Moderator's Note: The new title of this thread is an acronym for:

"In which we explore the subtle vaguaries of computer chess, the ELO system, and other largely philisophical related topics.

Or:

ELO, HOW DOES IT WORK???"

__________________________________________________ _____________________

In Game1 1997

Why did deep blue play 38...kg7 and not 38...rd6
The deep blue software was very weak IMO.
I'm not sure how Kasparov would of won after 38...rd6


FEN: 3rrk2/8/2p3P1/1p2nP1p/pP2p3/P1B1NbPB/2P2K2/5R2 b - - 1 38


Uploaded with ImageShack.us

I have no idea why one move or another would be better in this position, but imo a computer program making a mistake in 1997 is not a revelation that requires all caps in the topic.

What does "very weak" mean? Compared to today's software, compared to Kasparov, compared to your expectations, compared to yourself?

Upon further looking at the position at hand, probably it's horizon was too short to see the danger in putting himself in the pin and should have tried to sac the knight for the two pawns (which it probably underestimated).

Well in this case one move might lead to an unclear position and the other might lead to a loss.
Why would a computer that see's 200 million moves per second make such a simple mistake.
If there was a mate in one I would expect deep blue to see it everytime.
Playing 38...kg7 is like missing a mate in one hence the caps

This just isn't true.

Kasparov made a big fuss about this match, he even created a movie.
In it he alluded that the IBM team cheated him.
Yet he doesn't mention this game in the movie
A computer is as much likely to miss a mate in one than to play 38...kg7 in this position
My only explanation is the computer wanted to play 38...kg8 and the operator put the wrong move in.
He said he missed a draw in game2, but in this game deep blue's operator gifted him the win

So is Kg7 bad because of g4?

This is, basically once again, just not true. Saying it twice doesn't make it so. It doesn't lose in 1-ply, therefore there could be many things that the computer was looking at where it got it wrong. Missing a mate in 1 is much much less likely for a computer or even a human.

yes, its bad because after g4 black loses

I put this in Houdini last night and at depth 14+ it recognizes that Kg7 is bad (about a +1.2 advantage that steadily increased to 1.9 by a depth of 19). Up until depth 13, though, it was neck and neck between Kg7 and other better moves. It doesn't seem very strange to me that a computer in 1997 might have evaluated the position the same as the best engine we have today does at depth 13.

Ok, so you spotted Kg7 as a mistake and immediately saw that black can defend with 1-Rd6?

How come you never took part in the supertournaments and showed these guys how things should be done?

Err... is this directed at me? I said I put it into Houdini and let it analyze the position. I have no idea why Kg7 is a mistake compared to Rd6, other than that you've stated it as a fact and Houdini (at sufficient depth) agreed, and apparently it has something to do with g4.

All I'm doing is comparing computer analysis using the best engine available today with what Deep Blue played, and the Deep Blue move is consistent with what the modern engine says at relatively, but not extremely, low depth. Therefore I don't find it shocking that Deep Blue would have made the move even at much higher depth, considering that engines have progressed a ton in the last 14 years. However the modern engine, upon looking deeper, also confirms that the Deep Blue move was a mistake. I'm not sure what else you're looking for?

I could post the depth 12 line from Houdini that shows Kg7 as the best move, and we could look through it to see what the engine is overlooking (until it looks deeper). There's a good chance Deep Blue's algorithms might have overlooked the same thing, and that might explain what Deep Blue was "thinking" when it made the move? Or it might explain nothing, since they're entirely different engines.

pretty sure he meant OP.

I was replying to OP who called deep blue weak, Kg7 a simple mistake and like missing mate in one of course.
It goes without saying that a 14 year old computer with a playing strength of "only" 2800 cant match Houdini today who have 3200 something.

Had it been directed to you, I would have quoted you.

That makes much more sense, lol. The (only?) downside of posting at work is that it can cause one to read posts with a little less care, and perhaps occasionally misunderstand the obvious. Sorry

My computer gives kg7
It then gives rd6 after 6 seconds of thought
My computer does 3.6 million positions per second
Remember Deep Blue was doing 200 million positions per second in 1997!
This move just shows how weak the software was 14 years ago.
I think deep blue was set to do a brute force search in the first game of the match, hence it playing this extremely weak move.


17/53 00:04 15,300,207 3,681,000 -0.52 Kf8-g7 g3-g4 h5-h4 Bh3-g2 Kg7-f6
17/53 00:06 22,025,023 3,679,000 -0.10 Rd8-d6 Rf1-e1 Rd6-d8 Bh3-g2 Ne5-g4+

Its fairly well known too that Deep Blues main strength was due to the massive calculating ability. The software was fairly weak compared to the best chess programs in 1997 like Genius, Junior, Hiarcs, Fritz and Mchess.

Two things:

1. Deep Blue was not 2800 strength. Kasparov basically choked.

2. I doubt that even Houdini is 3200 strength, assuming we're talking about FIDE.

First, we really can't know what any engine's FIDE strength is, because the ELO rating system is purely relative. The top engines don't play enough serious games against humans to estimate their ELO strength accurately. They're clearly stronger than the best humans, but we can't say much beyond that. Second, if Houdini were 3200 ELO, that would put it 400 points above the top five humans. This would mean that if a top-five player were to play a serious match at full strength against Houdini, they would score about 9%. I doubt that if you gave one of the top five players a strong incentive to play their hardest they would score less than 20% or so against Houdini in a match.

First of all. Deep Blue from 1997 probably had 2800 since it beat Kasparov although its hard to prove for sure since it never played enough games.
What is known however is that the best PC-programs at that time had almost 2550-2600Elo and Deep Blue was far stronger than that.

2, All serious rating lists have the best computers at 3200+ so youre fairly alone in your opinion or its up to you to prove otherwise if you dont think so:

http://ssdf.bosjo.net/list.htm

Also, the FIDE rating system is pretty far from optimal when dealing with players with huge rating differences. If a program plays a match with a top player today, its basically a struggle to see if the top player can hang on to a draw or not. There is no way he can win.
Usually when the human player is white (The fide rating system does not even take into account the advantage of having white.) his best bet is probably to choose a drawing line rigth in the opening.

Engines do not only try to draw the game. Humans could.

The engine will be happy with a draw if the choice is to play on with a -0,01 pawns worse position... (Depending somewhat on its settings of course.)

Engines do not avoid tactics.

I dont understand what the **** you are talking about.
A computer basically never choices a sligthly worse but complicated position in favour over a forced draw.

The computer also never chooses to play a simpler line that exchanges a lot pieces if it is not the best one.

The ELO rating system is relative, not absolute. Very few serious tournament games are played between humans and computers, which means that the engines form a virtually isolated player pool. Thus, human and computer ratings are not comparable. Deep Rybka 4's rating of 3216 is a predictor of its performance against other engines in its rating pool--nothing more, nothing less.

Of course, I acknowledge the superiority of the top engines. However, to quote an engine's ELO rating when discussing the relative strength of humans versus computers is just sophistry.

Yes, and basically all complications objectively favours the player with a tempo up (White.), which means that a human player with the white pices who knows the theory many many times can get a draw fairly easy vs a top engine...