Author: Robert Hyatt
Date: 12:50:56 12/04/00
Go up one level in this thread
On December 04, 2000 at 14:22:00, Vincent Diepeveen wrote: >On December 03, 2000 at 08:28:55, Osorio Meirelles wrote: > >> >> How well would these programs play if they has the same speed of Deep >>Blue ? Would Deep Blue probably be crushed ? Any guesses on a 20 game >>match ? >> >>Thanks, >> >>Osorio > >Deep blue searched between 11 and 13 ply, which fullwidth without >hash last 6 ply is an incredible achievement, as you can see in its >logfiles. As I have pointed out before, that is _wrong_. Just look at the DB logs from the kasparov match. depth=11(5) means a _total_ of 16 full-width plies were searched. Not 11. 11 in software, 5 in the chess processors. This is all common knowledge, confirmed by every member of the DB team... > >So we can talk long and get big arguments again, but basically >you need to press your processor in 0.60 hardware to compare it, >and i definitely have a too large evaluation to even think of >putting it in hardware, might not fit on the 0.60 chip :) > >Not to mention that getting EGTB in search is gonna be tough also. > >Hashtables, my program is BASED upon having hash, so pretty hard >to miss this too. > >As deep blue was without hashtables and nullmove we can't compare >its NPS either with todays programs anyway, not to mention that >very little programs can run on 32 processors not to mention that >no one gets system time on 32 processor systems (with each >processor 16 hardware chess processors attached). > >Best we can do is look to its move and see whether our progs make >so many mistakes. My personal opinion is that Kasparov must be send >to Mars for his FM a-look-like play, i mean if i would lose like >that against chessprograms with a few draws from english systems >missing chance after chance to win, then this is explained by: >"diepeveen just has 2254 FIDE, so i don't blink with my >eyes if he makes 5 to 6 mistakes a game, whereas the average >top grandmaster makes at most 1 to 2 bad moves a game". > >How Kasparov managed to win game 1 in the second match >is still a big mystery for me, as he obviously was offering to >lose but the computer didn't take his generous offer and managed >to fatally weaken its pawn structure. > >So we should instead focus on the many bad moves it played as excellently >described in issue june JICCA 1997 by Seirawan who gives move after move >a question mark from both kasparov and deep blue. Note that with ?! he >means that a move is dubious. > >I might lose in national competition a game if i make 1 move >marked with ? or 2 with ?!. I'm just 2271 national rated, so national >master soon. Fide master i also get blindfolded, IM will be a bit >tougher though. To beat an IM i need to play a game with at most a >SINGLE ?! or ? from my side unless the IM is real old. > >Basically the only reason why deep blue beated kasparov was because of >kasparov himself and that a4 move in game 6, opening the position. >I think in 1997 very >little PC programs played such aggressive moves, whereas nowadays >aggressive tuning of most programs is very common. > >On the other hand i can't imagine a nowadays program that doesn't know >that a doubled pawn on g2 g3 or g7g6 isn't so bad something Deep Blue >is doing wrong in at least 2 games. Also first game it never gets >the idea of playing e5 in opening, unimaginable for me that any >nowadays program would prefer moves like e6? there. Bad moves like h6? >i might imagine if pawnstructure code of a program is weak or nonexisting. > >So we basically can only conclude very few things: > a) search depth of deep blue was ok, 11 to 13 ply, > but not uncommon nowadays, though some programs like tiger and > rebel seem to forward prune immense amount of nodes, still they'll > get depths very comparable or deeper as this. > b) how do we ever compare loads of hardware with efficient > working software programs, those are nodes a second not comparable! > c) In this CCC group > we'll have only a bunch of nerds with except for a few exceptions like > an Uri Blass very little chess insight > to analyze any game, not to mention doing statements about the > the moves produced by deep blue; and if we basically we need to > rely on Seirawan's chess technical comment for the many obvious > openings setup mistakes made on both sides, thereby taking into account > that he wrote it seemingly pretty optimistic for IBM as he was paid > by them. > >The IBM versus Kasparov match is clearly the only match in history that >drew a lot of attention without the games getting analyzed very well. > >If i ask the average AI dude who talks about deep blue as if it's his >pocket machine whether he read some reports on the match, then i already >get a negative answer. Not to mention chess technical reports. > >Everyone focusses on some silly comment from the side of Kasparov about IBM >cheating. Obviously that's only to cover his own dirty behind. >If you study the games it was not cheating at all. It was kasparov >who clearly on chessbase magazine 58 if i remember well who was doing >as if he played a 5 year old kid, without thinking of the consequences, >focussing on getting himself some media attention for a few 'exciting' >games. > >But especially that last part: studying the games, never in history >has a match been so little studied as the ibm-kasparov games. > >If i ask chessplayers whether they studied them, i always get negative >answers. > >If i ask after the 20th match game Kasparov-Karpov Lyon 1990 then >most people will remember it. In the computerchess world we even are >regurarly busy with a position from this match: NOLOT #1 Nxh6!! Kasparov > >Pathetic that such a great player has played such a few bad games against >IBM, well the first so many matches he played against the computer he could >get away with it even, but obviously game 6 in the match decided, >nevertheless: "Where are all those chesstechnical >analysis to call this match a serious match?" > >Obviously is that we'll never see again a match again of Deep Blue by IBM >as it's buried forever, the simple reason being that stocks/shares of IBM >got up an arguable 22% right after deep blue won. If i look now at >the internet: http://finance.yahoo.com/q?s=ibm&d=b > >Then marketshare of IBM is 172 billion dollar. > >22% of 172B = let's say around 35B dollar that's at stake playing >another match if they lose it their market share may drop that 22% >again as investors might get convinced that the new cpu's are not >so good as they thought they were... > >And that deep blue was in fact hardware processor from 0.60 micron >was never clear to those dudes and i give you less as a 0% chance that >we'll ever see something playing under IBM-deep blue flag with some >camera's close, more likely the processors are given away to >some hardcore chesscomputer designers from which we can only >hope they'll play a few games with it on the internet, something >by the way promised by IBM which they never did of course. > >Note that with nowadays big progress on many sides of the game >it's sure much tougher to win with an old program. > >We could see that clearly at dutch open also where very courageously >Duck joined. He had made a NEW openingsbook for it which didn't >perform bad, but he searched 8 ply fullwidth or something. > >So he obviously had some serious problems despite that some people >like me were real bad prepared. In a drawn line it was completely >outsearched to start with and basically that lost for it together with >some positional aspects, which both programs didn't understand too well >bye the way. > >He was searching 8 ply against me over 10. Now that's on a nowadays >machine. So getting 11 to 13 for deep blue in 1997 fullwidth was real >good IMHO. I don't search 13 ply fullwidth for sure except pawn endgame, >yet i completely would annihilate deep blue for sure with DIEP, >especially seeing what it did do wrong, and not even caring for the fact >that my openingspreparement would be real bad.
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.