Author: Graham Laight
Date: 02:19:39 01/04/00
Go up one level in this thread
Hi Fernando,
I understand your concerns, but I don't agree.
If I get time tonight, I will search my old copies of SS for the article(s)
about how these calculations were made. However, I would like to make a couple
of points to be going on with:
* I don't think that you should put all the burden of proof onto our side of the
argument - our side is the one that has by far the heftier weight of evidence.
SS and SSDF have spent many years working out how to score computers on the Elo
scale. Our opposition seems to be saying that at GM level, something special
happens which prevents the rules of rating that work well in lower ranges from
operating correctly.
However, they base this on a "feeling" of how well the computers "seem to
understand positions", and that fact that they can occasionally beat them in
self administered games. I have no objection to this evidence, and I'm glad that
this evidence has been put forward, because it contributes to our understanding
of the issues involved - which is the primary reason for my enthusiasm for the
CCC. However, it seems to me that the evidence put forward on our side of the
case is better.
* The DB v GK ('97) match showed that, although the computer displayed some
"positional weakness" in its moves, other strengths it had were able to overcome
this deficiency in this particular match. Therefore, "conceptual" (as Albert
puts it) ability may not be enough by itself to guarantee a win in chess.
So - while I agree that absolute mathematical proof that computers are well into
the GM range, and might even be starting to approach to super GM range, does not
yet exist, it doesn't mean that this isn't the truth of the situation.
By the way, GMs out there ought to be challenging the computers while they still
can. Firstly, it's good publicity for the GMs, and secondly, the task of
challenging them and beating them is not getting any easier!
-g
On January 03, 2000 at 16:30:25, Fernando Villegas wrote:
>Correct. Maths means nothing if there is not a clear thing or concept behind
>them. But this, of course, live all us in the beginnig: how strong they really
>are. Problem is that the games between human and comps we have are not only few,
>but scarcely representative of the human kind that play chess in a whole. To put
>it simple; if comps plays publically only some games againts GM, we cannot know
>much about the program's strenght by the same reason you cannot say a man is
>short or tall if you compare him only with baketball players. Any system of
>measurement imply more or less some divertisity of that thing measured. Then you
>can say where it is in the general distribution curve of the quality measured.
>In chess a rating is not only the result of how you does againts Kasparov, but
>with many kind of players; if not, we does not get a point and so we all appears
>as patzers, with 0000 Elo rating. So even if we get many GM to play againts
>comps, we are not going to haver more than a very geneal, inaccurate measure of
>how strong they are.
>Fernando
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.