From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Tue Aug 01 2006 - 14:59:21 MDT
Scott Yokim wrote:
>
> "It appears to us that actual scientific practice is guided by instincts
> that have not yet been fully recognized, much less analyzed and
> justified. We must take into account not only the logic of science, but
> also the sociology of science (perhaps also its soteriology). But this
> is so complicated that we are not even sure whether the extremely
> skeptical conservatism with which new ideas are invariably received, is
> in the long run a beneficial stabilizing influence, or a harmful
> obstacle to progress."
Or as Marcello once said: "Science is the magic that lets us understand
magic."
> Maybe each new hypothesis that scientists generate is initially given an
> evidence of something like -40 dB? And the RPOP should do the same? I
> suppose the initial value has evolved over time, and could be rising?
1) Having a clearly defined initial credence of -40 dB followed by
proper integration of evidence, would be a *vast* improvement over the
current situation in science. The problem here is not prior
probabilities; those wash out. The problem is selective attendance to
evidence, failure to integrate consistently, and subjective parameters
in informal models.
2) The RPOP should not do the same. You can't use ad-hoc methods in an
SI, not if you want to live. There are principled methods for dealing
with this problem such as Kolmogorov complexity and Vapnik-Chervonenkis
dimension; neither of these quite suits my purposes, but both are a vast
improvement over a human programmer sitting down, thinking a bit, and
then inputting a parameter of -40 dB.
-- Eliezer S. Yudkowsky http://intelligence.org/ Research Fellow, Singularity Institute for Artificial Intelligence
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT