From: Vladimir Nesov (robotact@gmail.com)
Date: Wed Jan 23 2008 - 15:28:01 MST
On Jan 24, 2008 12:23 AM, <x@extropica.org> wrote:
> On Jan 23, 2008 12:58 PM, Matt Mahoney <matmahoney@yahoo.com> wrote:
> >
> > AIXI/Occam's Razor favors the simplest algorithm, not the fastest program.
>
> PLEASE don't promulgate that misinterpretation of Occam.
>
> We should prefer the model positing the fewest **ontological**
> entities, not that which is "simplest."
Nope, it's all about length of description, and doesn't hold in
general if description isn't very good. It kinda works because
description tends to succintly represent information picked from
described domain, and this way length of description starts to
indicate probability of items from that domain. Thus shortest
representations tend to represent more probable hypotheses, and you
are usually better off with more probable hypothesis. But you can as
well devise a representation that doesn't reflect probabilities well,
especially for most probable hypotheses, and end up making Occam's
Razor heuristic fail. So stick with probabilities.
-- Vladimir Nesov mailto:robotact@gmail.com
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:01 MDT