Author: Dave Gomboc
Date: 10:56:40 04/04/99
Go up one level in this thread
On April 04, 1999 at 13:32:30, Mike Hood wrote: >On April 04, 1999 at 11:21:55, Dave Gomboc wrote: > >>On April 04, 1999 at 10:02:37, Adnan wrote: >> >>> >>>I think Deep Blue at best is not better than 2650 to 2700 Fide. People are >>>overestimating it's strength just because on few games. The twelve games that it >>>played in 96 and 97 weren't really impressive. Kasparov *lost* the 97 match by >>>playing weak. But as far as the quality of games is concerened, Deep Blue made >>>silly blunders, even tactical blunders, something which computers are supposed >>>to be at best. For example, Deep Blue's tactical blunder in game 2 that would >>>have forced a draw, or tactical blunder in game 6, 1996, where it could not even >>>calculate a simple combination accurately and allowed 22. Bxh7+. >>> >>>If took real pity, I would rate it at 2650 to 2700 Fide -- AT BEST. >> >>You're certainly entitled to your opinion! You won't convince me without some >>concrete variations, though! Maybe provide the FEN position after Black's 21st >>move, a winning PV for White, including replies to important deviations for >>Black? Or did I miss a previously demonstrated win posted here? >> >>Dave Gomboc > >Adnan's point of view is valid. Deep Blue has played too few games to be >accurately judged. Any ELO ratings are rough estimations at best. Dave and Adnan >have different opinions that have to be left unjudged, unless IBM decides to >leave Deep Blue active for a long enough period of time to be examined. And >don't forget... IBM's newest supercomputers are bigger and faster than the one >that played against Kasparov, so the next incarnation of Deep Blue will play >even better. > >All the same, I consider myself a sceptic, as far as the strength of Deep Blue >is concerned. Is the program's strength what you'd expect from a computer so >powerful? To ask it another way, how strong would Fritz play on a computer with >that power? That's a naive question, because it's impossible to compare a single >processor PC with a multi-processor supercomputer, but I think you understand my >point. PCs have been relatively weak for years, so PC chess programmers have >worked hard on developing smart algorithms to make the best of the available >power. I remember playing against Psion on an 8 Mhz 286. That was a brilliant >program in its day. I fear (although I hope I'm wrong) that the Deep Blue >programmers rely too heavily on the power of the computers they have at their >fingertips, and not on the optimization of every last Assembler instruction. > >Dancer Adnan made a specific claim: that it made a tactical blunder in game six. I'm asking him to back it up, anyone can make a claim, but <shrug>. I didn't even bother with the game 2 "blunder", the search depth to see that Kf1 was only drawing was enormous and had many delaying quiet moves. The Deep Blue developers state in "Search Control Methods in Deep Blue" (March 1999 -- and I haven't forgotten about you, Ernst!) that their policy was to harness the tremendous search power to provide as selective a search as possible. That certainly doesn't leave the impression in my mind that they're trying to brute-force everyone, any hype about its nodes per second count notwithstanding. Assember optimization only buys you a constant-time improvement. Algorithmic improvements can be worth much more. Sure, maybe you can squeeze 25 elo out of hacking 80x86 assembly, but today's PCs are of a speed that I don't think it's necessary to do this anymore. Indeed, there are very successful programs that don't: I recall a discussion on CCC some time ago (involving at least Amir Ban and Bob Hyatt). Amir stated that Junior (one of the best programs today) was written in C++ and that the speed loss relative to C was negligable. (Disclaimer: I hope I haven't misremembered.) Dave Gomboc
This page took 0.01 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.