You can easily find out yourself. Run an engine-engine tournament between two Stockfish engines, one with all that stuff and one without all that stuff. I think you'll find that there's not that much of a difference in results.
Are Humans still better at Chess?
Sort:
and NO.
for all the fuss, opening books aren't Really SO helpful.
seeing tactics & Promoting pawns is how the game is won and lost. (generally)
May 21, 2016
0
#4
EscherehcsE wrote:
You can easily find out yourself. Run an engine-engine tournament between two Stockfish engines, one with all that stuff and one without all that stuff. I think you'll find that there's not that much of a difference in results.
Has anyone tried this? I think it would be interesting to see the results.
May 21, 2016
0
#5
xt8088 wrote:
EscherehcsE wrote:
You can easily find out yourself. Run an engine-engine tournament between two Stockfish engines, one with all that stuff and one without all that stuff. I think you'll find that there's not that much of a difference in results.
Has anyone tried this? I think it would be interesting to see the results.
I did it a while back. I can't remember where I put the results, but I have vague recollections of about only a 20 or 30 elo difference. I guess I could do it again, but I'd have to decide on the time control and number of games to run.
[Edit] - Oh, and I'd have to find a good competitive opening book. I normally use neutral books, but that wouldn't work for this test.
May 21, 2016
0
#6
They've had human vs engine matches where the engine has no book. They've also had it where the human has access to an opening book. They've also done human vs engine and the human gets a weaker engine to use for suggestions and so no simple tactics are missed.
It didn't go well for the human in any of these.
An interesting hypothetical is GM that calculates as fast as engine vs engine that calculates as slow as a human... obviously the engine is crushed.
The engines could be used to generate a pretty solid opening book from scratch at this point, I would think. Put a well-studied opening position into Stockfish, and if you let it run for a while it will generally come up with the main line or a slightly less popular alternative that scores well.
May 21, 2016
0
#8
From the initial position you'd almost never get a main line out of an engine... even if after you show the lines to the engine it will agree they're best.
They're not designed to play the opening.
Chess engines are designed to evaluate chess positions. Opening positions are chess positions. There might be some positions that chess engines (so far) have difficulty evaluating accurately, but there's nothing special about the opening that prevents chess engines from finding the best moves. Except that you don't usually get highly tactical positions early in the opening so it doesn't play to chess engines' strength either. But if you give Stockfish with no opening book a long time (like an hour) on a reasonably fast computer to play each move, I suspect it's going to end up playing mainline theory most of the time.
A while back I let Stockfish 6 run on the starting position for a few hours, these were the top 5 lines it found.
1. d4 Nf6 2. Nf3 e6 3. c4 b6 4. Nc3 Bb4 5. e3 Bb7 6. Bd3 O-O 7. O-O d5 8. a3 Bxc3 9. bxc3 Ba6 10. cxd5 Bxd3 11. Qxd3 Qxd5 12. Bb2 Rd8 13. c4 Qe4 14. Qe2 Nbd7 15. h3 c5 16. Rfd1 h6 17. a4 Qg6 18. d5 Ne4 19. a5 exd5 20. cxd5
+ (0.18) Depth=36/53 0:196:04 59896 MN
1. Nf3 Nf6 2. d4 e6 3. c4 b6 4. Nc3 Bb4 5. e3 Bb7 6. Bd3 O-O 7. O-O d5 8. a3 Bxc3 9. bxc3 Ba6 10. cxd5 Bxd3 11. Qxd3 Qxd5 12. Bb2 Nbd7 13. c4 Qe4 14. Qb3 Qg6 15. Qa4 Rfd8 16. Rad1 h6 17. h3 c5 18. dxc5 Nxc5 19. Rxd8+ Rxd8 20. Qxa7
+ (0.17) Depth=36/53 0:196:04 59896 MN
1. e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6. Bb5+ Bd7 7. Bxd7+ Nbxd7 8. O-O Be7 9. dxc5 Nxc5 10. Nb3 Ne6 11. Be3 O-O 12. c3 Re8 13. Re1 a6 14. Nbd4 Ne4 15. Nd2 Nf6 16. Nf5 Qd7 17. Bd4 Nxd4 18. Nxd4 Bd6 19. h3 Ne4 20. Nxe4 Rxe4
+ (0.12) Depth=36/53 0:196:04 59896 MN
1. g3 e6 2. Nf3 Nf6 3. Bg2 d5 4. O-O Be7 5. d4 O-O 6. c4 Nbd7 7. b3 b6 8. cxd5 exd5 9. Nc3 Bb7 10. Bb2 Re8 11. e3 c6 12. Rc1 a5 13. Qc2 Bd6 14. Rfe1 Bb4 15. a3 Bd6 16. e4 dxe4 17. Nxe4 Nxe4 18. Rxe4 Nf6
+ (0.07) Depth=36/53 0:196:04 59896 MN
1. c4 e5 2. g3 Nf6 3. Nc3 Bb4 4. Bg2 O-O 5. Nf3 d6 6. O-O h6 7. d3 Be6 8. Bd2 Nc6 9. a3 Bxc3 10. Bxc3 Qd7 11. b4 Bh3 12. b5 Ne7 13. a4 Rfe8 14. e4 a6 15. Re1 axb5 16. axb5 Rxa1 17. Qxa1 Ng6 18. Bxh3 Qxh3
+ (0.05) Depth=36/53 0:196:04 59896 MN
Top two results transpose to the same line in the Queen's Indian Defense, third is the French Tarrasch, fourth is a Reti or something (not sure how to classify it) but plenty of theory on the position after move 6 (1... e6 admittedly unusual then it transposed back to something more common), and the fifth is a line in the English Opening but maybe dubious for white.
As you go deeper in these lines the move quality probably degrades, but the first several moves still tend to be theory.
May 21, 2016
0
#11
JSLigon wrote:
Chess engines are designed to evaluate chess positions. Opening positions are chess positions. There might be some positions that chess engines (so far) have difficulty evaluating accurately, but there's nothing special about the opening that prevents chess engines from finding the best moves
In other words you don't know how engines play chess, you just know that they do play chess... somehow. I got it 
But ok, I'll read the rest of your post now.
May 21, 2016
0
#12
Yeah, it will find some book moves of course. But "book" or "theory" is huge so that's no real accomplishment. Theory also includes many bad openings after all. Usually the openings engines choose are bland. The ones you posted aren't main lines, or even main defenses.
Engines are mainly tuned for the middlegame. Even in middlegames they'll superficially value things like space and piece activity when it doesn't matter. A big problem is the horizon effect.
May 24, 2016
0
#13
xt8088 wrote:
EscherehcsE wrote:
You can easily find out yourself. Run an engine-engine tournament between two Stockfish engines, one with all that stuff and one without all that stuff. I think you'll find that there's not that much of a difference in results.
Has anyone tried this? I think it would be interesting to see the results.
I just finished a 100-game test using Stockfish 7 x64 (one thread only, 40 moves in 4 minutes, repeating time control). I installed the Tomcat opening book and 5-piece Syzygy EGTBs for one engine. I don't know anything about the Tomcat book; I just chose it because it can be downloaded from Chess.com. Results were 53.5 % for the booked engine, which is about +24 elo. (100 games isn't a real accurate sample, but it should be good enough to get a rough idea.)
I may run similar tests with a few other books just to see if I get radically different results with other books.
Jun 4, 2016
0
#14
I just finished testing a few other opening books, and the results are:
Arena books:
Tomcat 53.5 %, +24 elo
DC (Dann Corbit) FTP Superbook v2 45 %, -35 elo
HIARCS Reference Book Lite 2550 47 %, -21 elo
Fauzi 2.2 58.5 %, +60 elo
Polyglot books
Diffusion 1.0.9 60.5 %, +74 elo
Morphius 12.9c 60.5 %, +74 elo
Top Tier 59.5 %, +67 elo
Stockfish 2.3.1 57.5 %, +53 elo
(For the Polyglot book tests, I decided to use the Stockfish DD x64 PA GTB engine, since it can use both polyglot books and the Gaviota tablebase files.)
Conclusions - It does appear that the choice of opening book can make a difference in performance compared to not using a book. I'm not sure how much of a benefit, if any, the endgame tablebase files were, since I was using a mechanical hard drive and only 5-piece tablebases.
Here's the question, isn't really a computer basically playing correspondence chess and the poor human is playing live chess?
I mean the computer has all the opening books, databases of games and maybe specifically all the games its actual opponent has ever played!
Without all that would the best of us puny humans still win?