GM says Rybka & Fritz weaker than best GMs in classical

[Uri Blass]

A GM can _way_ out-calculate a computer along sharp tactical lines for the most part,

That is not true for today's hardware and top engines !!!!!.

Matthias.

Easy to say, but programs have a fixed and fairly shallow horizon. GMs sometimes calculate variations to 40-50 plies. Programs don't.

If you mean in correspondence chess, aided by engines, you are right.
Otherwise there is no proof for your claim.

Statistics prove that you are wrong:
if the GMs were that strong, they would not be blundering as frequently as they do in important human-human tournaments.
Just pick _any_ one of such tournaments and count the blunders at _shallow_ depth. There is no way they can calculate variations to 40-50 plies without using engines.

Matthias.

I'm only going to say this one more time and then move on to other topics. Correspondence play in the 1960's and 1970's did _not_ use computers, yet the variations were calculated out to depths that I considered impossible. Berliner had some gems and did _not_ need a computer to help him, since none were available back then.

GMs blunder because they are human, and make mistakes in time pressure, or when distracted, etc. But overall their moves are quite good. Otherwise I guess all the "greatest games" books need to be trashed as too full of blunders to be useful?

This is one of those "impossible to prove" situations so there is little point in continuing the back and forth discussion. You can have the last word. I still believe GM players are far stronger than computers overall in terms of positional play. And in the case of certain types of very deep and forcing tactics. Computers don't miss anything within their search horizon, and this horizon has gotten deep enough that they give GM players great trouble now. A GM can certainly calculate as deep as or deeper than a computer program. But at a cost of mental energy, and eventually fatigue will decide the outcome in many games... it is more a case of the GMs losing than it is a case of the computers winning... and it now happens frequently enough that GMs are beginning to not do very well against computers. Whether they would do better with one game a week is unknown since such an event would take forever.

Bob, don't waste your time with them, they know far more than we do.

That's not polite.
In a conversation where 2 sides disagree, one side states its arguments and the other answers by trying to refute them and provide better. Etc, and conversation continues.
By saying, like you did, with an ironic way that you know better, and also expecting everyone to agree with what you say is ridiculous.
You can't force everyone to agree with you and if they don't, to start being ironic against them is a bad thing....
You should try to refute their statements instead.

He can't,the truth and the statistical data are obvious for everyone to see....
No more human strategical superiority nonsense,set on the table and prove it,or let one of your GM searching 500 plies ahead do it....

As I said, I have given up trying to explain. I've not said that GMs can beat programs OTB very often, although they can still do it. I have simply explained _how_ the computer is beating the GM. And based on my experience, my reasoning is correct. I've watched it over the years. We started beating GMs at blitz in the late 70's, not because we outplayed them, but we let them make mistakes that turned out to be fatal. Today the same idea is holding on, and it is approaching the point where a GM simply can not play with enough accuracy over a long enough period of time, to hold off the computers. But to say that computers can positionally outplay top GM players is just silly, for anyone with enough chess skill to understand the ideas involved...

A computer is so bad at recognizing weak pawns, weak squares and such. But its tactics carry it anyway...

IM larry kaufman in OTB chess has a different opinion.

http://rybkaforum.net/cgi-bin/rybkaforu ... 3#pid71643

"The key to Rybka's strength (and other programs too, though to a lesser degree) is that a decent evaluation after a very deep search will produce better positional play than a super evaluation after a much shallower or narrower search. No practical number of takebacks will change this; the human GM will be outplayed positionally despite his better evaluation, because he cannot possibly compete in the realm of search"

Uri

[bob]

A GM can _way_ out-calculate a computer along sharp tactical lines for the most part,

That is not true for today's hardware and top engines !!!!!.

Matthias.

Easy to say, but programs have a fixed and fairly shallow horizon. GMs sometimes calculate variations to 40-50 plies. Programs don't.

If you mean in correspondence chess, aided by engines, you are right.
Otherwise there is no proof for your claim.

Statistics prove that you are wrong:
if the GMs were that strong, they would not be blundering as frequently as they do in important human-human tournaments.
Just pick _any_ one of such tournaments and count the blunders at _shallow_ depth. There is no way they can calculate variations to 40-50 plies without using engines.

Matthias.

I'm only going to say this one more time and then move on to other topics. Correspondence play in the 1960's and 1970's did _not_ use computers, yet the variations were calculated out to depths that I considered impossible. Berliner had some gems and did _not_ need a computer to help him, since none were available back then.

GMs blunder because they are human, and make mistakes in time pressure, or when distracted, etc. But overall their moves are quite good. Otherwise I guess all the "greatest games" books need to be trashed as too full of blunders to be useful?

This is one of those "impossible to prove" situations so there is little point in continuing the back and forth discussion. You can have the last word. I still believe GM players are far stronger than computers overall in terms of positional play. And in the case of certain types of very deep and forcing tactics. Computers don't miss anything within their search horizon, and this horizon has gotten deep enough that they give GM players great trouble now. A GM can certainly calculate as deep as or deeper than a computer program. But at a cost of mental energy, and eventually fatigue will decide the outcome in many games... it is more a case of the GMs losing than it is a case of the computers winning... and it now happens frequently enough that GMs are beginning to not do very well against computers. Whether they would do better with one game a week is unknown since such an event would take forever.

Bob, don't waste your time with them, they know far more than we do.

That's not polite.
In a conversation where 2 sides disagree, one side states its arguments and the other answers by trying to refute them and provide better. Etc, and conversation continues.
By saying, like you did, with an ironic way that you know better, and also expecting everyone to agree with what you say is ridiculous.
You can't force everyone to agree with you and if they don't, to start being ironic against them is a bad thing....
You should try to refute their statements instead.

He can't,the truth and the statistical data are obvious for everyone to see....
No more human strategical superiority nonsense,set on the table and prove it,or let one of your GM searching 500 plies ahead do it....

As I said, I have given up trying to explain. I've not said that GMs can beat programs OTB very often, although they can still do it. I have simply explained _how_ the computer is beating the GM. And based on my experience, my reasoning is correct. I've watched it over the years. We started beating GMs at blitz in the late 70's, not because we outplayed them, but we let them make mistakes that turned out to be fatal. Today the same idea is holding on, and it is approaching the point where a GM simply can not play with enough accuracy over a long enough period of time, to hold off the computers. But to say that computers can positionally outplay top GM players is just silly, for anyone with enough chess skill to understand the ideas involved...

A computer is so bad at recognizing weak pawns, weak squares and such. But its tactics carry it anyway...

I agree that in the beginning when computers only started to beat GM's it was the case but not today for the top programs.

Remember that we talk about the top programs of today and not about Crafty.

You can look at the games of rybka against GM's to see that rybka positionally outplay GM's in most of the games.

I do not claim that it is because of better evaluation function and it may be because of search but the facts remain.

Uri

Sorry, but I have actually followed many of the games as they were played out, and Rybka, just like all computers, occasionally pulls a real lemon out of its hat and advances a pawn that should never have moved, for just one example problem. But it plays tactically well enough to make it difficult to exploit some of the truly dumb moves it makes (and no, it doesn't make a lot of dumb moves, but against a GM it only takes one bad move to lose the game, unless your tactics are so good that you can find a counter-mistake made by the GM and exploit it.

You just will never find computers that enter into a 10 move sequence to start a plan and the first 10 moves are based solely on the idea of preventing the opponent from being able to use a key square, so that the plan will work later. A GM can lay out a plan for the entire game, a computer lays out a plan for exactly one move because if it doesn't see something happening it doesn't know how to anticipate that it will happen in another 10-20 moves. Fortunately that is not enough of an edge for the humans, and computers are winning. But they are winning solely based on applying so much pressure on the human to avoid mistakes, that mistakes get made.

[bob]

As I said, I have given up trying to explain. I've not said that GMs can beat programs OTB very often, although they can still do it. I have simply explained _how_ the computer is beating the GM. And based on my experience, my reasoning is correct. I've watched it over the years. We started beating GMs at blitz in the late 70's, not because we outplayed them, but we let them make mistakes that turned out to be fatal. Today the same idea is holding on, and it is approaching the point where a GM simply can not play with enough accuracy over a long enough period of time, to hold off the computers. But to say that computers can positionally outplay top GM players is just silly, for anyone with enough chess skill to understand the ideas involved...

A computer is so bad at recognizing weak pawns, weak squares and such. But its tactics carry it anyway...

This flawed assessment must come from Crafty seeing it has not been at the top for a quarter century. Seeing that you are talking the talk are you prepared to walk the walk? The only reason Kasparov and Kramnik were not humiliated like Adams is that they put in ridiculous conditions like getting the opening book/specific copy of program to be used.

Come back when you can get your attention span to cover at least 5 minutes. There is no "talk to talk" nor any "walk to walk". I explained _how_ the computers are winning. so what walk is there to walk.

SImply pay attention and read what I write, not what you want to imagine that I write...

[bob]

A GM can _way_ out-calculate a computer along sharp tactical lines for the most part,

That is not true for today's hardware and top engines !!!!!.

Matthias.

Easy to say, but programs have a fixed and fairly shallow horizon. GMs sometimes calculate variations to 40-50 plies. Programs don't.

If you mean in correspondence chess, aided by engines, you are right.
Otherwise there is no proof for your claim.

Statistics prove that you are wrong:
if the GMs were that strong, they would not be blundering as frequently as they do in important human-human tournaments.
Just pick _any_ one of such tournaments and count the blunders at _shallow_ depth. There is no way they can calculate variations to 40-50 plies without using engines.

Matthias.

I'm only going to say this one more time and then move on to other topics. Correspondence play in the 1960's and 1970's did _not_ use computers, yet the variations were calculated out to depths that I considered impossible. Berliner had some gems and did _not_ need a computer to help him, since none were available back then.

GMs blunder because they are human, and make mistakes in time pressure, or when distracted, etc. But overall their moves are quite good. Otherwise I guess all the "greatest games" books need to be trashed as too full of blunders to be useful?

This is one of those "impossible to prove" situations so there is little point in continuing the back and forth discussion. You can have the last word. I still believe GM players are far stronger than computers overall in terms of positional play. And in the case of certain types of very deep and forcing tactics. Computers don't miss anything within their search horizon, and this horizon has gotten deep enough that they give GM players great trouble now. A GM can certainly calculate as deep as or deeper than a computer program. But at a cost of mental energy, and eventually fatigue will decide the outcome in many games... it is more a case of the GMs losing than it is a case of the computers winning... and it now happens frequently enough that GMs are beginning to not do very well against computers. Whether they would do better with one game a week is unknown since such an event would take forever.

Bob, don't waste your time with them, they know far more than we do.

That's not polite.
In a conversation where 2 sides disagree, one side states its arguments and the other answers by trying to refute them and provide better. Etc, and conversation continues.
By saying, like you did, with an ironic way that you know better, and also expecting everyone to agree with what you say is ridiculous.
You can't force everyone to agree with you and if they don't, to start being ironic against them is a bad thing....
You should try to refute their statements instead.

He can't,the truth and the statistical data are obvious for everyone to see....
No more human strategical superiority nonsense,set on the table and prove it,or let one of your GM searching 500 plies ahead do it....

As I said, I have given up trying to explain. I've not said that GMs can beat programs OTB very often, although they can still do it. I have simply explained _how_ the computer is beating the GM. And based on my experience, my reasoning is correct. I've watched it over the years. We started beating GMs at blitz in the late 70's, not because we outplayed them, but we let them make mistakes that turned out to be fatal. Today the same idea is holding on, and it is approaching the point where a GM simply can not play with enough accuracy over a long enough period of time, to hold off the computers. But to say that computers can positionally outplay top GM players is just silly, for anyone with enough chess skill to understand the ideas involved...

A computer is so bad at recognizing weak pawns, weak squares and such. But its tactics carry it anyway...

IM larry kaufman in OTB chess has a different opinion.

http://rybkaforum.net/cgi-bin/rybkaforu ... 3#pid71643

"The key to Rybka's strength (and other programs too, though to a lesser degree) is that a decent evaluation after a very deep search will produce better positional play than a super evaluation after a much shallower or narrower search. No practical number of takebacks will change this; the human GM will be outplayed positionally despite his better evaluation, because he cannot possibly compete in the realm of search"

Uri

I don't see how that disagrees with what I wrote at all. "very deep search". That is exactly the issue.

[bob]

Oh God, I must say it:
Terry and Bob are right in this case -- cause they have looked on the other side too unlike you.
They have expirience that you missed -- you, Mathias, Wael (sorry, doc!): experience of knowing GM's in live.
So I can repeat my words from some posts below -- you underestimate grandmasters, guys. You strongly underestimate them.

I like to believe that GMs are as strong as their hypothetic superiority is portrayed in this thread, but they blunder too frequently.
Going by statistics, we have to say they can't be that strong.
There is no data to prove me wrong.

If I could look 40...50 plies ahead, my blunders would be after 30th ply, not at ply 6.

Matthias.

No it wouldn't. You are thinking "full-width plies" not "plies". A selective search program could look 50 plies deep and make all sorts of tactical blunders by excluding (forward-pruning) critical moves. Humans are _far_ better at doing this than computers. But the depths computers now reach are enough to compensate for their lack of selectivity. But in the right kinds of positions, the GM can simply out-search the computer along the _critical_ paths. There are way too many endgame positions in Fine's "Basic Chess Endings" book that no computer has a chance of finding, yet a human did it 50 years ago or more when there were no computers to use at all...

[Mike S.]

Nevertheless, I HOPE you are aware that even the best humans in the 2700+ class will NEVER win a long tournament time match against any good engine (incl. Crafty), running on a normal off-the-shelf hardware.

The misunderstanding is:

Yes, there are many positions which GMs understand much better. - But there are even more practical positions where humans will play (minor, or big) blunders, against an engine which calculates effective 15+ plies.

So, these matches are NOT decided by the FEW positions where top humans are stronger, but by the MANY complicated practical positions where chess programs are stronger.

[Matthias Gemuh]

No it wouldn't. You are thinking "full-width plies" not "plies". A selective search program could look 50 plies deep and make all sorts of tactical blunders by excluding (forward-pruning) critical moves. Humans are _far_ better at doing this than computers. But the depths computers now reach are enough to compensate for their lack of selectivity. But in the right kinds of positions, the GM can simply out-search the computer along the _critical_ paths. There are way too many endgame positions in Fine's "Basic Chess Endings" book that no computer has a chance of finding, yet a human did it 50 years ago or more when there were no computers to use at all...

GMs are indeed better at most endgames than engines, without even calculating very deep. However today's engines win the games before the GMs get a chance of getting that far.

Your point in the whole thread is thus that
in GM vs. engine matches the engines can be expected to be the match winners, but it is the GMs that are stronger (as the thread title says). Right ?

Matthias.

[swami]

GM says...GM's say..

"Chess is 99 percent tactics." - Teichmann

"Chess is really 99 percent calculation" - Soltis

Modern day engines are definitely far advanced in positional chess, I agree with Uri re: Rybka is a good positional engine...Everyone knew engines are far better in tactics/calculations.

Also, History of Man vs machine in classic conditions always favoured computers... now the engines are even better what with MP and more improvement.

[Terry McCracken]

Bob, you're going to get metacarpal trying to explain the subtle realities to some of these people. It's not worth it

Terry

[bob]

Nevertheless, I HOPE you are aware that even the best humans in the 2700+ class will NEVER win a long tournament time match against any good engine (incl. Crafty), running on a normal off-the-shelf hardware.

The misunderstanding is:

Yes, there are many positions which GMs understand much better. - But there are even more practical positions where humans will play (minor, or big) blunders, against an engine which calculates effective 15+ plies.

So, these matches are NOT decided by the FEW positions where top humans are stronger, but by the MANY complicated practical positions where chess programs are stronger.

As I have said, many times now, yes the machines are superior. I was explaining _why_ they are superior. It is not from their inferior evaluations when compared to humans, it is because of their great long-term ability to avoid simple tactical mistakes. I have seen computers get zonked with a deep tactical idea they completely missed (kingside attacks come to mind, but also I have seen the same in some endgames). But I have also seen humans make somewhat shallow tactical errors as well, due to human weaknesses that have been covered many times. If a GM _could_ play at the top of his game for 5 hours a day for a week, he could win a match against computers consistently. But he can't do that, as has been shown repeatedly. The longer the match, the worse the human seems to play.

So, again, computers win, but not for the "positional excellence" that is being attributed. Even Rybka has a very simple evaluation as has been revealed in the past... It relies on speed, a good search, and a solid evaluation to play at the level it plays at...