Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Tiger against Deep Blue Junior: what really happened.

Author: Robert Hyatt

Date: 19:19:54 07/26/00

Go up one level in this thread


On July 26, 2000 at 17:00:12, blass uri wrote:

>On July 26, 2000 at 15:28:49, Robert Hyatt wrote:
>
>>On July 26, 2000 at 14:38:14, Alvaro Polo wrote:
>>
>>>On July 26, 2000 at 14:07:26, Ed Schröder wrote:
>>>
>>>>On July 26, 2000 at 07:31:12, Dave Gomboc wrote:
>>>>
>>>>>You can believe any of the scholarly sources of information with Hsu and/or
>>>>>other members of the DB team on it, like IEEE Micro for instance, or the paper
>>>>>from an AAAI workshop last year.  Or, you can believe the web page.  It's your
>>>>>call, Ed.
>>>>>
>>>>>Dave
>>>>
>>>>Okay... I get the point. Here is what I mean with all those contradictions
>>>>between the IBM pages and other papers, statements and I will be detailed.
>>>>
>>>>This forum is about playing strength mainly, 90%? more..?
>>>>
>>>>For this reason only DB is a hot topic even after 3 years. It is about
>>>>WHO is the STRONGEST. That is the MAIN question here and you can taste
>>>>in almost every posting.
>>>>
>>>>We want to know which program is the best TODAY. This topic for instance
>>>>started right after the Dortmund event, "is DJ stronger than DB" and off
>>>>we went... another new endless DB discussion. New DB discussions will
>>>>happen again and again until the question WHO is the strongest TODAY will
>>>>be answered. Not really something to look forward when candidate number
>>>>one is not present anymore.
>>>>
>>>>Now what bothers me is the following:
>>>>
>>>>PRELUDE: in this posting "IBM" stands for the whole DB team, Hsu and co
>>>>included.
>>>>
>>>>198x - 1995: DT/DB dominates computer chess against 386/486/6502 machines.
>>>>
>>>>1995: DT lost the world championship in Hong Kong (just 3.5 out of 5)
>>>>
>>>>1991: DT doing 7 million nodes (source IBM)
>>>>1995: DT doing 7 million nodes (source IBM)
>>>>      DT doing 2 million nodes (source Hsu)
>>>>
>>>>1996: IBM claims 100 million nodes.
>>>>
>>>>100 million nodes fixed a huge image problem IBM had because of the loss
>>>>of last year.
>>>>
>>>>It did the trick at least in my mind. Surely Hong Kong was a mistake and
>>>>it would never happen again. So there was no question about it who was
>>>>the strongest. Note that our minds are feed with information that comes
>>>>from a commercial source and this information is still playing a huge
>>>>role here in CCC even today trying to answer the question, WHO is the
>>>>strongest. Who can beat 100 million nodes a second? Answer: none and
>>>>this dominated our minds, even today.
>>>>
>>>>Until last year I had no single reason to doubt the numbers (or any other
>>>>information) that came from IBM and took everything for granted that came
>>>>from IBM. Today I doubt the jump from 2 million to 100 million. IBM had
>>>>a very good reason. Image was on steak after the Hong Kong debacle. IBM's
>>>>answer: 100 million. Problem solved. Most of us happily fell asleep.
>>>>
>>>>Now I am not the person to start conspiracy theories but this one touches
>>>>the soul and existence of CCC which is playing strength and WHO is the
>>>>STRONGEST.
>>>>
>>>>IBM has proven being a sloppy partner when the topic is giving right
>>>>information. I like to mention all the contradictions as pointed out in
>>>>the previous postings of the last days.
>>>>
>>>>I also like to point to the habit of IBM to come up with explanations
>>>>every time they lose or make a bad move. The last example of this are
>>>>the 6 games of DB-JR in Paderborn last year. The furious answer came a
>>>>few months later high lighted as follows, literally:
>>>>
>>>>  "Ed Schröder has misused this opportunity at will, to make false
>>>>   advertisement for his program".
>>>>
>>>>The story is pointed out on: http://www.rebel.nl/reb-css.htm
>>>>
>>>>Note this page needs an update as at the time I gave the case a rest.
>>>>
>>>>In email with Hsu one of the authors of Deep Blue it became clear
>>>>to me that at least one of the main items of their defence (Deep Blue
>>>>Junior thinking only 1 second per move) was not true as Hsu admitted
>>>>that Deep Blue Junior at least used "panic time" when Deep Blue Junior
>>>>found itself in trouble. Apparently this confession came because in the
>>>>discussion I kept my point straight up having seen Deep Blue Junior
>>>>thinking 10-15 seconds frequently, having witnesses of that too.
>>>>
>>>>I still keep my point straight up Deep Blue Junior using it's time
>>>>control in the same way as any other chess program playing blitz.
>>>>
>>>>After that confession I stopped believing the whole story Deep Blue
>>>>Junior just being a DEMO program of 2200 elo. Every time the machine
>>>>loses it is something else. Saying first "one second" as a main point
>>>>of your defence and take it back later doesn't sound good.
>>>>
>>>>Furthermore the documentation that came with the program (nor on the
>>>>screen) did not mention any of the playing strength restrictions as
>>>>pointed out by Hsu in the CSS article (see the URL above) so we took
>>>>the thing for real otherwise we would not have touched the machine.
>>>>
>>>>I take these 6 games for real coming close to DB-JR real playing strength.
>>>>
>>>>These 6 games were played out of curiosity. The opportunity was there
>>>>to learn something about the machine and we were expecting to be beaten
>>>>badly by the monster. The opposite happened leaving Christophe, the
>>>>spectators and myself in astonishment behind.
>>>>
>>>>There is also no doubt in my mind any other strong (amateur) program
>>>>could beat the Deep Blue Junior in Paderborn. I don't want any credit
>>>>for these games. This is just for the sake of the truth.
>>>>
>>>>Note: I have nothing against Hsu in fact I removed the 6 games from the
>>>>Rebel Home Page on request of Hsu. Not that I was in agreement with that
>>>>but I did it out of respect for him and his achievements and the joy his
>>>>program has given me and the whole chess community.
>>>>
>>>>For me 2 points are most crucial:
>>>>
>>>>a) 2M (1995) to 100M (1996) real or not.
>>>
>>>I don't know. Probably 100M are real (my opinion only)
>>>
>>>>b) Paderborn 1999 (DEMO program or not)
>>>>
>>>
>>>Christophe's results (1.5-1.5 against Chess Tiger on a Pentium 150Mhz with Tiger
>>>thinking 20 times more time than DBJ) seem to indicate that it was the real DB
>>>Junior, not a demo.
>>>
>>>Alvaro
>>>
>>
>>Not really.  His hardware was certainly available in 1997.  And in 1997 a
>>single crippled version of a DB chip played the famous match that causes
>>so much discussion here.  And did quite well.  So his result was really
>>quite good, even against a broken 1 chip version of the program, when
>>you think in that context.  DB junior is a real horse, as many GM players
>>can attest to after losing to it during the tuning process for the 1997 DB
>>match.  I saw it beat several GMs like drums at exhibition matches at
>>conferences...
>
>I do not know about public tournament time control of Deep blue Junior against
>GM's and this is the real test.
>
>I am sure that tiger(p150) can also beat GM's at fast time control(at least GM's
> that did not buy tiger and do not know about the weaknesses of tiger)
>
>Uri


DB Jr wasn't beating GM players at blitz.  It was beating them at longer
time controls.  I think the last match I watched was something like game/30
but with an increment on the clock.  It was short enough to make it interesting
to spectators, but not so short as to become a tactical oversight issue.



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.