Author: Robert Hyatt
Date: 06:19:13 05/10/00
Go up one level in this thread
On May 10, 2000 at 02:30:56, Peter Kappler wrote: >On May 09, 2000 at 22:02:06, Robert Hyatt wrote: > >>On May 09, 2000 at 11:58:35, Bruce Moreland wrote: >> >>>On May 08, 2000 at 13:31:02, Robert Hyatt wrote: >>> >>>>On May 08, 2000 at 11:10:51, blass uri wrote: >>>> >>>>>On May 08, 2000 at 10:37:54, Robert Hyatt wrote: >>>>><snipped> >>>>>>Absolutely not. First, it would give Kasparov a chance to see how deep the >>>>>>thing searches, how it extends. How it evaluates some positional >>>>>>considerations. It would be a decided advantage for Kasparov had he had this >>>>>>kind of information. Injecting it into the middle of the match would have >>>>>>definitely tainted the results. >>>>> >>>>>If the machine is strong enough then knowing how deep the thing search and how >>>>>it extends and how it evaluates positional considerations could not save >>>>>kasparov from losing. >>>>> >>>>>Uri >>>> >>>> >>>>No.. but if they are close in strength, it could be a deciding point for the >>>>match. >>> >>>Looking at those printouts wouldn't have helped him a damned bit. >>> >>>This thread is a giant troll, isn't it? >>> >>>bruce >> >> >>I totally disagree. I had an IM get a copy of Crafty, played it a bunch of >>games offline, and 'learned' some things about how it evaluated (and >>mis-evaluated) things. And then he returned to ICC and did much better against >>Crafty (and its clones) than he should have. He later told me what he had >>found, and how he had found it, and commented that "had the program not been >>available so that I could play it, watch its analysis, see how it evaluated >>various things, I could not have learned how to exploit certain weaknesses." >> >>I believe Kasparov could do that _far_ quicker than the IM did. He pointed >>this out after match one. I would love to know how a program evaluates passed >>pawns vs king safety, space vs center control, pawn structure vs piece mobility, >>and so forth. I think it could be a decisive advantage if the player is already >>strong enough to be a 'problem' for the computer. Kasparov would be a 'problem' >>for _anybody_. > > >The question was how much *looking at the printouts* would have helped Kasparov. > >Letting Kasparov play a bunch of offline test games against DB would *obviously* >have been a huge advantage for Garry, but that's completely different than >looking at some PVs and scores from one game. > >--Peter Again, I disagree. A strong chess player can learn a _lot_ about a program from looking at a complete log of one or two games. Particularly when that player knew what _he_ was thinking about during the game. Now he has some insight into how DB 'thinks'. I had a non-titled chess player watch the "trojan horse activity" on ICC, and he watched me stop it completely. He wasn't much of a programmer, but could read C reasonably well. He studied the source for crafty, looked at the range of scores my code could produce to avoid this attack, and he worked out a very nice plan to bust it wide open. All by studying about 20 lines of code and thinking about what they did, and how this could be exploited. He was quite proud when he wrecked several crafty clones on ICC (He carefully avoided mine so I wouldn't notice and fix it.) Later he told me what he had done after it became obvious that someone had told me and I had fixed it. :) You would be surprised what good chessplayers can learn by reading the code, looking at it play, or studying detailed output. The information content is _not_ "close to zero" to them...
This page took 0.03 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.