Could Carlsen beat engine - Chess Forums - Page 3

GnrfFrtzl

Mar 11, 2015

0

#41

TheGreatOogieBoogie írta:

GnrfFrtzl wrote:

TheGreatOogieBoogie írta:

Yes! He's the best player in the history of the world and machines don't know how to plan. He can steer the game into channels where human thinking trumps brute force calculation. While he can't calculate nearly as fast a an engine engines have a horizon effect of 20 moves whereas Carlsen can see deeper than that, resulting in a more accurate assessment.

Not trying to diss the guy, but how on Earth could Carlsen see deeper than an engine?
I understand the difference between the human thinking and pure engine, but there's no human that can see 20 moves deep.

Alekhine sometimes looked 30 moves ahead as shown sometimes in his best games collection.

There is absolutely no way that anyone can look 20-30 move deep.
10, I can believe, but even that is extremely rare.
If you're talking 20 move deep, you're talking thousands of lines and variations.
No human can see that.
If you give them a pen and paper and a few hours, fine, I'll buy it.
But not without those.

HilarioFJunior

Mar 12, 2015

0

#42

GnrfFrtzl wrote:

LegendLength írta:

Out of interest is it possible to calculate the maximum rating possible for either fide or some other system of ratings?

I'm sure you could do it for a game of 1 king and 1 queen each, so it should be extendable for a normal game right? Or would you have to work it out empirically by making a perfect engine first?

I'm sure there is no maximum rating.
Ratings will go on climbing higher and higher until someone works out a better rating system.

There's a linear correspondence between the quality of moves from a player and his elo rating. Hence there's probably a maximum rating. According to some researchers it's 3571.

http://www.cse.buffalo.edu/~regan/papers/pdf/Reg12IPRs.pdf

HilarioFJunior

Mar 12, 2015

0

#43

TheGreatOogieBoogie wrote:

GnrfFrtzl wrote:

TheGreatOogieBoogie írta:

Yes! He's the best player in the history of the world and machines don't know how to plan. He can steer the game into channels where human thinking trumps brute force calculation. While he can't calculate nearly as fast a an engine engines have a horizon effect of 20 moves whereas Carlsen can see deeper than that, resulting in a more accurate assessment.

Not trying to diss the guy, but how on Earth could Carlsen see deeper than an engine?
I understand the difference between the human thinking and pure engine, but there's no human that can see 20 moves deep.

Alekhine sometimes looked 30 moves ahead as shown sometimes in his best games collection.

Yes, Carlsen and Alekhine (as well as superhumans) can see many moves ahead, but they can't see objectively as an engine. Well, even engines can't see everything due to pruning! Actually when they say that "Carlsen can see 20 moves ahead" it means that he can intuitively see some final positions after 20 moves, but not every one, not clearly as an engine would do, and without brute force.

MervynS

Mar 12, 2015

0

#44

Highly doubt any of the best players of all the eras would do well against Stockfish, Houdini or this new positional engine Komodo I see on ChessBase.

I can see humans getting an edge in the opening and middle game in enough games. The problem with computers however, since they calculate so well, is their defensive tenacity. I also think engines also tend to be very objective unless tweaked, if simplifying is clearly the best play, they will simplify even though winning chances may disappear.

GnrfFrtzl

Mar 12, 2015

0

#45

HilarioFJunior írta:

GnrfFrtzl wrote:

LegendLength írta:

Out of interest is it possible to calculate the maximum rating possible for either fide or some other system of ratings?

I'm sure you could do it for a game of 1 king and 1 queen each, so it should be extendable for a normal game right? Or would you have to work it out empirically by making a perfect engine first?

I'm sure there is no maximum rating.
Ratings will go on climbing higher and higher until someone works out a better rating system.

There's a linear correspondence between the quality of moves from a player and his elo rating. Hence there's probably a maximum rating. According to some researchers it's 3571.

http://www.cse.buffalo.edu/~regan/papers/pdf/Reg12IPRs.pdf

'There's probably' is not a definite.
Well, we're just theorisin.
I'm using the farmer's common sense:
"Just when you think an end hasben' reached, there comes a kid that shows a new way."

vittyvirus

Mar 12, 2015

0

#46

GnrfFrtzl wrote:

I understand the analogy, but I still think it's inaccurate.
A computer can store million times more data, can calculate deeper and faster, and also lacks any kind of psychological disadvantage.
The only times I've seen humans beating engines is when they use 'anti-engine' methods and create weird lines that a computer simply chooses not to follow.

Of course he can with proper training and opening choice. There are 2000sh people who have beaten Stockfish easily using an opening of their choice (usually Kings Indian) and anti-commputer tactics plus playing positions where planning is more important.

vittyvirus

Mar 12, 2015

0

#47

Wernher-von-Braun wrote:

Justs99171 wrote:

mosey89 wrote:

DarkVlader wrote:

Both engines are above 3300. Carlsen is below 2900.

This is incorrect, engine ratings are not FIDE ratings therefore they are not comparable.

Having said that, its unlikely that a human player can defeat a strong engine in a classical game. However in a correspondence game I expect most GMs could defeat an engine.

I don't buy into any entity having a 3000+ rating. We've seen Fischer play almost perfect at 2785 and Kasparov play almost perfect at 2851.

This couldn't be further from the truth. Carlsen, Fischer, Kasparov, etc, made lots of mistakes. They just played better than their opponents. Ratings inflate, yes, but that has nothing to do with comparing two people of the same era. Rating inflation matters when you are comparing people (or engines) from different eras, such as comparing Fischer's rating to Carlsen's.

3300 is their CCRL rating. CEGT rates them below 3100. SSDF rates them 3400+

Jion_Wansu

Mar 12, 2015

0

#48

Yes he can. Here is why. All engines play the exact same move order if you play the exact same move order against it in previous games. Engines can't think

vittyvirus

Mar 13, 2015

0

#49

Jion_Wansu wrote:

Yes he can. Here is why. All engines play the exact same move order if you play the exact same move order against it in previous games. Engines can't think

No they don't. There is a issue called "Search Instability" in chess engines. Theoritically they should report exactly the same variation, node counts etc. for the same position searched twice. You can, pick up Stockfish, let it analyse the starting position to a good depth, then restart and search the same position again with Stockfish to the same depth. I bet you'll see a different variation, if not a different best move. For example see this with Houdini 4 Pro:

FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
Houdini 4 Pro:
20/41 00:10 33,385k 3,240k +0.18 d2-d4 d7-d5 Bc1-f4 Bc8-f5 e2-e3 Nb8-c6 Ng1-f3 e7-e6 Bf1-d3 Bf5xd3 Qd1xd3 Ng8-f6 Nb1-c3 Bf8-d6 Bf4xd6 c7xd6 a2-a3 Qd8-b6 Nc3-b5 Nf6-e4 O-O O-O c2-c4 d5xc4 Qd3xc4

FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
Houdini 4 Pro:
20/41 00:09 31,384k 3,198k +0.23 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 O-O Bf8-c5 Nf3xe5 Nc6xe5 d2-d4 a7-a6 Bb5-a4 b7-b5 d4xe5 Nf6xe4 Ba4-b3 Qd8-h4 Bc1-e3 O-O Be3xc5 Ne4xc5 Bb3-d5 Ra8-b8 Nb1-c3 Bc8-b7 Qd1-f3 Bb7xd5 Nc3xd5

And if YOU search to depth 20, your node counts, score, and the varition will be different.

NM Dale

Mar 13, 2015

0

#50

Is Carlsen familiar with the A.R.B system?

vittyvirus

Mar 13, 2015

0

#51

Wernher-von-Braun wrote:

vittyvirus wrote:

Jion_Wansu wrote:

Yes he can. Here is why. All engines play the exact same move order if you play the exact same move order against it in previous games. Engines can't think

No they don't. There is a issue called "Search Instability" in chess engines. Theoritically they should report exactly the same variation, node counts etc. for the same position searched twice. You can, pick up Stockfish, let it analyse the starting position to a good depth, then restart and search the same position again with Stockfish to the same depth. I bet you'll see a different variation, if not a different best move. For example see this with Houdini 4 Pro:

FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
Houdini 4 Pro:
20/41 00:10 33,385k 3,240k +0.18 d2-d4 d7-d5 Bc1-f4 Bc8-f5 e2-e3 Nb8-c6 Ng1-f3 e7-e6 Bf1-d3 Bf5xd3 Qd1xd3 Ng8-f6 Nb1-c3 Bf8-d6 Bf4xd6 c7xd6 a2-a3 Qd8-b6 Nc3-b5 Nf6-e4 O-O O-O c2-c4 d5xc4 Qd3xc4

FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
Houdini 4 Pro:
20/41 00:09 31,384k 3,198k +0.23 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 O-O Bf8-c5 Nf3xe5 Nc6xe5 d2-d4 a7-a6 Bb5-a4 b7-b5 d4xe5 Nf6xe4 Ba4-b3 Qd8-h4 Bc1-e3 O-O Be3xc5 Ne4xc5 Bb3-d5 Ra8-b8 Nb1-c3 Bc8-b7 Qd1-f3 Bb7xd5 Nc3xd5

And if YOU search to depth 20, your node counts, score, and the varition will be different.

Nice post, but um remind me to never play a correspondence game with you, lol.

I don't play correspondence anymore, because I don't have consistent internet, and I find chess programming more interesting.

Jack_of_Clubs

Mar 13, 2015

0

#52

Would someone like Carlsen be able to gain any kind of advantage by studying how Stockfish is programmed/coded in preparation for a match?

vittyvirus

Mar 13, 2015

0

#53

Jack_of_Clubs wrote:

Would someone like Carlsen be able to gain any kind of advantage by studying how Stockfish is programmed/coded in preparation for a match?

Very much, especially by studying evaluation. Stockfish is weak in quiet midgame positions involving a lot of planning, like those occuring after King's Indian Defence. Stockfish also is a agressive pruner. But I bet SF would murder Jose Raul Capablanca and all other players at endgames.

Joona Kiiski on where Stockfish is weak:

* Classical king's Indian kind of blocked positions where "black" slowly develops dangerous attack. Stockfish playing with "white" has no clue what is happening until it's far too late.

* Stockfish is far too optimistic in late midgame positions where it has material advantage, but its king has no real shelter and queens are still on board. The risk that opponent is able to force perpetual check is great, but Stockfish has no understanding of this.

* Stockfish greatly overvalues connected passed pawns when opponent can create a blockade with his pieces. Luckily this doesn't happen very often in practice.

* Stockfish is too optimistic in late endgame with very few pawns, like KNP vs. KN or KBPP vs. KBP which often liquidate to draw.

* Stockfish still lacks basic endgame knowledge for some trivial drawn positions like KQ vs. KP when pawn is on "c2" or "a2"

Jion_Wansu

Mar 13, 2015

0

#54

Then the chess engines that chess.com uses are stupid. They fall for the same tricks

HilarioFJunior

Mar 13, 2015

0

#55

GnrfFrtzl wrote:

HilarioFJunior írta:

GnrfFrtzl wrote:

LegendLength írta:

Out of interest is it possible to calculate the maximum rating possible for either fide or some other system of ratings?

I'm sure you could do it for a game of 1 king and 1 queen each, so it should be extendable for a normal game right? Or would you have to work it out empirically by making a perfect engine first?

I'm sure there is no maximum rating.
Ratings will go on climbing higher and higher until someone works out a better rating system.

There's a linear correspondence between the quality of moves from a player and his elo rating. Hence there's probably a maximum rating. According to some researchers it's 3571.

http://www.cse.buffalo.edu/~regan/papers/pdf/Reg12IPRs.pdf

'There's probably' is not a definite.
Well, we're just theorisin.
I'm using the farmer's common sense:
"Just when you think an end hasben' reached, there comes a kid that shows a new way."

But it's an evidence.

Jack_of_Clubs

Mar 13, 2015

0

#56

vittyvirus wrote:

Jack_of_Clubs wrote:

Would someone like Carlsen be able to gain any kind of advantage by studying how Stockfish is programmed/coded in preparation for a match?

Very much, especially by studying evaluation. Stockfish is weak in quiet midgame positions involving a lot of planning, like those occuring after King's Indian Defence. Stockfish also is a agressive pruner. But I bet SF would murder Jose Raul Capablanca and all other players at endgames.

Joona Kiiski on where Stockfish is weak:

* Classical king's Indian kind of blocked positions where "black" slowly develops dangerous attack. Stockfish playing with "white" has no clue what is happening until it's far too late.

* Stockfish is far too optimistic in late midgame positions where it has material advantage, but its king has no real shelter and queens are still on board. The risk that opponent is able to force perpetual check is great, but Stockfish has no understanding of this.

* Stockfish greatly overvalues connected passed pawns when opponent can create a blockade with his pieces. Luckily this doesn't happen very often in practice.

* Stockfish is too optimistic in late endgame with very few pawns, like KNP vs. KN or KBPP vs. KBP which often liquidate to draw.

* Stockfish still lacks basic endgame knowledge for some trivial drawn positions like KQ vs. KP when pawn is on "c2" or "a2"

Very interesting. On that basis, I'd like to think there are reasons to be optimistic that a World Champion can compete against the strongest chess programmes with enough preparation.

AngeloPardi

Mar 13, 2015

0

#57

GnrfFrtzl wrote:

TheGreatOogieBoogie írta:

GnrfFrtzl wrote:

TheGreatOogieBoogie írta:

Yes! He's the best player in the history of the world and machines don't know how to plan. He can steer the game into channels where human thinking trumps brute force calculation. While he can't calculate nearly as fast a an engine engines have a horizon effect of 20 moves whereas Carlsen can see deeper than that, resulting in a more accurate assessment.

Not trying to diss the guy, but how on Earth could Carlsen see deeper than an engine?
I understand the difference between the human thinking and pure engine, but there's no human that can see 20 moves deep.

Alekhine sometimes looked 30 moves ahead as shown sometimes in his best games collection.

There is absolutely no way that anyone can look 20-30 move deep.
10, I can believe, but even that is extremely rare.
If you're talking 20 move deep, you're talking thousands of lines and variations.
No human can see that.
If you give them a pen and paper and a few hours, fine, I'll buy it.
But not without those.

In endgames it perfectly possible do calculate down a line 20 moves ahead, but it's only due to the very limited number of moves available.

And each time you decide to exchange a bishop for a knight and to cripple your opponent pawn structure, you are looking twenty moves forward - of course, this doesn't mean that you have calculated every variation.

GnrfFrtzl

Mar 13, 2015

0

#58

AngeloPardi írta:

GnrfFrtzl wrote:

TheGreatOogieBoogie írta:

GnrfFrtzl wrote:

TheGreatOogieBoogie írta:

Yes! He's the best player in the history of the world and machines don't know how to plan. He can steer the game into channels where human thinking trumps brute force calculation. While he can't calculate nearly as fast a an engine engines have a horizon effect of 20 moves whereas Carlsen can see deeper than that, resulting in a more accurate assessment.

Not trying to diss the guy, but how on Earth could Carlsen see deeper than an engine?
I understand the difference between the human thinking and pure engine, but there's no human that can see 20 moves deep.

Alekhine sometimes looked 30 moves ahead as shown sometimes in his best games collection.

There is absolutely no way that anyone can look 20-30 move deep.
10, I can believe, but even that is extremely rare.
If you're talking 20 move deep, you're talking thousands of lines and variations.
No human can see that.
If you give them a pen and paper and a few hours, fine, I'll buy it.
But not without those.

In endgames it perfectly possible do calculate down a line 20 moves ahead, but it's only due to the very limited number of moves available.

And each time you decide to exchange a bishop for a knight and to cripple your opponent pawn structure, you are looking twenty moves forward - of course, this doesn't mean that you have calculated every variation.

So we're talking about different stuff, then.
I understand that looking into 10-20 moves is possible if the moves are limited and are forcing (endgames or mating nets).
But an engine sees all lines and variations at any position.
This is what a human will never be able to do.

Jack_of_Clubs

Mar 13, 2015

0

#59

I'm not sure that a computer does search equally deep across all lines. Doesn't it truncate lines that are considered inferior?

vittyvirus

Mar 13, 2015

0

#60

Jack_of_Clubs wrote:

I'm not sure that a computer does search equally deep across all lines. Doesn't it truncate lines that are considered inferior?

Yes. If it'd search all lines, then even the best computers won't go beyond depth 8 for the starting position. Most of the pruning is done by the alpha-beta algorithm, which prunes without affecting accuracy of search. Null-moves and LMR are other very popular reductions, but they affect accuracy of the search. For example, null-moves don't work in zunzuangs (dunno the spelling).

It's thus a important point that computer search isn't perfect.