Author: Robert Hyatt
Date: 22:21:27 10/12/99
Go up one level in this thread
On October 12, 1999 at 21:10:10, Dave Gomboc wrote: >On October 12, 1999 at 20:03:58, Amir Ban wrote: > >>On October 12, 1999 at 19:24:31, Dave Gomboc wrote: >> >>>On October 12, 1999 at 18:51:25, Amir Ban wrote: >>> >>>>On October 11, 1999 at 21:25:30, Manuel Rodriguez Blanco wrote: >>>>> >>>> >>>>snip >>>> >>>>>Rebel Company about the accusations: >>>>> >>>>> >>>>>We were not aware of any restriction that playing against Deep Blue Junior was >>>>>forbidden. Deep Blue Junior was there and we took the opportunity to find out >>>>>more about this program. The result was posted as NEWS no more no less and we >>>>>don't have (nor had) any intention to include Rebel's victory over Deep Blue >>>>>Junior in our advertisements. >>>>> >>>>> >>>>>We can't confirm the "one second" time control of Deep Blue Junior. Deep Blue >>>>>Junior indeed played on a "one second" time control using its default time >>>>>control but raising the time control caused Deep Blue Junior to think a lot >>>>>longer (up to 10-15 seconds). >>>>> >>>> >>>>This point is important, and I didn't see it discussed. If so, Hsu is >>>>misinformed about what features were made available to users of DBjr, and Rebel >>>>played a version that was much stronger than he thought. I don't remember if >>>>they played equal time controls, but if they did, it may well be that the >>>>contest was fair or close to fair (we need to assume that the DBjr server was >>>>not overloaded, which to me seems likely, because it was hardly used intensively >>>>at the stations I saw in Paderborn). >>> >>>The run-on sentence "The program runs on a single chip, [has] a thinking time of >>>one second per move (including initialization and independently of the thinking >>>time set by the user interface), without permanent brain and using a primitive >>>evaluation function." seems to discuss this point. (It also makes me wonder if >>>the statement was translated into German for the magazine, then back into >>>English. :-) >>> >> >>This was said without taking into account Schroder's statement from above about >>the time control. It seems to contradict what Hsu & Campbell say, unless you >>make the improbable assumption that the interface asked for time control and >>then emulated it while really thinking only one second (what for ?). It's more >>probable that they weren't directly involved with setting up this demo and were >>misinformed about its capabilities. > >The "impedance mismatch" between the back-end and the front-end does seem weird. >Presumably they weren't involved in the creation of the interface. So the >question is whether what the front-end shows is right, or the people who worked >on the back-end are right. If they worked on the back-end, then I'll buy their >explanation. It does seem quite illogical though, so the possibility you raise >has some weight. Bob posted once that "Hsu was asked to..." set up the demo. >So my belief boils down to "do I believe Bob?", and in this case, I do. based on my discussion with Hsu, I would 'guess' (emphasis on that word) the following: he/Murray/etc were directly responsible for the single-chip 'engine'... because who else could actually modify that code. I doubt they had anything to do with the front-end that connected this stuff to the web. Amir mentioned that the 'lobby machines' weren't heavily used to account for time variability. I would only mention that the 'net' is a lot bigger than just the machines at Paderborn. Supposedly this thing gets a _lot_ of usage. But if only 10 people use it at the same time, each move will take 10 seconds + network lag. There is only one 'engine' and one 'chip'. No way to avoid slowdowns. And whether the front-end allows time changes or not, Hsu seemed pretty positive about how the 'engine' works, as he explained it to me, then to Friedel 2-3 months later, in _exactly_ the same terms. > >>>>I wouldn't consider Rebel beating DBjr a surprise. Even assuming full-DB to be >>>>the equal of Kasparov (doubtful), DBjr should be much weaker, and not more than >>>>Rebel. Besides, isn't Rebel's record against rated players better than DBjr's ? >>>>I don't know the statistics, but I got the impression that DBjr's record is not >>>>too good. >>> >>>FWIW, I remember an article in Chess Life where GM Soltis annotated a game that >>>he won over DBjr. IIRC, he said that in the several months it had been touring, >>>only 2 other people had won a game. (Time control was G/15.) >>> >>>I did say FWIW, but I'll grant that it's probably not worth much. >>> >>>>I think it's pretty low to say or imply that Ed played DBjr for cheap publicity. >>>>Obviously he did that out of curiosity. It would make better business sense to >>>>concentrate on the WCCC rather than play improvised games in the hall, but >>>>people who are curious do what is intersting, not important. It's clear from Hsu >>>>& Campbell's letter and the clarification from Friedel that they are not curious >>>>in the least, and that they don't give a damn about their peers respect. That's >>>>a good enough reason not to respect them, and I don't. >>>> >>>>Amir >>> >>>I agree re: curiosity. I don't agree re: peers. Possibly your idea of who >>>their peers are is different than mine. >>> >> >>Their peers are people who take an interest in computer chess for its own sake, >>or at least their fellow competitors. Ed is certainly one of their peers. >> >>Amir > >I think that if you work toward beating the world champion for a decade and >finally accomplish it, only to have GK and a whack of others call you a cheater, >it would be a natural reaction to start giving less weight to other people's >opinions -- as a self-defence mechanism, if nothing else. > >All the same, I've met Murray Campbell and had one extended discussion with him. >I've seen him interact with Tony Marsland and Jonathan Schaeffer. I've heard >him answer all kinds of questions about DB. I've never received the impression >that he didn't care about the respect of other people. He's always seemed like >a very polite person to me. > >Some people believe DB is a cut (or ten) above the micros, while others don't >agree. IBM doesn't let the machine play, which means that DB team members are >not going to be able to convince remaining skeptics that DB is really strong... >so why should they expend the effort? It would just be a futile exercise, and >they're smart enough to know that. > >Both Hsu and Benjamin are apparently writing books... hopefully this will give >us some more information. > >Dave Hsu's will be a great read. I proof-read it for him, and as my wife will tell you, I read it front-to-back without putting it down. Very interesting reading for all, when it comes out. BTW I agree about all the DB guys. I have known Murray almost forever, and could tell you how I became the first ever programmer to use PVS, albeit accidentally at an ACM event, at his urging. And I've known Hsu since 1988. I wouldn't call any of them 'close friends' but I definitely respect what they have done.
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.