chess knowledge vs. search depth

[Dann Corbit]

Would it be fair to assume that the need for positional knowledge (or atleast positional considerations) in chess programs decreases as search depths increase? If so, does it decrease exponentially or at a constant rate?
I think Tarrasch, Nimzovich and Fred Reinfeld would be rolling in their graves if they saw the technique of a 2950+ elo program that would beat any of them effortlessly.
Todays programs seem to thrive on doubled, tripled, isolated and otherwise horrible looking pawn structures. Poisoned pawns are a special delicacy. Is it the massive search depths that are causing this utter disregard for positional considerations? Will it become even more pronounced? By extension, would a perfect player need any knwledge at all, other than how the pieces move?
Opinions?
Peter

I see the opposite of your conclusion. Five years ago, chess programs were very bad in gambit situations (and so the opening books avoided them at all costs). Today, if you let the strongest engines think at 40 moves / 2 hrs, they can understand even gambit opening book positions.

The strongest chess programs do not try to capture poisoned pawns (unless it is not as toxic as it looks at first glance). They develop knights before bishops, even if you remove the opening book, and leave rook and queen at home until substantial development has occurred.

In short, I think that the latest crop of chess programs at slow time controls play far more like GMs than the programs of a few years ago.

I guess that at correspondence time control, there is no move so brilliant that some chess program won't make it today, given sufficient hardware (e.g. Rybka 2.3.2a on 8 CPU monster will probably see all the LCT II moves {if they really are the best} at 24 hours/move -- of course I am guessing but I think it is a good guess).

P.S.
The GMs will also snack on a pawn if you leave it hanging, even though they could spend the tempo developing and will sometimes accept a doubled pawn if they see some other benefit connected with the capture.

[Dann Corbit]

A chess engine can play perfectly with no evaluation at all except win/loss/draw if it can search far enough.

As Robin smith explained this is not correct if perfectly means beating weak players.

The engine with no evaluation except win/loss/draw can allow weak players to get easy draw(by playing poor moves that still draw and later force repeatition).

Uri

It may need to search 12000 plies, but if it can search perfectly it will play perfectly. Essentially, a perfect search with no evaluation except win/loss/draw is just a proof search like that used in checkmate solvers.

[Uri Blass]

A chess engine can play perfectly with no evaluation at all except win/loss/draw if it can search far enough.

As Robin smith explained this is not correct if perfectly means beating weak players.

The engine with no evaluation except win/loss/draw can allow weak players to get easy draw(by playing poor moves that still draw and later force repeatition).

Uri

It may need to search 12000 plies, but if it can search perfectly it will play perfectly. Essentially, a perfect search with no evaluation except win/loss/draw is just a proof search like that used in checkmate solvers.

This "perfect" play is enough to draw with everybody but may not be enough even to beat me because the engine may allow me easy draw instead of trying to go to positions when it is hard to find the draw.

Uri

[Dann Corbit]

A chess engine can play perfectly with no evaluation at all except win/loss/draw if it can search far enough.

As Robin smith explained this is not correct if perfectly means beating weak players.

The engine with no evaluation except win/loss/draw can allow weak players to get easy draw(by playing poor moves that still draw and later force repeatition).

Uri

It may need to search 12000 plies, but if it can search perfectly it will play perfectly. Essentially, a perfect search with no evaluation except win/loss/draw is just a proof search like that used in checkmate solvers.

This "perfect" play is enough to draw with everybody but may not be enough even to beat me because the engine may allow me easy draw instead of trying to go to positions when it is hard to find the draw.

Uri

If the search does not find a win when a win is possible and instead finds a draw, then the search is not perfect.

It is the same as a checkmate search program that cannot find a checkmate when a checkmate is possible.

Checkmate programs do not (in general) contain an evaluation function other than win/lose/draw. And yet they always find the win if it is present and they are allowed enough time to search.

In a similar way, a perfect search *must* find a win if a win exists or the search is imperfect.

[Uri Blass]

A chess engine can play perfectly with no evaluation at all except win/loss/draw if it can search far enough.

As Robin smith explained this is not correct if perfectly means beating weak players.

The engine with no evaluation except win/loss/draw can allow weak players to get easy draw(by playing poor moves that still draw and later force repeatition).

Uri

It may need to search 12000 plies, but if it can search perfectly it will play perfectly. Essentially, a perfect search with no evaluation except win/loss/draw is just a proof search like that used in checkmate solvers.

This "perfect" play is enough to draw with everybody but may not be enough even to beat me because the engine may allow me easy draw instead of trying to go to positions when it is hard to find the draw.

Uri

If the search does not find a win when a win is possible and instead finds a draw, then the search is not perfect.

It is the same as a checkmate search program that cannot find a checkmate when a checkmate is possible.

Checkmate programs do not (in general) contain an evaluation function other than win/lose/draw. And yet they always find the win if it is present and they are allowed enough time to search.

In a similar way, a perfect search *must* find a win if a win exists or the search is imperfect.

Of course it finds a win when win is possible but this is not enough for more than a draw against non perfect player.

It seems that I have to give an example with a diagram to explain it.

[d]4r1k1/8/8/8/8/8/8/4KRB1 w - - 0 1

A GM may play Kd2 and win against me because I will blunder later when your perfect player that has no evaluation may blunder by Kf2 when I draw easily by Rf8+ and Rxf1

Drawing move in a draw position may be a practical big blunder that cost you 1/2 point and this is the reason that search with no evaluation is not enough.

Uri

[fern]

Knowledge in this, as in anything, is a kind of a elaborate, sophisticated shortcut to replace what we cannot do, simply to see all as it is. They are rules of thumbs even if with pretensions of being something more. If you can see -trough search. you does not need them.
The question is: how much search is necessary for what?

My best
Fernando

[Uri Blass]

Knowledge in this, as in anything, is a kind of a elaborate, sophisticated shortcut to replace what we cannot do, simply to see all as it is. They are rules of thumbs even if with pretensions of being something more. If you can see -trough search. you does not need them.
The question is: how much search is necessary for what?

My best
Fernando

Search is not going to tell you how to help your opponent to blunder.

If you have a won position then deep search is enough and you do not need evaluation but
If you have a drawn position then you need evaluation to help your opponent to blunder.

Uri

[Dann Corbit]

A chess engine can play perfectly with no evaluation at all except win/loss/draw if it can search far enough.

As Robin smith explained this is not correct if perfectly means beating weak players.

The engine with no evaluation except win/loss/draw can allow weak players to get easy draw(by playing poor moves that still draw and later force repeatition).

Uri

It may need to search 12000 plies, but if it can search perfectly it will play perfectly. Essentially, a perfect search with no evaluation except win/loss/draw is just a proof search like that used in checkmate solvers.

This "perfect" play is enough to draw with everybody but may not be enough even to beat me because the engine may allow me easy draw instead of trying to go to positions when it is hard to find the draw.

Uri

If the search does not find a win when a win is possible and instead finds a draw, then the search is not perfect.

It is the same as a checkmate search program that cannot find a checkmate when a checkmate is possible.

Checkmate programs do not (in general) contain an evaluation function other than win/lose/draw. And yet they always find the win if it is present and they are allowed enough time to search.

In a similar way, a perfect search *must* find a win if a win exists or the search is imperfect.

Of course it finds a win when win is possible but this is not enough for more than a draw against non perfect player.

It seems that I have to give an example with a diagram to explain it.

[d]4r1k1/8/8/8/8/8/8/4KRB1 w - - 0 1

A GM may play Kd2 and win against me because I will blunder later when your perfect player that has no evaluation may blunder by Kf2 when I draw easily by Rf8+ and Rxf1

Drawing move in a draw position may be a practical big blunder that cost you 1/2 point and this is the reason that search with no evaluation is not enough.

Uri

He does not need to evaluate because the search will see the draw in the case of a blunder. You could leave a stipulation to play on until there are zero wins in the tree.

[Dann Corbit]

Knowledge in this, as in anything, is a kind of a elaborate, sophisticated shortcut to replace what we cannot do, simply to see all as it is. They are rules of thumbs even if with pretensions of being something more. If you can see -trough search. you does not need them.
The question is: how much search is necessary for what?

My best
Fernando

Search is not going to tell you how to help your opponent to blunder.

If you have a won position then deep search is enough and you do not need evaluation but
If you have a drawn position then you need evaluation to help your opponent to blunder.

Uri

This can also be deduced from the tree that only has outcomes in it. Always move towards the drawn branches with the most outcomes of +1 in them until all outcomes are =0.

[smirobth]

Knowledge in this, as in anything, is a kind of a elaborate, sophisticated shortcut to replace what we cannot do, simply to see all as it is. They are rules of thumbs even if with pretensions of being something more. If you can see -trough search. you does not need them.
The question is: how much search is necessary for what?

My best
Fernando

Search is not going to tell you how to help your opponent to blunder.

If you have a won position then deep search is enough and you do not need evaluation but
If you have a drawn position then you need evaluation to help your opponent to blunder.

Uri

This can also be deduced from the tree that only has outcomes in it. Always move towards the drawn branches with the most outcomes of +1 in them until all outcomes are =0.

This would be a big positive step towards avoiding lines like 1.f3 and 2.Kf2, but I still think it is not the optimum solution for inducing mistakes in order to beat the current crop of top GMs and software. Such a strategy would encourage moves that dragged the game on as long as possible (since long games have many more moves, thus many more potential blunders and potential +1 leaf nodes) even if it meant that no serious presure was being put on the opponent and blunders were less likely to occur. For example such a program might play moves that closed the position and avoided exchanges even in positions where doing so creates no real difficulties for the opponent. Also such a program might be willing to repeat the exact same drawn game multiple times. Thus I believe such a program might achieve far fewer wins against GMs and current GM level software than current PC software achieves. It would probably win many games (and obviously lose none) and so it would be quite strong. But I still think relatively weak players might get more draws against such a program than they can against current top level programs. Just look how well some relatively weak humans do against programs when they can retain all the pawns on the board and lock up the position. Now imagine they are playing a program that itself tends to head towards such positions.