book for Stockfish just to follow symmetrical positions only, the chances of losing becomes extremely slim.

1. e4 e5 ( force Stockfish to play Berlin, Ruylopez). Sicilian create asymmetry in position and it create some chance of losing. The same strategy to draw with white is much more easy.

[Dann Corbit]

LC0 went on to draw the next 6 games.
A 60 ply search of the opening position after 1.b4 by Stockfish with 60 threads and full 6 man tablebase file with important 7 man files gives a score of 0.0.
My statistics say that the net outcome is about equal for both sides, agreeing with the evaluation.

[duncan]

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.


On the other hand, current Stockfish with contempt may be slightly higher rated than 32 men TB in Round Robin rating pool. Prettty crazy right? Actually 32 men TB doesn't store statistics of winning chances or knowledge of maximising chances to win which will subsequently lose some elo. For example, in KRB vs KR, 7 men TB will randomly throw away one rook as it knows that all moves are draw. ( KB vs KR is still draw). Carry behaviour of how this 7 men TB play into 32 men TB, we can see TB will be randomly choose the first move 1. a3 or f3 or h3 or e4 (assuming that except 1.g4, all other 19 moves are likely to be draw). Subsequently, 32 men TB will choose next move like. 1. a3 e5 2. a4 ( It still knows as a draw but it throw away practical winning chances)

Quote: "The size of 8-man tablebases will be 100 times larger than the size of 7-man tablebases. To fully compute them, one will need about 10 PB (10,000 TB) of disk space and 50 TB of RAM. Only the top 10 supercomputers can solve the 8-man problem in 2014. The first 1000-move mate is unlikely to be found until 2020 when a part of a TOP100 supercomputer may be allowed to be used for solving this task."

I cant comprehend a 32 man table base and the storage needed for it.

From the point of view of hardware, today it could be solved for a million dollars worth of computer ram and hard drive

[duncan]

LC0 went on to draw the next 6 games.
A 60 ply search of the opening position after 1.b4 by Stockfish with 60 threads and full 6 man tablebase file with important 7 man files gives a score of 0.0.
My statistics say that the net outcome is about equal for both sides, agreeing with the evaluation.

Did you do the 60 ply search?

[duncan]

Maybe but I think it's very unlikely. I think if it was the case that perfect chess look so amazingly different we would somehow see at least some of it sometimes with some engine stumbling into it or some correspondence player showing it.
As it is correspondence chess is completely hopeless when it comes to winning against competent opponent who just wants to draw with black. All correspondence wins come from:

<snips>

This is sort of begging the question. We only have perfect information of a trillion trillionth etc positions.
Correspondence player are stumbling in the dark of the galaxies of chess positions, so how can they hope to even of having a hint of finding the hypothetical pawn capture on move 45?

[Dann Corbit]

LC0 went on to draw the next 6 games.
A 60 ply search of the opening position after 1.b4 by Stockfish with 60 threads and full 6 man tablebase file with important 7 man files gives a score of 0.0.
My statistics say that the net outcome is about equal for both sides, agreeing with the evaluation.

Did you do the 60 ply search?

No, stockfish did

[duncan]

The comparison is silly.

You don't need to be able to play every single legal 32-men position perfectly to never lose a game from the start position as white. You only need to draw or win from positions that follow sensible white moves your book/engine chooses to play. If black plays to equalize, it makes white's job of not losing even easier.

To hope for a win, black must play for complications, but this makes black's position worse against a competent enough opponent.

You are talking about black with a 32 man tablebase using complication tricks. So what is the issue ?

Furthermore, the positions that are really hard to play perfectly feature a strong advantage for one side and/or big imbalances. This is not the case of the start position, which is very symmetrical and extremely balanced.

The hypothesis that perfect chess would look nothing like what Stockfish or Leela play today is baseless. Looking at how engines changed human opening theory is clear evidence. Humans play very imperfect chess, and the details of many lines were changed with the advent of powerful engines. But the main structures have remained unchanged. The Ruy Lopez, the QGD, the Sicilian, the Nimzo... were thought to be great openings in that neither side made any inaccuracy. Stockfish and Leela today support this assessment. Engines from 2030 will still support it.

[duncan]

LC0 went on to draw the next 6 games.
A 60 ply search of the opening position after 1.b4 by Stockfish with 60 threads and full 6 man tablebase file with important 7 man files gives a score of 0.0.
My statistics say that the net outcome is about equal for both sides, agreeing with the evaluation.

Did you do the 60 ply search?

No, stockfish did

Did you do the 60 ply search using stockfish?