Back with another revisiting of recent posts and your comments.
Regarding this much-commented entry on computer opening books, I got an e-mail from a man who knows, Chrilly Donninger. He came to fame as the programmer of Nimzo and is now running the hardward-based Hydra chess project (formerly Brutus).
His machine just beat the elite program Shredder convincingly in a match. In that great report Donninger makes the points he made in his e-mail: Hydra uses a drastically truncated book, going out on its own after move 10 in most cases. He explains "It is just to set the program on the right track. No special book tricks, play just the main line." That's certainly a good start. A rumor on the computer chess streets said that the well-funded Hydra team in the UAE prepared for the match by meticulously going through the Shredder book looking for weaknesses, a la Kasparov. But looking for weak spots up to move ten wouldn't be useful enough to bother with, although you could certainly find lines your opponent isn't comfortable with. More relevant is that Shredder is a commercially available program, making it relatively easy to prepare for if you want.
Regardless of rumors, Hydra outplayed Shredder even when it wasn't winning out of the opening as in the first few games. It is safe to postulate that Hydra, running on as many cards as it did in Abu Dhabi, is the strongest chess machine extant. It would be interesting to let it play a few hundred games to see if it can even lose to a PC program, and I assume the Hydra team has done that and more.
I'm up to my eyeballs in chess programmers these days. Friend and former KasparovChess Online colleague Shay Bushinsky is in town on vacation with his family. Shay, together with Amir Ban, is half of the Israeli programming team behind the current computer world champion Junior. He's enthusiastic about the next version of Junior, number nine, number nine, number nine...
Today Mark Crowther produced issue TWIC issue 512, which is two to the power nine. Number nine, number nine ....
Seems to me that the fact that Hydra outplayed Schredder with a truncated opening book simply proves that finally computers can re-invent their own opening theory. The good point being that the resulting position will probably suit better the program, even if a mere mortal would disagree on the choice.
I hope this approach will be adopted by other programmers as well for future computer-computer matches.
I completely agree with Philippe.
I add that the real interest for a human player in computer vs computer matches is not of course to be a fan of this or that program.
Interest I think is in discover new improvements in positions, particularly in the openings: for this reason I do not agree with the "book burning" idea but I'd like a computer book where the moves are stored according to the program itself.
If Hydra during a game is out of book at move 11 and choose a move after for example 4 minutes this move must be stored in its book so that next time in the same position he'll play it suddenly...
(Instead of wasting time repeating the analysis to discover what already discovered previous time:
during a match without opening book all the audience, even patzers, can say "now the program will play 9.h3". The program, the best chess player in the room, deprived of his knowledge as HAL in 2001 SPACE ODYSSEY, the Kubrick's flm, needs time to think...after some minutes he discovers and plays what everibody knew, 9.h3, boring no?).
Better: after the game let Hydra analyze position at move 11 again and not for 4 minutes but for 4 hours. If the move is still the same let it stored in the book else rewrite it becouse the analysis after 4 hours has more validity than a 4 minutes one.
A good idea should be a book where moves are stored with a flag about the number of chess nodes that the program needed to find it: in this case the tree of a computer opening book could show his contents at different levels of evalutation: when a program is selected to play openings at 100% of his strenght it always choose the top level move, else can choose one of the lower variations to avoid the effect "oh...my stupid computer...he always plays the same moves" .