Author: Robert Hyatt
Date: 07:47:26 08/02/01
Go up one level in this thread
On August 02, 2001 at 01:38:14, Bruce Moreland wrote: >On August 02, 2001 at 00:46:24, Robert Hyatt wrote: > >>Deep Thought beat the living hell out of every computer chess program for a >>period of 10 years. It lost one game to Fritz in 1995. DB2 was 100 times >>faster, with a better evaluation. To claim that a program that can only >>search 2M nodes per second is as good or better is nothing but pure nonsense, >>no matter how it gets justified. > >It drew with WChess in the same tournament, I say in order to give Dave his >often overlooked due. > >I don't care whether it makes sense to say it or not. All of us have a right >not to have the ghost of Deep Blue sticking its thumb up our butts for the rest >of our lives. I don't follow. There were three "greats" in computer chess. Slate during the 70's and early 80's, Thompson during the late 70's and early 80's, and then Hsu from 1987 forward. Deep Thought and Deep Blue pretty well proved themselves to be the cream of the crop, from 1987 forward. They didn't win in Hong Kong in 1995. They really didn't want to enter, because they were working hard on the DB1 system for the kasparov match the next year. But they did, and they lost "the game". They were a dominant force in computer chess for 11 years until they 'disbanded' after the last Kasparov match. I don't see how that will ever "go away" any more than the polio vaccine developer will be forgotten either. It doesn't bother me... In a few years, DB2 will be just an amazing machine for 1997, but quite common for (say) 2010. It never bothered me that Belle was unbeatable for a couple of years, that chess 4.x was unbeatable for several years, etc. It just made me work harder. Until I finally caught Belle. Of course along came Hsu and caught me and then some. And someone will come along and catch him one day too. > >>It is a _lot_ different... because Frans knows better, as do I and many >>others. I doubt he would have played them a best 2-of-3 in 1995. He took >>the win and ran. Because he _knew_ he could not repeat such a lucky break >>a second time... nor could any other program either. > >He didn't have to play them a second time. That was round five. He won. >Because he'd lost in the first round, and won the last four, he had four points. > DT2 had won its first three games, drew the fourth, and lost the fifth. That's >3.5 points. I've never heard anything that says that you have to give someone >with a half-point less a rematch. > What I said was that Frans won that game. He would _not_ have been interested in extending that to "best 2 of 3". Because he would know that the chance of losing the other two games would be very high. I wasn't suggesting a playoff in that event at all. I was suggesting that Frans knew that he had been very lucky. _I_ know he was very lucky. Since you can count all the micros that beat DB over a 10 year span on 2 fingers and have a spare. >Fritz was at the next open event, in 1999. DT/DB could have shown up and played >there, no problem. I don't know what Frans thinks about his chances in that >game. Perhaps he would expect to lose. But you can't fault him for being >"scared" or "running" when he's played in every event since then, and DT/DB have >played in *zero*. > I hope I didn't imply he was "afraid". I hope I did imply that "he is smart enough to realize how dismal his chance of beating the real Deep Blue 2 machine would really be." >>I only have to think about the 200M nodes per second to realize how good it >>was, even had I not _seen_ it play games. To think today's programs are even >>close is just something I can't believe anyone would take seriously. At best >>it is marketing hyperbole. At worst it is dishonest. Either end of the thing >>is _bad_. > >It's a big number, but life is full of upsets. One of the rules of sport is >that you have to play the game, you can't just declare yourself the winner based >upon everyone's perception of you. Not with one game, no. But for a protracted match, do you think you could beat them? I don't. Not on any hardware I can touch today. For one game? Even if they are 400 points better than me I have a 1 in 16 chance of winning one game. Which isn't horrible. But I have essentially no chance of winning a match... > >>>If they want the reputation, let's see them rejoin the community and play a >>>bunch of games. >> >> >>You can only do what the people in charge allow. They _did_ compete from 1987 >>until the last ACM event that was held. They did quite well during that span >>of time, proving that they were simply in a class by themselves. In the last >>ACM event, they had to forfeit a game due to a monster power failure at the >>Watson center, yet they _still_ won the tournament outright with no tie-break >>system needed. > >A wonderful result, but the ACM events are also not a permanent title. If you win every one of them, it is "close" to permanent. > >>>If they don't want to do that, it's hard to provide substantitive >>>counter-argument against the "Deep Blue Sucks" people, and I don't see why any >>>of us should try. >>> >>>Bob, they left and they aren't coming back. They didn't make much of a real >>>legend before they left. Posthumous attempts to create a legend for them >>>shouldn't be our business. > >>Didn't make a legend? Were you present in 1987, 88, 89, 90, 91, 92, 93, 94, >>etc? I was present and saw what they did to everybody that showed up. Over >>and over. Any AI book names DB as the strongest chess engine yet built. Until >>someone seriously challenges that 200M node per second number, the books will be >>right... > >I think it's important to identify what we are talking about here. DT was what >played between 1987 and 1994, and no, I wasn't there for that. actually DT2 played starting around 1991 or 1992. That was the first dual-chess processor version. it might have even been earlier. > >DT2 was some sort of partial prototype of DB, and it played in 1995. Yes, I was >there for that. > >DB played 12 games in 1996 and 1997 verus Kasparov. I saw these on the internet >like everyone else. > >DT was dominant in its day, but I don't think it would be as dominant now. >We're solving some of the Nolot positions in less time than it took DT in circa >1994, and DT isn't known as a postitional monster. Correct. But DT was a 2M node per second machine. DB2 was 100x+ faster. What does _that_ do to DT's Nolot solutions? It would get most in a tournament game time control. We aren't doing that for more than a couple. > >DT2 was 3.5 out of 5.0, which may have been a fluke but there it is. > >DB is 0 out of 0 so far against computers. That record doesn't convince me of >anything. I don't think anyone should have to declare them permanent champion >for the next twenty years or however long it takes for us to match their node >rate. > >Maybe the Spruce Goose was a wonder to fly, but all it took was the one short >hop, and there it sits in a warehouse. > >Maybe DB was wonderful against computers. It's against any sane scientific >principle to say YES or NO based upon the evidence, which is nil. We not only >don't have games played by the thing, we don't have any public examples of how >fast the thing solves any test positions. I disagree. DT was definitely wonderful against computers. DB2 was 100x faster and had more intelligence. It _has_ to be better. Kasparov smashed deep thought 2-0. He beat DB1 (but didn't smash it) and he lost to DB2. That also suggests DB2 is much stronger that DT. And as late as 1995 I don't believe any micro program had any significant chance against DT at all, based on several years of games it had played. Yes, bad things happen. I lost to Dave that same year I believe. My program tried to play the game with multiple kings on the board due to a SMP problem. It failed. But that doesn't mean that it wasn't strong as hell when it was "right"... I would much prefer to rely on 10 years worth of data from DT as opposed to "The Game" in Hong Kong. > >I don't say they suck, but I do say that without a lot more than they've >produced so far, it's correct to remain agnostic about them. It's not possible >to know how strong they are against computers. Any assertion is based upon very >thin evidence. > >bruce I don't claim "they are god" either. But I do claim (1) they were as good at chess as _anybody_ currently doing computer chess. Perhaps better, but certainly not worse; (2) they were dominant when they competed. More so than _any_ other computer chess machine around including chess 4.x in the 1970's. (3) their machine did not "suck" as many want to say; (4) their machine was not based on a stupid search, plus a primitive evaluation. The first one was, yes. But it evolved over 10 years into something much more "aware". It had to to play against Kasparov.
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.