Super GM blunder score will increase every year

[Dann Corbit]

I see it all the time.
One program will think things are even and the other program sees +3 pawns.
It is not hard to find positions like these with this:
http://home.pacific.net.au/~tommyinoz/gameanalyser.html

This happens particulary when the gap between the two programs is rather big in terms of strength....but there are a lot of exceptions of course....

Ok Doc. You did it again.
You said you really wanted me to let you be my friend, and I am trying.
I did not say BamBam vs Rybka 3. I did not say duo core vs quad core.
Of course I meant fairly equal hardware and a fairly equal program.
Like I requested from Dan. Please post a one move blunder by a top program in tournament (long) time control with decent hardware.
I posted one the other day, I will try to find it again.

kgburcham

I see them all the time. Unfortunately, I rarely keep my Winboard debug logs.

Now, by blunder, I mean that program A sees clearly a winning move and program B has no idea it is coming or program A makes a bad move and program B jumps all over it. You will see it (though very rarely) even in a big mismatch like some program 150 Elo below Rybka playing Rybka and Rybka is stunned.

[Rolf]

I see it all the time.
One program will think things are even and the other program sees +3 pawns.
It is not hard to find positions like these with this:
http://home.pacific.net.au/~tommyinoz/gameanalyser.html

This happens particulary when the gap between the two programs is rather big in terms of strength....but there are a lot of exceptions of course....

Ok Doc. You did it again.
You said you really wanted me to let you be my friend, and I am trying.
I did not say BamBam vs Rybka 3. I did not say duo core vs quad core.
Of course I meant fairly equal hardware and a fairly equal program.
Like I requested from Dan. Please post a one move blunder by a top program in tournament (long) time control with decent hardware.
I posted one the other day, I will try to find it again.

kgburcham

I see them all the time. Unfortunately, I rarely keep my Winboard debug logs.

Now, by blunder, I mean that program A sees clearly a winning move and program B has no idea it is coming or program A makes a bad move and program B jumps all over it. You will see it (though very rarely) even in a big mismatch like some program 150 Elo below Rybka playing Rybka and Rybka is stunned.

Dann, something is wrong in all the above. Therefore I wanted to explain it to kgburcham, but he's too impolite.

Perhaps I can clarify this with you at least.

Look please, I dont deny that such things happen like you described with one prog showing this or that and another prog has a naive output on the display. But can we call this "blunder"? I hope we agree that this is a different blunder than the usual human-like blunder.

Now let me please ask you what a single displayed value means? Is it always a typically (wrong) chess-wise evaluation of that program? Why do you think so?

I see many other more tech depending reasons for such findings. It's not blunder then. Although you can call it blunder if you insist. But then we should differentiate between a chess blunder and a tech blunder.

Apart from all this it's known that some progs are intentiously tuned this way and to then conclude that with rising depth the "blunders" will also increase, this is a fallacy. Because that would now mix up chess and tech blunders.

What I want to say is mainly that there dont exist one move chess blunders in computerchess and if you see this happening it's either a bug or a technical "confusion" out of whatever, you name it, hash, multi-processor or other parameters.

NB that I react on pure CC examples from play. As Burcham had mentioned the match between Deep Junior and Shredder. Read above on page one. Then it seems that the basic topic has been confounded with human "blundering" as computer programs perceive the situation. But this is a different topic. But you had answered this by stating that one reason were the rising depth of machine evaluation which then resulted in higher (negative) scores. But this is about humans played blunders and NOT one-move-blunders by programs!

[S.Taylor]

Here is another Super GM 4 point blunder in the latest tournament.
As hardware and programs improve every year these blunders will just keep getting higher in value. For example this same blunder might show 10 points in 2010. Or maybe a mate eventually.
I am sure some here think that they realized how Gm have blundered since the cavemen played chess. I have not realized how often GMs blunder untill the last few years using programs. I am amazed how often a 2700 player will blunder.

This subject means alot to me, as I have been increasingly observant about how most people, though some more than others, asses a human.
OK, so a trained human can be a proffessional at certain endeavors. That's very good. He must never make mistakes, especially dentists and other such jobs on which we depend on a human for dear life!!!
But this "non mistake making" should be accompanied by great patience and deddication and realization of responsibility to a patient, as per any kind of doctor, or plane pilot with passengers etc. And even though self-confidence can also help the positive mental aptitude, it still needs to go together with the realization of human inadequacy, and hoping for G-d's continued help.
If not for G-d, a human cannot do a single thing!
So I believe that this is a lifetime work to always be aware of this.
We must indeed take responsibilities. But part of that, is inner humility.
And the need for this is becoming more and more apparent via chess vs computer chess! So this is very valuable!

[BubbaTough]

The word blunder is very confusing to me. People talk about the 'point swing', but the number of 'points' is in accordance to an arbitrary scale chosen by the programmer. The only real importance in judging a move is does the move transfer the theoretical outcome of the game (the game was drawn but after that move now white wins...or white was going to win but after that move now black is going to, etc.). Assumably, the higher the 'point swing' the larger the probability of a change in theoretical outcome. The largest point swings are usually when transposing into a situation where the computer sees mate (for example, going from a draw to a tablebase assisted mate in 120).

All this said, if we want to label a blunder as something that changes the outcome of the game assuming best play, every game won or lost must have a blunder in it (assuming the game starts out drawn). The only reason some games don't have large point swings is if neither side is sufficiently good at calculating/understanding the game to spot where the decisive change in the theoretical status of the game was. When you get better computers/programs/humans/goats and look over the existing 'blunder free' games, you will see large 'point swings' as the entity conducting analysis is better at spotting when the theoretical transition in outcome occurs.

Now perhaps (in fact almost assuredly) there are people that don't think a blunder should be defined as a change (always for the worse, a move cannot change the theoretical status of the game for the better) in the theoretical status of the game. In that case, we need a better definition then of blunder to achieve some consensus on whether GMs really 'blunder' all the time. I would posit that such a definition should not be linked to computer evaluation, since people were blundering far before computers were around.

-Sam