Author: Bruce Moreland
Date: 17:32:56 09/20/05
Go up one level in this thread
On September 20, 2005 at 16:16:15, Robert Hyatt wrote: >On September 20, 2005 at 04:06:56, Bruce Moreland wrote: >>No credit for imaginary work. > >Deep thought wasn't "imaginary". It beat the hell out of me and everyone else >for almost 10 years. And produced a 2650+ performance rating against human GM >players over 25 consecutive games... > >Deep Blue also wasn't imaginary. It beat plenty of GM players in exhibition >matches all over the world. I watched one at a supercomputing conference, for >example. My intent was to say that you can't extrapolate from there. Once they stopped, they don't continue getting better. So if we give them an Elo of 2700 at the time they quit, and someone else achieves 2750 ten years later, DB has not been earning "interest" on their Elo for 10 years. The thing that is 2750 is better. >DB 2 also wasn't imaginary. Unless Kasparov's imagination somehow made him >lose... We have watched the Dream Team wipe out and lose games, despite being comprised of the best talent on earth. There weren't enough games to know for sure that Kasparov could not solve the machine and/or overcome his own brain farts. IBM of course never had any interest in clarifying. Kasparov scored the first basket, IBM scored another, and then shouted "We win!", took their ball, and went home. We're scientists and engineers here, at least kind of. But this DB thing got into satanic marketing. Not the same goals or language or anything. Step one when discussing DB is to agree that it is possible to mis-apprehend the beast, and that IBM prefers this to be true. I refuse to give much credit to a machine whose owners want us to misunderstand it. I feel absolutely no burden to give them the benefit of the doubt on anything. I wouldn't acknowledge that the thing could play through to mate without crashing. >>We didn't see enough of DB to judge. Who is to blame for this? They are, >>although to be fair, "they" may be IBM corporate bean-counting deal to some >>extent. Should they benefit from this by becoming permanent hypothetical >>computer chess world champion? Hell no. > >I'm not sure I agree about the "enough to judge". I personally got to see far >more of the thing than I wanted to see, starting at the first event I played >them in in 1987 and stretching forward through the last ACM event ever held, in >1994. I wish the "final product" had been left standing for a while, but even >without it, it is pretty clear it was a dangerous box. As many human GM players >will certainly attest. We know a whole whole lot more about Fritz. Is "Chauncey Gardener" cut out to be Vice President of the United States? Or is he really "Chance, the gardener", who has come across some great clothes and is too confused to respond directly when people ask him questions? Of course it is possible to find out, but you have to be able to have a CONVERSATION with him in order to do it. I think it highly likely that the thing can play great chess, but I cannot *concede* that it can do it, or can do it consistently. Not enough evidence, by a long shot. You go back and forth on this issue of DB or DT. If it plays a bad game and ends up buried up to its ankles upside-down in dirt, that was DT. If it beats every micro back in dinosaur-time, that was DB. It's either one continuous project or it's two different projects. It can't be both. So you either saw enough of it, in which case that bad game against Fritz, and the other bad game against WChess, go on its record, or all you saw were the games against Kasparov, in which case we have a dozen games from something that could have played tens of thousands by now. Nothing can be concluded from a dozen games. >yes, most programs today, on hardware such as what I used in the WCCC, are also >very dangerous. But we are certainly not "far ahead of deep blue" otherwise we >would all be producing 3000+ performance ratings in long GM games... We aren't. > Yet. So while we might be (today) in the same ballpark, we _are_ in the same >ballpark. We haven't left them in our dust (yet). I've regularly seen search >speeds of 20M nodes per second on the WCCC box I used. I have seen 40M on an >8-way dual-core. That is under a factor of 10 slower than DB. So its "edge" >today would certainly have eroded to something fairly small. Unless they had >continued the project... > > > > > >> >>You don't get to win if you don't play. >> >>bruce > > >When you think about it, they both played _and_ won. Maybe not as much as we'd >like, but certainly DT played plenty and showed its strength. The rest is an >issue with the IBM bean-counters most likely... They have not played *since* then. They haven't played for 8 years. bruce
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.