Computer results vs human results

[ganstaman]

I was about to post this in the other thread, but figured this forum could better handle a new thread than a derail:

Originally Posted by Do it Right

It's also worth mentioning that high level computer chess has very few draws in general. The drawing phenomenon seems to be largely a human one and not a direct consequence of the game.

To go further off topic, there's something strange I noticed about a few computer tournaments that I was hoping someone could help clarify for me. What I noticed was that black seemed to win more often than white. Here are some example computer tournaments I could find (I am showing you all the ones I looked at just now, so no bias in my selection):

WCCC 2006: +32 -40 =21
15th World Computer Chess Championship Tournament: +21 -23 =19
15th World Computer Chess Championship Tournament: +17 -16 =12
10th International Computer Tournament: +24 -25 =19
2011 World Computer chess Championship Tournament: +13 -9 =14

I then found 3 human tournaments for quick comparison:
London Classic 2011: +9 -8 =19
Tata Steel 2012: +22 -17 =52
US Championship 2012: +24 -12 =30

So what's going on here? It is amazing to compare the number of draws, but the victories with black confuse me a bit. Does anyone have better stats in a database somewhere to see if this is a real effect or something?

[ganstaman]

Hmmm, I just had an idea of how to check this out on chessgames.com. I searched for any player with "computer" in the name and got this: http://www.chessgames.com/perl/chess...=&eco=&result=

In these games, white went: +1628 -1271 =1011

Still, I linked to 3 tournaments above with more black victories than white victories. Has that ever happened in a human tournament?

[EGarrett]

I was unable to find a page for Houdini or Rybka on chessgames, which irritated me to no end. However, it's interesting to see how few draws there are in the computer games. Obviously they have don't any pride or fear of defeat, so essentially they just try to rip each other up every game.

[Do it Right]

To my knowledge computers don't draw games where draws are not forced. If humans weren't allowed to agree to draws in positions where the draw is not clearly forced then the number of decisive games would likely skyrocket.

[John_Douglas]

The % of draws is also a factor of rating difference. What do the numbers look like if you exclude games with rating difference of more than 50 points? My guess would be that the number of draws in computer chess would be significantly higher than what is shown here.

[ganstaman]

Originally Posted by EGarrett

I was unable to find a page for Houdini or Rybka on chessgames, which irritated me to no end.

You can search for either of them from the home page just as you would for any other player:

http://www.chessgames.com/perl/chessplayer?pid=100439
Oh, I take that back, Houdini doesn't have a page there, but Rybka definitely does.

Also, the responses so far talk about draws, but I was really interested in the black wins. I haven't yet been able to find a human tournament with more wins from the black side than white side, yet I have no trouble doing so with computer tournaments. I can't figure out why this would be.

[John_Douglas]

Originally Posted by ganstaman

You can search for either of them from the home page just as you would for any other player:

http://www.chessgames.com/perl/chessplayer?pid=100439
Oh, I take that back, Houdini doesn't have a page there, but Rybka definitely does.

Also, the responses so far talk about draws, but I was really interested in the black wins. I haven't yet been able to find a human tournament with more wins from the black side than white side, yet I have no trouble doing so with computer tournaments. I can't figure out why this would be.

Well, neither side has any objective advantage in the initial position, so the only way a win is possible is due to some mistake. Computers probably are equally likely to make mistakes with either colour, but maybe human players are more likely to make mistakes with Black due to a psychological aversion to defending.

Just one possible explanation.

[EGarrett]

Originally Posted by ganstaman

You can search for either of them from the home page just as you would for any other player:

http://www.chessgames.com/perl/chessplayer?pid=100439
Oh, I take that back, Houdini doesn't have a page there, but Rybka definitely does.

Also, the responses so far talk about draws, but I was really interested in the black wins. I haven't yet been able to find a human tournament with more wins from the black side than white side, yet I have no trouble doing so with computer tournaments. I can't figure out why this would be.

Yes, while reviewing the Rybka/Houdini match, I noticed how often Houdini won with black (and did so by winning a bishop).

The computers never really play to equalize or defend, nor are they ever afraid, so they always look to attack (as Fischer did in his prime), so winning with black would be fairly common I'd imagine. White may keep the edge in terms of the most wins, but when white doesn't win, it's probably just as likely to lose outright to the computer with black as it is to draw.

[valenzuela]

Maybe the opening books for black are better.

[ganstaman]

Originally Posted by John_Douglas

but maybe human players are more likely to make mistakes with Black due to a psychological aversion to defending.

Originally Posted by EGarrett

The computers never really play to equalize or defend, nor are they ever afraid, so they always look to attack (as Fischer did in his prime), so winning with black would be fairly common I'd imagine.

Originally Posted by valenzuela

Maybe the opening books for black are better.

These guesses seem as good as any. I wonder if one could use them to improve their own score from the black side.

[meekrab]

Originally Posted by John_Douglas

Well, neither side has any objective advantage in the initial position

It's White's move, that's a pretty big objective advantage.

I think these tournaments are just statistical outliers combined with some pretty terrible engines losing a lot of games against Rybka as White.

[ganstaman]

Originally Posted by meekrab

I think these tournaments are just statistical outliers combined with some pretty terrible engines losing a lot of games against Rybka as White.

The same engines lose to Rybka as black, so that's not it.

I don't think they can just be brushed off as outliers, either. I've noticed this before and only posted a handful of tournaments -- I believe there are more computer tournaments like this. And like I said, I can not find a single human tournament with more black wins than white wins despite how many human tournaments there have been (and no one has yet found one for me either). Why is it so easy to find these computer tournaments but so impossible to find a human one?

Here's a bunch more for fun from Linares:
2004: +6 -3
2005: +9 -6
2006: +15 -11
2007: +12 -6
2008: +16 -9
2009: +9 -6
2010: +8 -1

If someone could find a human tournament with more black wins, then maybe I'd be satisfied that this is just a fluke. Still, I basically picked 5 computer tournaments at random and 60% of them had more black wins. I've now posted 10 human tournaments, somewhat at random (I clearly just went through however many Linares tournaments I could find easily, so they weren't random in that sense), and 100% of them have more white wins. This doesn't seem that odd to you?

I've never seen anyone talk about this before, so it does make me wonder if I'm making something out of nothing, but it still feels fishy to me.

[PyramidScheme]

clearly its better to be black. lol

[Noir_Desir]

last candidates matches, +1 -2

i think humans are just used to be satisfied with a draw with black and this affects their decision-making. Disregarding how big white's initial advantage really is, hardly any good player would deny that it is far easier to play white than black. The advantage of the first move means that it is easier to get structures you like, get a space advantage, complicate play or force simplifications both of which is often difficult with black without making major concessions.

[mangler241]

Here's a tournament ( only 30 games ) for which Black won 6 and White only 5:

http://www.chessbase.com/newsdetail.asp?newsid=6444

so there must be at least dozens/hundreds of tournaments like the above where there are more Black wins.


For Houdini, Rybka et al, compared to humans, Black is more likely to at least equalize; when humans are under pressure ( usually because of White's initiative if Black committed some inaccuracies ), they don't always defend well.