GM says Rybka & Fritz weaker than best GMs in classical

[Tony Thomas]

I still dont understand the whole point of limiting the search of the computer were as no limitation is applied to the search of the human...I think you are being slightly partial here..Is it really the computers fault that the human cant really see things very fast and does no search in a short time control?

[ozziejoe]

I don't know about 1 second per move, but there is clear evidence that human gm performance does not degrade much at short time controls (at least it degrades much less than you would expect). See Burns (2004) psychological science, chess and speed of processing.

Eg., that article concludes that 81% of chess skill is accounted for by how players performed with less than %5 of the normal time available. It looks like chess performance has more to do with quick pattern recognition processes than time consuming, deep search.


Here is another interesting quote

Gobet and Simon (1996) used ratings
to analyze nine simultaneous exhibition matches played by Gary
Kasparov against teams of four to eight weaker players. Because he
played each team member simultaneously, Kasparov had much less
time available than did each of his opponents. Gobet and Simon found
that Kasparov’s median rating based on his performance in these
matches was 2646, whereas his regular tournament rating at that time
was 2750. They stated, ‘‘In view of the slight extent to which lack of
time lowered the quality of Kasparov’s play in the simultaneous
matches, we conclude that memory and access to memory through the
recognition of cues is the predominant basis for his skill’’ (p. 54; but
see Lassiter, 2000, and Gobet & Simon’s, 2000b, response).

[Matthias Gemuh]

I don't know about 1 second per move, but there is clear evidence that human gm performance does not degrade much at short time controls (at least it degrades much less than you would expect).

That is true. The explanation is that positional play is the decisive element at short time controls. There is no deep tactics to penalize inaccuracies in that positional play.

At long time controls, positional play has to be much more accurate, otherwise deep tactics will dominate the game.

Matthias.

[Uri]

Top GMs have an eiditic memory and fantastic pattern recognition abilities. They can recall games and positions with extreme accuracy and in abundant volume. This is why they also play blind chess so good.

But still, in a 1 minute game the computer can outplay the GM. Fritz had 3500 in bullet on the playchess server which is 100 points higher than any GM who plays on the playchess server which demonstrates that computers are very strong in bullet.

[Uri]

That is true. The explanation is that positional play is the decisive element at short time controls. There is no deep tactics to penalize inaccuracies in that positional play.

At long time controls, positional play has to be much more accurate, otherwise deep tactics will dominate the game.

Matthias.

But isn't positional play itself based on tactics?

Positional chess is the same as strategy and is based on long term tactics. Without strategical play, a chess game is just a series of tactical tricks. A good strategy binds together the tactics, and enables a player to make methodical progress towards victory.

[Matthias Gemuh]

But isn't positional play itself based on tactics?

Positional chess is the same as strategy and is based on long term tactics. Without strategical play, a chess game is just a series of tactical tricks. A good strategy binds together the tactics, and enables a player to make methodical progress towards victory.

Positional play is based on tactics for engines.
For humans, it's the other way round : good positional play gradually leads to tactics.

Matthias.

[bob]

I still dont understand the whole point of limiting the search of the computer were as no limitation is applied to the search of the human...I think you are being slightly partial here..Is it really the computers fault that the human cant really see things very fast and does no search in a short time control?

There was no point given for playing games like that. My point was this hypothetical question:

given less than a second, with a non-tactical position, which move would you think better, one suggested by a computer, or one suggested by a top GM?

My point was that for non-tactical positions (AKA positional play) the GM will have a _much_ better quick grasp of the position and the plan required for that position. His positional judgement is _far_ better than a computers simple evaluation.

And before we go there, yes, in a tactical position, I would trust the computers move over the GM's, but that wasn't the subject, it was _positional_ skills and there is just no comparison between a computer and a top-level GM in that regard. Computers are unable to use all the games they have played in the past, in any effective way, to suggest plans in this position, while humans do so without thinking about it.

[Dr.Wael Deeb]

I still dont understand the whole point of limiting the search of the computer were as no limitation is applied to the search of the human...I think you are being slightly partial here..Is it really the computers fault that the human cant really see things very fast and does no search in a short time control?

There was no point given for playing games like that. My point was this hypothetical question:

given less than a second, with a non-tactical position, which move would you think better, one suggested by a computer, or one suggested by a top GM?

My point was that for non-tactical positions (AKA positional play) the GM will have a _much_ better quick grasp of the position and the plan required for that position. His positional judgement is _far_ better than a computers simple evaluation.

And before we go there, yes, in a tactical position, I would trust the computers move over the GM's, but that wasn't the subject, it was _positional_ skills and there is just no comparison between a computer and a top-level GM in that regard. Computers are unable to use all the games they have played in the past, in any effective way, to suggest plans in this position, while humans do so without thinking about it.

All you wrote is true here,no doubt,but....
We are talking about a classical game of chess,a match under long time controls and the computer using a cutting edge hardware,then all your theories are irrelevant here....


Computers are unable to use all the games they have played in the past, in any effective way, to suggest plans in this position

What about the learning function,even Crafty has one of the best learning features in the computer chess scene,RomiChess,another one with a revolutionary new abilities,so....
Don't forget that the time is near when a program will use a daatbase to find a position which is playing currently and it will use it much better and faster than the poor human being

[bob]

I still dont understand the whole point of limiting the search of the computer were as no limitation is applied to the search of the human...I think you are being slightly partial here..Is it really the computers fault that the human cant really see things very fast and does no search in a short time control?

There was no point given for playing games like that. My point was this hypothetical question:

given less than a second, with a non-tactical position, which move would you think better, one suggested by a computer, or one suggested by a top GM?

My point was that for non-tactical positions (AKA positional play) the GM will have a _much_ better quick grasp of the position and the plan required for that position. His positional judgement is _far_ better than a computers simple evaluation.

And before we go there, yes, in a tactical position, I would trust the computers move over the GM's, but that wasn't the subject, it was _positional_ skills and there is just no comparison between a computer and a top-level GM in that regard. Computers are unable to use all the games they have played in the past, in any effective way, to suggest plans in this position, while humans do so without thinking about it.

All you wrote is true here,no doubt,but....
We are talking about a classical game of chess,a match under long time controls and the computer using a cutting edge hardware,then all your theories are irrelevant here....


Computers are unable to use all the games they have played in the past, in any effective way, to suggest plans in this position

What about the learning function,even Crafty has one of the best learning features in the computer chess scene,RomiChess,another one with a revolutionary new abilities,so....
Don't forget that the time is near when a program will use a daatbase to find a position which is playing currently and it will use it much better and faster than the poor human being

No they are not irrelevant. I gave an example where IM Mike Valvo blew deep thought out in a correspondence match, the game was played on r.g.c.c back somewhere around 1990 or so so it is something that can be found. Yes today's computers are better than 1990 deep thought. But top GMs are at least equally better than Mike was at chess skill.

Computers are winning games based on strong tactics and average positional skills. The difference between a computer and GM can be characterized as follows:

1. Positional skills favor the GM by a significant margin. GMs know where weaknesses will be important, where they are not, which wing should be exploited, when to attack, when to defend, when to trade, when to not trade, etc. Computers have general rules for the above but there are plenty of exceptions that a strong GM will understand far better. Lock up a position (typical anti-computer play) and the computer is helpless, while a human will think "OK, I have one possible pawn break _here_ but I need to get everything into an optimal position to take advantage of the break once I play it. It might take 20 moves to do this and since a computer can't see the end result, it won't even start down the road, unlike a GM. That's a significant edge to the GM.

2. Tactical skills favor the computer although not in the way you think. A GM can _way_ out-calculate a computer along sharp tactical lines for the most part, they can simply search far deeper than our current full-width searches can reach. But here's the rub, the deeper the tactics, the more side-paths there are and humans tend to make mistakes along these paths as they get distracted, tired, etc. In short, a computer sees almost _everything_ that is within its search horizon, which gives it an advantage. But unfortunately that search horizon is not as deep as a GMs horizon in lots of positions, so the two tactical abilities are somewhat different.

3. Then there are positions like Shirov's bishop sac several years ago that was based both on deep analysis and also on positional considerations, and it was proven to be a forced win later, but computers had no clue until they were given enough time to see all the way to the end or they were forced to follow the winning line until they reached that point.

The problems humans face today against computers is that the computers don't miss much that happens within their search horizon tactically. And that horizon has become fairly distant. Humans can still out-calculate the computer, but it takes extreme effort to do it move after move, and somewhere along the way there is a good chance that the human will make a mistake and get jumped on. If we could somehow take a human brain and convert it into an electronic equivalent, which would eliminate all the weaknesses, I would rely on the electronic human completely and programs would have great difficulty winning games when the 'human factor' is eliminated and fatigue has no effect on the game.

That is where we are today. And as the depths continue to increase as hardware improves, the program's "no mistake horizon" keeps getting pushed farther away. And probably one day this "horizon" will be pushed beyond the point where any human player can reach it, giving the computers almost total dominance in chess. But that has not happened yet and it isn't close. Fortunately for computers, humans are relatively frail and they are making enough mistakes to have difficulties as a match becomes fatigue-city...

As far as learning goes, I have removed positional learning from crafty completely. It simply doesn't work. A human can look at two positions and if he knows one is a win, and the second is the same except for one pawn on a different square which has no effect, he generalizes that the second is won as well. Not computers. How many positions do you think one would need to memorize to have decent coverage in a chess game? for humans, thousands of patterns is enough. for a computers, we would need tera-terabytes of data to begin to approach what a human's ability to generalize/extrapolate can accomplish with much less data...

Book learning works, but we are not talking about learning which openings seem to be better or worse...

[Matthias Gemuh]

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.