Top engines without tablebases don't play as good as top GMs

[pichy]

Endgame play can involve very long-range planning, and computers are still not terribly good at that.

On the other hand, many games, including those at GM level, are decided by mistakes that no modern engine would make.

And many times pre-computer GM analysis of endgames is faulty or incomplete.

--Jon

I agree that many non top GMs make mistakes, but I would put my money on GM Aronian or Kramnik against any of the top 5 engines without Nalimov or any other tablebases to play 6 ending positions in 10 minutes per move.

There is no question the top programs are far superior to even the best humans in the overall game and ELO.

I would not say in all stages, ....

Yes, that is what I said too but you cut that part.

... the only reason why top engines can easily beat Top GMs is because first they have an intensive opening and in the middlegame they hardly make mistakes;but if you take the tablebase away and present any top engines with 10 positions played by lets say Capablanca vs Alekhine or Tarrasch vs Thorold or GM Kramnik versus Kasparov and provide the engine and Kramnik or Aronian with 10 minutes per move the GM will definitively beat the top engine provided that the position are equal in material.

There is no single "only reason" - there are many reasons now that computers are superior. The top computers now just outplay the top Grandmasters period. I have played over many games of top Grandmasters (using Komodo or other programs) and what I see is that the one or both players make what the computer considers a slightly weak move, the score drops a little but not fatally, but then it happens again, and then again and pretty soon it's a lost game. In general it will be several inaccuracies. In many game the winning player returns the favor with inaccuracies of his own but sooner or later one of the players prevails.


You played over many games with top Grandmasters versus Komodo and you probably noticed that by the time the game has reached the endgame stage Komodo has already a won position due to the tactical superiority of your engines, and the rest of the game should be very easy for Komodo, but what you have not done is to give Komodo several positions that were played by great endgame players like Capablanca or Tarrasch where the position is equal and match those endgame positions against top GM like Kramnik or GM Aronian providing ten minute per move to avoid losing by a blunder since most endgames blunders by top GMs were caused due to time trouble when the players that blundered was running out of time.



It used to be said that computers are all about "tactics", but tactics is really a human term which to generalize a bit means they "don't miss anything within their horizon" and these days tactics is also positional play. An opportunity to make a pawn slightly weaker is not "tactics" or "combination play" in the traditional sense but it is what computers do much better than humans. Or it could be "combination" to free a cramped position that the human missed. Humans are still superior in "understanding" what cannot be seen or directly computed but that doesn't buy much any longer.

As far as openings are concerned, modern opening theory is based a LOT on computer analysis, and a couple of centuries of human analysis do not provide much that computers cannot figure out for themselves in seconds or even less than a second. And humans would be poor in the opening if they didn't study and memorize opening like crazy, so I see no justification for considering that computers should be forbidden to consider in advance what they should play while thinking that is natural for humans to do. How is that reasonable?

Databases make very little difference in the strength of the programs. At best perhaps a tiny boost, but there is a school of thought that they play worse with databases too due to overheads and such.

I can present you with more than 10 positions played by Capablanca or alekhine, or any other top endgame elite where the engine don't not select the correct move withouth using tablebases.

Of course you can cherry pick positions like this, but the only thing that matters is the complete game. If you could start human/computer matches from positions which human understand better of course you will find that humans will do better.

Nobody questions the fact that computers have certain weaknesses. Even Larry who is very pro-computer says in closed positions computers play like weak club players. What you fail to consider is that one could take some other subset of the game and make a very strong case that humans cannot play chess very well and computers will put them to shame because they have too many weaknesses. It would be just as valid to say that. But none of it really matters except actual results.

I cannot help but think of the Monty Python "Life of Brian" movie where one of the characters cried, "what have the Romans ever done for us?" His buddy said, "well they did make a difference with the crime rate", then it was "other than the crime, what have the Romans ever done for us?" Then it was the sanitation, the roads, etc. Finally he had to say, "other than the roads, the sanitation system, the crime, the aqueduct, education, irrigation system, medicine, clean water, WHAT HAVE THE ROMANS EVER DONE FOR US?"

This reminds me of that because you are systematically claiming that the only reason computers are better is because of the very things they are good at, and yet that it's somehow not relevant. Oh sure, if you take away the book, the endgame databases, the human errors, limit the game to positions humans are better at, give humans 10 minutes per move, THEN LET'S SEE HOW WELL THE COMPUTER DO!!! Is there any other advantages you want to take away from computers to prove they are actually inferior?

[Don]

Are you serious? Kasparov use to play these simple positions in his sleep, so did many of the elite past and present.

How many super grandmasters post here? None?

To advance the discussion beyond a shouting contest, let's discuss the "Tarrasch" point of view that Re3 is superior to Re7+, a view that is not shared by Houdini.

Can you explain the following:
1) How does white win after 1.Re3 if black replies Kd6 and Rd5.
2) How does black have "good drawing chances" after 1.Re7+ Kd6 2.Rxg7 Rxd3 3.h4 Rd1+ 4.Kh2 Rd2 5.Rxh7 Rxb2 6.h5 - Houdini thinks that the final position is a straightforward win for white.

Robert

I think players tend to "over-romanticize" top players. When I was a kid I read books that made it seem like these guys were practically god's.

We have the ELO system which does a fair job of quantifying the skills of the top players and hopefully that demystifies them somewhat. But whether people like it or not, computers have passed them (and it's been a few years already) and the gap will only get wider.

Don

Computers have passed humans only in tactics. Planning and , generally speaking, strategy are on humans' side.

Computer have also passed humans in ELO.

Often computers fail endgames in which you need a right plan more than a long calculation.

Regards

Computer outplay humans in endgames too. There are specific weaknesses computer have which sometimes show up, but it's not enough to prevent the large ELO superiority they have over humans.

[Don]

I think players tend to "over-romanticize" top players. When I was a kid I read books that made it seem like these guys were practically god's.

If a Steinitz or Lasker were to play a top-five engine on a cluster in a say 100 game match, would the score for the first half of the match be different from the score for the second half?

I believe these players would lose badly but I think they would improve in the second half of the match. Probably they would figure out what to do to not lose quickly and give themselves a chance for an occasionally draw.

The superiority is not so great that a human won't win a game once in a while, but it's substantial.

[Dragulic]

Steinitz or Lasker

The superiority is not so great that a human won't win a game once in a while

So you believe the ELO model breaks down here?
A cluster Rybka performs above 3400. Steinitz performs below 2600.
"once in a while"....

[Mike S.]

1349: Tarrasch,S - Thorold,E, Manchester 1890
4R3/2k3pp/pp6/8/3r4/3P4/PP4PP/6K1 w - - 0 26

Analysis by Fritz 13 (Atom N455/1.66 GHz only):

26.Re7+ Kc6 27.Re3 Rd7 28.Kf2 Kd5 29.Rf3 Rc7 30.Ke3 g6 31.Rf8 Rc2 32.Rf2 Rc1 33.Rf7 Re1+
+- (1.89) Depth: 10/21 00:00:00 79kN
+/= (0.39) Depth: 14/31 00:00:09 3941kN
26.Re3 Rd5 27.Kf2 Kd6 28.Rg3 g6 29.Rf3 Rb5 30.b3 Rc5 31.Ke3 Kd5 32.Rf7 Rc2 33.Rd7+ Ke6 34.Rxh7 Rxa2 35.g4 Kd5 36.Rd7+
+/= (0.58) Depth: 14/31 00:00:14 6128kN
+/= (0.59) Depth: 18/33 00:00:51 22007kN, tb=4

Siegbert Fritz Tarrasch.

(Note that Fritz already decided for Re3 before a few tablebase hits happened.)

MultiPV(3) from empty hash:

Analysis by Fritz 13, after depth 17 was finished:

1. +/= (0.61): 26.Re3 Rd5 27.Rg3 g6 28.Rh3 Rd7 29.Kf2 Kc6 30.Rf3 a5 31.Ke3 Re7+ 32.Kd2 Kd5 33.Rf8 a4 34.Rb8 Kc5
2. +/= (0.47): 26.Re7+ Kd6 27.Re3 a5 28.Kf2 Rd5 29.Re8 Rc5 30.Kf3 Rc2 31.Re2 Rc1 32.Ke4 Ra1 33.b3 Rh1 34.g4 Rd1
3. +/= (0.46): 26.Rh8 h6 27.Rh7 Rxd3 28.Rxg7+ Kc6 29.Rg6+ Kc7

However, after some interactive analysis:

Analysis by Fritz 13:

1. +/- (1.12): 26.Re7+ Kc6 27.Rxg7 Rxd3 28.h4 Rd1+ 29.Kh2 h5 30.Rg6+ Kc7 31.g4 Rd5 32.Rg5 Rd2+ 33.Kg3 Rd3+ 34.Kf4 hxg4 35.h5 Rd2 36.h6 Rxb2 37.h7 Rh2 38.Rg7+ Kd6 39.Kxg4
2. +/- (0.76): 26.Rg8 Rd7 27.Kf2 g6 28.Ra8 Kb7 29.Rh8 Kc6 30.Ke3 Re7+ 31.Kd2 Rf7 32.Kc3 Rf2 33.Rxh7 Rxg2
3. +/= (0.66): 26.Re3 Kd6 27.Rh3 h6 28.Rg3 Ke5 29.Rxg7 Rxd3 30.Re7+ Kf6 31.Re2 Kf5 32.Kf2 h5 33.g3 Rd1 34.Ke3 Kg4

I also tried some other engines, and conclude: Re7+ is ok, and Re3 is somewhat artsy...

[pichy]

1349: Tarrasch,S - Thorold,E, Manchester 1890
4R3/2k3pp/pp6/8/3r4/3P4/PP4PP/6K1 w - - 0 26

Analysis by Fritz 13 (Atom N455/1.66 GHz only):

26.Re7+ Kc6 27.Re3 Rd7 28.Kf2 Kd5 29.Rf3 Rc7 30.Ke3 g6 31.Rf8 Rc2 32.Rf2 Rc1 33.Rf7 Re1+
+- (1.89) Depth: 10/21 00:00:00 79kN
+/= (0.39) Depth: 14/31 00:00:09 3941kN
26.Re3 Rd5 27.Kf2 Kd6 28.Rg3 g6 29.Rf3 Rb5 30.b3 Rc5 31.Ke3 Kd5 32.Rf7 Rc2 33.Rd7+ Ke6 34.Rxh7 Rxa2 35.g4 Kd5 36.Rd7+
+/= (0.58) Depth: 14/31 00:00:14 6128kN
+/= (0.59) Depth: 18/33 00:00:51 22007kN, tb=4

Siegbert Fritz Tarrasch.

(Note that Fritz already decided for Re3 before a few tablebase hits happened.)



Please try it without tablebase, even if fritz decided does not mean that it will stick with it.


MultiPV(3) from empty hash:

Analysis by Fritz 13, after depth 17 was finished:

1. +/= (0.61): 26.Re3 Rd5 27.Rg3 g6 28.Rh3 Rd7 29.Kf2 Kc6 30.Rf3 a5 31.Ke3 Re7+ 32.Kd2 Kd5 33.Rf8 a4 34.Rb8 Kc5
2. +/= (0.47): 26.Re7+ Kd6 27.Re3 a5 28.Kf2 Rd5 29.Re8 Rc5 30.Kf3 Rc2 31.Re2 Rc1 32.Ke4 Ra1 33.b3 Rh1 34.g4 Rd1
3. +/= (0.46): 26.Rh8 h6 27.Rh7 Rxd3 28.Rxg7+ Kc6 29.Rg6+ Kc7

However, after some interactive analysis:

Analysis by Fritz 13:

1. +/- (1.12): 26.Re7+ Kc6 27.Rxg7 Rxd3 28.h4 Rd1+ 29.Kh2 h5 30.Rg6+ Kc7 31.g4 Rd5 32.Rg5 Rd2+ 33.Kg3 Rd3+ 34.Kf4 hxg4 35.h5 Rd2 36.h6 Rxb2 37.h7 Rh2 38.Rg7+ Kd6 39.Kxg4
2. +/- (0.76): 26.Rg8 Rd7 27.Kf2 g6 28.Ra8 Kb7 29.Rh8 Kc6 30.Ke3 Re7+ 31.Kd2 Rf7 32.Kc3 Rf2 33.Rxh7 Rxg2
3. +/= (0.66): 26.Re3 Kd6 27.Rh3 h6 28.Rg3 Ke5 29.Rxg7 Rxd3 30.Re7+ Kf6 31.Re2 Kf5 32.Kf2 h5 33.g3 Rd1 34.Ke3 Kg4

I also tried some other engines, and conclude: Re7+ is ok, and Re3 is somewhat artsy...

[Mike S.]

Fritz 13 didn't stick to it anyway; see the last output in my first reply. Nor did any other engine I tried, e.g. Stockfish 2.2.2 (= no tbs.).

[MM]

Are you serious? Kasparov use to play these simple positions in his sleep, so did many of the elite past and present.

How many super grandmasters post here? None?

To advance the discussion beyond a shouting contest, let's discuss the "Tarrasch" point of view that Re3 is superior to Re7+, a view that is not shared by Houdini.

Can you explain the following:
1) How does white win after 1.Re3 if black replies Kd6 and Rd5.
2) How does black have "good drawing chances" after 1.Re7+ Kd6 2.Rxg7 Rxd3 3.h4 Rd1+ 4.Kh2 Rd2 5.Rxh7 Rxb2 6.h5 - Houdini thinks that the final position is a straightforward win for white.

Robert

I think players tend to "over-romanticize" top players. When I was a kid I read books that made it seem like these guys were practically god's.

We have the ELO system which does a fair job of quantifying the skills of the top players and hopefully that demystifies them somewhat. But whether people like it or not, computers have passed them (and it's been a few years already) and the gap will only get wider.

Don

Computers have passed humans only in tactics. Planning and , generally speaking, strategy are on humans' side.

Computer have also passed humans in ELO.

Computer have a different elo. You can't find a elo rating that includes humans and machines and even if it should exist it would be wrong for a simple reason: to have a rating list including humans and machines, humans and machines should play each other continuosly, so the comparison would we right. That's not the actual situation.

Regards

[Kingghidorah]

Endgame play can involve very long-range planning, and computers are still not terribly good at that.

On the other hand, many games, including those at GM level, are decided by mistakes that no modern engine would make.

And many times pre-computer GM analysis of endgames is faulty or incomplete.

--Jon

I agree that many non top GMs make mistakes, but I would put my money on GM Aronian or Kramnik against any of the top 5 engines without Nalimov or any other tablebases to play 6 ending positions in 10 minutes per move.

There is no question the top programs are far superior to even the best humans in the overall game and ELO.

Databases make very little difference in the strength of the programs. At best perhaps a tiny boost, but there is a school of thought that they play worse with databases too due to overheads and such.

There are area's of course where each is far superior, for example closed positions are difficult for computers. But from the standpoint of a 10 game match for instance without databases of any kind a Rybka class program will have an easy match victory against the very best the human race has to offer.

I agree

[pichy]

Are you serious? Kasparov use to play these simple positions in his sleep, so did many of the elite past and present.

How many super grandmasters post here? None?

To advance the discussion beyond a shouting contest, let's discuss the "Tarrasch" point of view that Re3 is superior to Re7+, a view that is not shared by Houdini.

Can you explain the following:
1) How does white win after 1.Re3 if black replies Kd6 and Rd5.
2) How does black have "good drawing chances" after 1.Re7+ Kd6 2.Rxg7 Rxd3 3.h4 Rd1+ 4.Kh2 Rd2 5.Rxh7 Rxb2 6.h5 - Houdini thinks that the final position is a straightforward win for white.

Robert

I think players tend to "over-romanticize" top players. When I was a kid I read books that made it seem like these guys were practically god's.

I believe that you "over-romanticize" your engine Komodo and you believe that top engines are invincible in all stages of the game, but in the endgame without tablebase they are still inferiors to the top human Gms like Arorian and Kramnik. I know that they make mistakes in some endings that engines will never make, but in most of those cases that they do is because they are in time trouble which will not happens if they are provided with a fix time control such 10 minutes per move. If Capablanca was alive I would bet you $10,000 per each ending setting versus your engine komodo without tablebase playing 5 or 6 endings in which the position is evaluated equal for both side with a time control of 10 minutes per move.

We have the ELO system which does a fair job of quantifying the skills of the top players and hopefully that demystifies them somewhat. But whether people like it or not, computers have passed them (and it's been a few years already) and the gap will only get wider.

Don

Computers have passed humans only in tactics. Planning and , generally speaking, strategy are on humans' side.

Computer have also passed humans in ELO.

Often computers fail endgames in which you need a right plan more than a long calculation.

Regards

Computer outplay humans in endgames too. There are specific weaknesses computer have which sometimes show up, but it's not enough to prevent the large ELO superiority they have over humans.