Author: Robert Hyatt
Date: 07:05:17 02/18/02
Go up one level in this thread
On February 18, 2002 at 01:08:49, Uri Blass wrote: >On February 17, 2002 at 23:07:02, Robert Hyatt wrote: > >>On February 17, 2002 at 19:11:05, K. Burcham wrote: >> >>> >>> >>>We have read lots of comments on Deep Blue, from lots of different people. >>>I think I know some of how you feel about Deep Blue and Deep Blue vs GM. >>>I Have tremendous respect for Deep Blue, and the people responsible for its >>>creation and performance. I would like to hear your comments, and answers to the >>>following questions, if you will. >>> >>>1.If Crafty would show all Deep Blue moves, in analysis mode, for all six games >>>vs Kasparov---Would you agree Crafty was at least equal with Deep Blue? >> >> >>No. Because I know what Crafty is capable of... and I know what DB was >>capable of. It is certainly _possible_ that some program might be able to >>reproduce _all_ of those moves. But that wouldn't mean that program was >>equal to deep blue, without something convincing in the way of speed and/or >>depth to show some sort of "equality" with what was an incredibly fast >>machine... >> >> >> >>>2.I dont think I have heard your opinion on this one. Do you say that Deep Blue >>>is GM level, Super GM level, IM level, master level? >> >>That is hard to answer. It _must_ be GM level because they proved that >>deep thought was GM level beyond _any_ doubt. Whether it is beyond that >>(into the super-GM level) is another question that 12 games doesn't quite >>answer... So beyond the "GM" question I simply have to say "I don't know..." >> >> >> >> >> >> >>>3.Do you say some of todays programs are getting the kns of Deep Blue on the >>>dual and quad machines? >> >>Absolutely not. DB averaged 200M nodes per second, peaking at one billion >>nodes per second. What program reaches 2M nodes per second, much less 200M? >> >> >> >> >>>4.Do you say that the top twelve programs today, cannot play on the level of >>>Deep Blue? >> >>I don't believe they can, no. Deep Thought (at around 2M nodes per second) >>produced a 2650 rating over 25 consecutive games vs GM players. These were >>all 40 moves in 2 hours type games. Deep Blue was at least 100X faster, >>not to mention a lot "better" in terms of evaluation re-design, etc.. >> >> >> >> >> >>>5.Do you say that the Opening Book and EGTB are not an issue when comparing Deep >>>Blue to todays programs? >> >>DB used endgame tables. It used them in 1988 and before. Even HiTech >>used them in 1985 so this wasn't unusual. DB's opening book was obviously >>very good, being created by a GM specifically for DB. >> >> >> >>>6.Is the reason you say no program today can play on Deep Blue level is because >>>of the Deep Blue chess software (program, search, etc).? >> >>"software" is a misnomer. DB was not "software" It was a symbiotic relation- >>ship between special-purpose hardware combined with software. The "whole" was >>very impressive.. >> >> >> >>>7.If you say that 6 is true, then can't we say that todays programmers do not >>>understand programming on the level that the Deep Blue programmers did? >> >> >>Nope... it just means that when you design special-purpose hardware, you can >>do things that are not possible when using a general-purpose microprocessor >>like the rest of us. That doesn't mean the DB guys were _better_. It just >>means they worked "outside the box" the rest of us are locked inside. >> >> >> >>>8.If you agree with 7, then is it true to say there is a big difference in >>>writing a chess program for very large hardware vs limited hardware? In other >>>words from a programmers point of view, is hardware a limiting factor, knowing >>>you must prune so as to reach a competitive depth so as to be equal to programs >>>of the same time period? For example: with a dual 1500mhz and 1 gig ram, do >>>programmers size their tree for certain results (depth)-----with a 10,000 mhz >>>and 8 gigs ram, the same programmer knowing he has huge hardware would allow his >>>tree to be also huge, very little pruning. The program could be doing a lot more >>>with hardware? The search could be very large with larger hardware. Is this >>>limiting todays programmers? >> >> >>Of course that is true. But not in terms of raw mhz. the computer can't do >>some things very well... but special-purpose hardware can. _that_ is the >>advantage they had. >> >> >> >>>9.If you say it is the software, then when do you expect programmers to "catch >>>up" with the Deep Blue programmers? >> >>This is going _way_ out into no-man's-land. When chess engines can search >>at 200M nps and beyond, they have a _chance_ to equal DB. Until then, we >>will all remain well behind them. > >I am not going to repeat my opinion about deep blue here > >only one comment: > >It seems that you assume in this answer that software improvement to have >better search rules than deeper blue are impossible. > >I disagree. > >I believe that a program can search less nodes and be better thanks to better >search rules. OK... all you have to do is to find _two_ programs A and B, such that A can beat B on 200X faster hardware, with the limitation that the main difference between A and B is search. And B must have the traditional search extensions used for the past 30+ years at least (recapture, out-of-check, passed pawn advances, etc). I too believe that a program _can_ search fewer nodes and be better. But I don't believe _any_ program today can close that 200X gap... > >x nodes of the programs of today are clearly better than x nodes of programs of >1997 and x nodes of programs of 2010 are going to be better than x nodes of >today. > >I believe that part of the improvement is better evaluation function but part of >the improvement is also better search rules. > >Uri
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.