Author: Albert Silver
Date: 11:14:17 04/20/01
Go up one level in this thread
On April 20, 2001 at 01:44:50, Mike S. wrote:
>On April 20, 2001 at 00:46:11, Dann Corbit wrote:
>
>>(...)
>>In any case, play against a chess program with takebacks. You will find that
>>you can eventually win unless you are a complete idiot. Now, just record the
>>winning moves and try again. Eventually (through a very simple method like
>>this) you can detect and exploit weaknesses or opening book errors.
>>
>>If anything, it is his strongest point.
>
>I'm afraid it will remain unclear forever (not that I'm too interested :o), if
>the first version, or draft of that contract included an obligation to hand over
>a complete package, including the opening book, or just the engine. IMO it would
>*not* make sense for BGN nor Kramnik to have the opening book, for obvious
>reasons (because they of course want to have a match, no killer book
>preparation), so I think they would have removed that from the contract as soon
>as it is explained (if necessary...) during negotiations. I think it's unusual
>to make such details of contracts public.
>
>If I want to take my chance and find parts of a contract I cannot agree with, I
>try to negotiate them away - that's what Amir Ban tried to explain to us.
>
>I do not assume, that BGN or any other organizer would let programs compete
>under different contracts. Whoever suspects this, should first bring evidence
>and talk later. So, do you think DF or DJ would agree to send their openings to
>Kramnik 3 month before the final match, without the right to change anything?
>This doesn't sound quite reasonable to me.
>
>Providing the engine alone isn't such a problem I think, because in this case
>Kramnik could not prepare for a small number of opening lines most likely to
>reach. I do not think he would profit much more from a new engine version, than
>from a current version, during his training.
>
>Furthermore: I'm only guessing and may be wrong, but I don't think that he will
>be able to use the match hardware for preparation. Therefore, he will only get a
>rough impression more or less of what his opponent will be; he'll know much less
>than he is used to know about the GM's he usually plays against. So lets not
>overestimate this preparation issue.
>
>Regards,
>M.Scheidl
I think you are grossly UNDERestimating the preparation issue. I think the
program under such circumstances is a dead duck. Give me a Sicilian defense with
lots going on and I could still prepare against the program, but if I were truly
out to kill the machine, I would choose consistently close structures stemming
from 1.f4 or 1.d4 f5 openings. No doubt he would cover his rear end to be sure
he could consistently do so, and don't believe for an instant he won't. I saw no
backing down against Kasparov as he stonewalled (figuratively speaking) the then
WC with the Berlin Defense, etc. Despite Kasparov's inability to break it, I
think the programs would suffer MUCH more. Furthermore, he is a true specialist
in such openings. He wrote the chapter on the Dutch Defense (1.d4 f5) in
Dvoretsky's Positional Play. He knows ALL the subtleties of such positions. And
worse, even in an inferior closed position he should be able to win or draw at
worst. These positions pretty much KILL those extra million NPS the hardware
would obtain.
As a match strategy from the programming team's side, here are a couple of
things I would do that would abide by the regulations as stated.
- First, the whole changes and upgrade issue. As I understand it, it IS now
permitted to continue working on the program but all changes must be forwarded
to the Kramnik team. Fine. I imagine that there is always more than one veriosn
being tested, honestly so, so that different items can be added, checked, and
removed (or kept). I'd send him a version. Any. Truly and honestly, it is the
current house veriosn being worked on. Truly and sneakily, it isn't the ONLY
house version being worked on. Truly and honestly, it is difficult to know
whether a new item being tested is better or worse so withholding it is not
contrary to the spirit (the latest and strongest version) of the program. I
would send him my conlcusively strongest version VERY close to the beginning of
the match. Be VERY careful about this though. DON'T underestimate Kramnik's
ability to prepare. Here's a small story for you: The Spanish GM Miguel Illescas
was once teamed with Kramnik for 2 days (only 2) for a special training session.
He was being well-paid of course. His job was to teach him what he knew about
the Tarrasch and Semi-Tarrasch openings, something he had been playing a long
time. He thought the time was completely insufficient, despite Kramnik's
reputation, and doubted what could be done. He said that he simply couldn't
believe it. In 2 days he himself had NO doubt that Kramnik was every bit of an
expert them it as he was. Kramnik's knowledge of the game was so collosal, and
memory, that 2 whole days turned out not to even be necessary. Above all, I
would try to have the version that played, to be the one that was most different
from the next-to-last version sent him.
- I would very carefully work on tweaks that would steer away from static
structures, because while these postional niceties in a program such as Junior
or Shredder work great against other programs or lesser players, they are a
deathwish against someone such as Kramnik. No question in my mind about this. I
would probably enlist (paid) someone who is an expert in this particular aspect:
Robert Hyatt. Whatever one might think of Crafty's relative strength, it is well
known that his program has many special implementations designed to cope with
just this problem. His extensive experience would be invaluable I would imagine.
You could do what you liked with it, but the feedback would be bound to be
helpful.
These are just a couple of things, but I'd definitely start thinking along these
lines.
Albert
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.