Joel on Computer Training
By GM Joel Benjamin   
November 5, 2007
Joel,
Was Deep Blue’s computational strength about 15 million nodes per second?

I think Intel PC’s are approaching Deep Blue. I have seen a four-core 2ghz platform achieve 5.5 million n/s with Deep Fritz 10. An Intel S5000xvn motherboard with dual X5365 (eight cores at 3ghx each) may be half way to Deep Blue.
Has Blue Gene ever been measured for its nodes/second chess capacity?

Joshua Friedel lost to Shabalov in the recent New England Masters.



On move 62 …Ke3??, Joshua missed his draw. In reviewing the entire game, it appeared to me that both players were making many moves that Fritz 10 thought were best. Would you say that computer training is getting to be a primary source of many players' lines today?

I am guiding my PC through 21- 24 ply per move for a month, no book. Considering an average game is 80 ply, 21 ply is 25%, so that’s pretty good. I collect the sublines of 20 sublines per move. Fritz 10 wants to play a Ruy Lopez Arc Angel variation. On the first move white’s 1. e4 was credited a .25pt advantage, but by move 10 Black has achieved -.3 advantage. I am going to play it out, but it may be that White has been too aggressive.

Thank you,
S. Warren Lohr



Dear Warren,

I’m not ashamed to say I can’t answer parts of your question.  Please keep in mind that my degree is in history.  But I will address the parts I think I understand.

1)    PCs have obviously greatly increased in power since 1997, but chess engines do their searches differently as well, which contributes to their greater search depth.

2)    Grandmasters don’t copy their opening moves from computer programs.  But they do test their opening ideas on the computer to assess their value and look for possible replies.

3)    Almost any grandmaster game will have a high correlation to almost any search engine.  Most good moves are fairly obvious to good players.  It was quite absurd when Topalov pointed to a high correlation of Kramnik’s moves to search engines as “proof” that Kramnik was cheating.  [Though agreement of specific moves can sometimes be relevant, especially for weaker players]

4)    Computer evaluations are at their most meaningless in the opening.  When there are no obvious tactics or material imbalance, judgment reigns.  Computers often have biases that make them stronger in the middlegame but produce unsightly moves in the opening on occasion.  In the game you  were testing on Fritz10, the differences in scores are so tiny that they have no relevance at all.

Ask Joel a question of your own at askgmjoel@uschess.org . Also be sure to read the submission guidelines.