| David Kane 2006-11-19, 9:06 pm |
|
"Louis Blair" <lblai@blackburn.edu> wrote in message
news:1162322358.246753.74960@e64g2000cwd.googlegroups.com...
quote:
> "... the rating scale itself is an arbitrary scale, open
> ended, . . . with no reproducible fixed points. ..."
> - Arpad Elo, Chess Notes, 1988.
> _
> Somebody once proposed that one could have a
> reproducible fixed point by having a program that
> would calculate all possible legal moves in a
> position and randomly select one. My guess is
> that this would not produce results that would
> be consistent enough to be helpful. What do
> other people think?
>
I have partcipated in long threads in the past
debating the principle of "rating the moves" as an alternative
to the Elo system, which rates results only. While not wishing
to reopen that debate there is an interesting article
touching on many related issues in
http://www.chessbase.com/newsdetail.asp?newsid=3455
As to your suggestion, I think we could safely fix the
rating scale by defining any machine algorithm as a certain
rating. I don't see why it would have to be so weak as you
suggest, and offhand it would be more sensible to use one that
plays more like a player in the middle of the distribution.
You'd still have to have people willing to play the machine in
tournament conditions. Whether there is really any
practical advantage in doing this, I don't know. This would
seem to be a good way to determine rating inflation/deflation
but other methods may be good enough. The one advantage
that pops to mind is that it could depoliticize ratings to some
extent. It's human nature for people to blame their rating
declines on the "unfair system".
|