Author: Robert Hyatt
Date: 07:49:46 05/29/03
Go up one level in this thread
On May 29, 2003 at 02:01:16, Ricardo Gibert wrote: >On May 28, 2003 at 23:01:16, Robert Hyatt wrote: > >>On May 28, 2003 at 19:36:27, Ricardo Gibert wrote: >> >>>On May 28, 2003 at 14:55:10, Robert Hyatt wrote: >>> >>>>On May 28, 2003 at 11:17:00, Ricardo Gibert wrote: >>>> >>>>>On May 28, 2003 at 11:00:06, Robert Hyatt wrote: >>>>> >>>>>>On May 28, 2003 at 00:57:00, Ricardo Gibert wrote: >>>>>> >>>>>>>On May 28, 2003 at 00:10:32, Robert Hyatt wrote: >>>>>>> >>>>>>>>On May 27, 2003 at 19:11:49, Jorge Pichard wrote: >>>>>>>> >>>>>>>>>After being completely out-played for the entire game, and with imminent defeat >>>>>>>>>on the horizon, Kasparov resigned the 2nd game rather than drag out the >>>>>>>>>humiliation. But Deep Blue had made a critical error, allowing Kasparov a >>>>>>>>>perpetual check. The analysis is quite deep and extends slightly beyond Deep >>>>>>>>>Blue's search horizon. And, apparently, also Kasparov's. Kasparov's team, which >>>>>>>>>included Grandmaster Yuri Dokhoian and Frederic Friedel, were faced with the >>>>>>>>>delicate task of revealing the news to Kasparov. They waited until lunch the >>>>>>>>>next day, after he had had a nice glass of wine to drink. After they revealed >>>>>>>>>the hidden drawing resource, Kasparov sunk into deep thought (no pun intended) >>>>>>>>>for five minutes before he conceded that he had missed a draw. He later claimed >>>>>>>>>that this was the first time he had resigned a drawn position. >>>>>>>>> >>>>>>>>>Six years later, which program can see the draw in the famous 2nd game of the >>>>>>>>>rematch? >>>>>>>>> >>>>>>>>>Jorge >>>>>>>> >>>>>>>> >>>>>>>>NO program sees this. It is about 60 plies deep. It is unlikely that a >>>>>>>>program will see it for quite some time to come, in fact... >>>>>>> >>>>>>> >>>>>>>Whenever this comes up, you always say "about 60 plies", but I can't find >>>>>>>anything to corroborate this. Let's see your analysis. >>>>>> >>>>>>The best analysis I specifically remember was posted on Ed's web site a few >>>>>>years ago. I didn't save it as it was not particularly "interesting" to me. >>>>>> >>>>>>It might still be available however... >>>>> >>>>> >>>>>http://www.rebel.nl/db2.htm >>>>> >>>>>He only gives 36 plies. A far cry from "about 60 plies" >>>> >>>> >>>>OK... then that may not be the deepest drawing line. >>>> >>>>However, he does point out the difficulty of finding one particular non-checking >>>>move way deep into the tree... >>> >>> >>>While one could argue whether or not a program like Yace or Fritz can actually >>>"see" the draw from the root position, there is little question that they would >>>actually "play" the draw. This is an easy experiment to perform and you will >>>find this includes the move you allude to. >> >>I've done that myself and I agree. However, following the moves to a draw >>is not the same thing as recognizing that the game is a draw at the point in >>question. >> >>Of course, just eventually stumbling into the draw by playing reasonable >>moves is a good first approximation. But I'd prefer to see a program understand >>from the beginning that this is drawn and why this is so... ie evaluate the >>various positions along the way correctly. >> >> >>> >>>There is a more interesting experiment that I would like to see Yace perform. It >>>would require modification to the program, however. That is to have it perform a >>>test to see if the position after the ...Qc1 move occurs plus to verify that it >>>correctly evaluates it. >>> >>>BTW, I think this type of test should be a standard feature of programs that can >>>be turned on/off. How else to know to determine with a high degree of confidence >>>whether a program "understands" a given position or not? >> >> >>It's a huge headache. I could do this in the 1970s because the tree size >>was small enough to print the _entire_ thing. Not today when it could >>easily pass one billion lines of output. And if you are talking about >>searching overnight, forget it. :) > >I agree that your take on what I said is completely impractical. > >I'm just talking about a test for *single* position plus a *single* associated >eval. > >In short, a boolean test to make sure that what a certain desirable thing does >indeed happen. All it would do is set a flag that you can check after the search >is stopped. That's it. Nothing elaborate. It could easily be done for a single position. However, the question is, "is that good enough to be useful?" The answer is not so clear. There is no telling how many times that particular position is reached, and what would be interesting is the backed-up evaluation each time it is hit during the search. However, the probability is that this would _also_ produce a lot of output for deep searches, due to transpositions.
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.