Author: Rolf Tueschen
Date: 12:46:53 06/28/02
Excuse me for the little dwlay in my answer to this important posting. CCC has a bad technique when you want to discuss older posts. You can't find them online. I deleted the more unfriendly note by Jose here in the quoted article because he has nothing to add to the content of the debate. Subject: Re: Normal distribution no way for machines of diff. generations QED From: Dann Corbit E-mail: dcorbit@solutionsiq.com Message Number: 234506 Date: June 07, 2002 at 14:50:39 And at the same time, some good points are raised by the dissenting voices. The matter of control in an experiment is a crucial one. I have seen many experiments run, even by a staff of PHD's, where the control of the experimental design was a very poor one. We should be forced to think about what is allowed to vary and what must stay the same. We should contemplate what we are measuring and also what we are not able to measure. It is also important to consider the overall experiment in with the conclusions which we want to draw. If more than one thing is allowed to vary, we cannot say which thing caused the improvement. ### Well, for science this is very trivial. Nobody would dare to argue against it. But the SSDF people do exactly this. Of course they say that SSDF is no science. But the irony is that Elo and its maths _is_ science. And now the execution of the whole SSDF: You can't simply take a scientifical method and make the "experiment" in your personal _unscientific_ way. And when critics come you answer that it's not about science. That is, here I am in opposition to Dann, simply not honest. Because on the base of the Elo maths the SSDF pretends that its "results" have a meaning. But Dann himself said that without exactitudiness you have no results at all. No matter if you use Elo maths or engage clairvoyants. Now the most important point is the betraying of more naive readers of the SSDF ranking list. Let me also repeat a very mean method of betrayal. SSDF gives the error margins. This is ridiculous because it is a pretending as if the SSDF is almost having scruples to pretend a thing. But the truth is that the whole practice of SSDF is nonsense, without any scientifical background. This may sound brutal, but for me the pretension of the SSDF is much more brutal, when every two months average people and computerchess lovers are deceived. We must not forget that every chessplayer knows of the Elo system and therefore the use of Elo in the SSDF is a serious betrayal. If you as a reader may doubt what I write here, then please try to understand what Dann just has written. This control thing is not some spooky bogus. No, it is the core of our complete experimental science whether natural or social. It is - if you want to have the impression for the importance - even more important than all fundaments of the Constitution of the United States of America. That is why it is so bad to see how arrogant the SSDF people behaved in this debate. They seem to think that they had the right and the reputation to do what they want with the Elo maths. But this is false. We all, included the SSDF people, all prgrammers who have programs in SSDF, the business, the media, we all are victims of a false ranking list. I think this isn't a minor important thing. Let's rest the debate. When I again started it I really believed and hoped that SSDF might understand the debate and might change its practice. Now I no longer have any hope. Rolf Tueschen #####(end of my article) For instance, the program version changes, the opening book changes and the hardware changes. Was it the program version change or the opening book change or the hardware change that resulted in a change in strength? We cannot say. We can only say "This system appears to be stronger than that system" but not 'why' the strength difference appears. One example of this problem is that people will see hardware increase in the SSDF list and use this to make projections about MHz increase to increase in ELO. Unfortunately, other variables have changed at the same time. So we cannot say with any certainty that the strength change was due to the hardware. We tend to make models in our mind and to simplify them to get simple answers. But Einstein said something pertinant: "Things should be made as simple as possible, but no simpler."
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.