Author: Albert Silver
Date: 04:12:24 12/24/99
Go up one level in this thread
On December 24, 1999 at 05:51:41, Graham Laight wrote:
>On December 23, 1999 at 20:46:54, Albert Silver wrote:
>
>>>>I would be fascinated to hear about this advantage it converted in the second
>>>>game.
>>>
>>>It obtained the advantage courtesy of its opening book in game 6. It converted
>>>it into a win.
>>
>>Reread my question.
>
>OK - I thought you meant the the second game it won.
>
>In game 2, the Spanish "Ruy Lopez" opening was used. Through good play, DB (as
>white) managed to maintain an advantage throughout the game - slowly squeezing
>the life out of GK's position. Finally, it obtained a position in which GK (and
>all the experts present) believed that he was lost, and so GK resigned.
>
>In fact, it was later shown that there was a complicated repetition draw
>available - but I don't think that this is relevant to this discussion.
>
>>Rebel Tiger has not been tested by the SSDF, you are mistaken.
>
>I refer you to http://www.rebel.nl - on this page, Ed clearly claims that Rebel
>Tiger has been tested by SSDF, and that this program now tops the SSDF rating
>list with 2696 points.
You are right about the claim, but the claim is wrong as well. SSDF tested Chess
Tiger 12 for DOS which not only is slower than the Windows version, but also did
not have the Rebel opening book. I think Rebel Tiger will be much stronger with
the added speed, and Noomen's top notch book.
The internal engine for CA5 IS for Windows, but the book I was using (I have a
newer one now but it needs testing) was quite basic. It wasn't really an issue
as my opening knowledge is quite basic as well, and Tiger didn't lose because of
the opening.
>
>>I have Chess Tiger the internal CA5 engine using a CA book, Fritz 5.32, Fritz 6
>>(only played a couple of blitzes so no notch on my quiver), Hiarcs 7.32. Rating?
>>No idea. Those were among the top though.
>
>Fritz 5.32 = 2574 +- 24 64Mb 200 Mhz
>Fritz 6 = not rated
>Hiarcs 7.32 = 2572 +- 27 64Mb 200 Mhz
>
>>>
>>>>>However, it's not good enough to beat them in the comfort of one's home.
>>>>
>>>>Really? I don't think that would make any difference. I wouldn't beat a GM in
>>>>the comfort of my house either.
>>>
>>>OK. But it has been statistically demonstrated that "self administered" IQ tests
>>>tend to score more highly than "overseen" IQ tests! Can't think why... :)
>>
>>What do statistics have to do with this? I don't have to go about proving I have
>>beat them, it's a _fact_ to me (your skepticism is your problem). I have never
>>beat (nor played) a human rated 2700, but I do not need to to know that I will
>>be pulverized on the board. For me that makes it CRYSTAL clear the programs are
>>not knocking on the 2700 door. I am not trying to convince you they are not, I
>>am telling you from my knowledge. You can believe it or not at your convenience.
>
>>>>>If he
>>>>>played them under competitive conditions, some extra considerations would come
>>>>>into play:
>>>>>
>>>>>* Some of the evaluation factors would be changed, so that he may not be able to
>>>>>predict their moves so accurately
>>>>
>>>>Predict? I honestly don't understand what you mean by this.
>>>
>>>I am suggesting that if you play your computer a lot, you'll learn to predict
>>>what moves it's going to make.
>>
>>I do not cook the games, and I play them infrequently.
>>
>>>
>>>>>* The whole thing, from opening books to evaluation factors could be tuned to
>>>>>produce an optimum game against HIM.
>>>>>
>>>>>This is the reality that GK faced against DB in May '97.
>>>>
>>>>What is your point? I never said I was unbeatable, not even that I had a
>>>>positive score against the damn things. :-) Only that I could NEVER have scored
>>>>a single point against 2700 human player. Ever.
>
>This is all good and valuable information which I will take into account in
>formulating my own opinions on the subject. But, at the risk of sounding
>disagreeable, but in order to clarify my thoughts, I have to offer the following
>observations:
>
>* The machines you play are only rated at about 2570 on the SSDF list - barely
>into the grandmaster range
You cited 200 Mhz machines as well. I suspect they do a little better on my
K6-2/350. Also, 2570 on the FIDE list is not "barely into the grandmaster
range". It constitutes the rating of a strong grandmaster. 2570 on the SSDF list
may barely be in the human IM range though.
>
>* I do not question your integrity (well - only slightly!) - but when one
>self-administers a test, there are many ways one can balance the odds in one's
>own favour. Perhaps looking at the computers' analysis, for example, to "pass
>the time while the computer's thinking", or maybe the odd move back because
>"that's not what I'd really have played", or possibly setting the computer to
>play aggressively because "it's more fun!". It may be that you're a man of 100%
>integrity - but unfortunately I've seen too many chess players (who don't favour
>the computers) play!
I can't attest to other players. When I have a new program, one of the first
things it gets (after I've had fun playing around with it), is a serious match,
which ends when I score my first victory. The match is played at 40/2 and I play
the game on my chessboard which I keep on the side.
>* You appear to be admitting that you lose most of your games against these
>programs
Yes, undoubtedly, but the point is I would be doing worse against a 2600+ human
player.
>
>* I've observed that some players who play computers regularly improve their
>playing strength without realising. It may be that if you enter a rated
>tournament, your own rating will rise!
Maybe. Doubtful. I have played in tournaments where my TPR was considerably
higher than my rating, and even so, lost rating points.
I have been quite inactive lately and would need to get back to the books to
hope for any real changes in my chess strength.
>
>>>>>If anti-computer chess is alive and well, why did IM Dan Hergot lose to Hiarcs
>>>>>in early '97 - to what is now an old version of Hiarcs on old hardware?
>>>>
>>>>He didn't play anti-computer chess.
>
>>>>>And why did GM Ruslan Scherbakov lose to Rebel Century?
>>>>>
>>>>
>>>>He didn't play anti-computer chess either. Furthermore, I never said a computer
>>>>could never beat a GM. That would be completely wrong. I only said they are not
>>>>GMs IMO.
>
>Most grandmasters did not think that DB played good positional chess against GK
>(except in game 2), and GK most definitely DID play anti-computer chess!
>
>>>>>And why did the computers beat the humans overall at the last Aegon tournament
>>>>>(1997)?
>>>>
>>>>Were the humans all GMs overall?
>>>>
>>>> Albert Silver
>>>
>>>No - but not all the computers were top notch either.
>>
>>What they do against weaker players is not the issue. You claimed they were
>>beating up on GMS. What was the score of the GMs rated over 2500?
>
>Unfortunatley, I don't know what the ratings of the top computers were at the
>time of the last Aegon. From memory, I think that the computers have risen quite
>sharply in the SSDF ratings over the last 2-3 years.
>
>>>Is there anyone from the SSDF following this thread?
>>>
>>>Can you tell us when the last time the ratings scale was checked for correlation
>>>with the FIDE rating scale, please?
>>
>>It never correlated with the FIDE scale. It _did_ correlate to the national
>>Swedish Elo ratings up to 1993 through SSDF's efforts to organize and include
>>games played against human opponents.
>>
>> Albert Silver
>
>As far as I know, nobody has seriously suggested that the Swedish Elo ratings do
>not correlate well with the FIDE ratings.
Correlating well with the FIDE list is not correlating with the FIDE list. The
human ratings that influenced the SSDF list were Swedish Elo ratings. Remember
also that FIDE ratings at the time only accepted minimum ratings of 2200. In any
case, since 1993 the pool of players has changed, and the ratings within measure
the relative strength of its players. This is not to say it has no value, but it
does not measure how well a computer will do against players of a different
pool, the FIDE list, which is also completely isolated as it does not include
computers of any kind. It only measures how well computers do against
themselves. That is why Shroeder's GM challenge will be the first really
revealing test on the relative strength between humans and computers at that
level.
Albert Silver
>
>-g
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.