Re: Metarationality (was: JOIN: Alden Streeter)

From: Samantha Atkins (
Date: Thu Aug 29 2002 - 17:25:13 MDT

Eliezer S. Yudkowsky wrote:

> They are unconscious, perhaps. But here I think you are using
> "irrational" simply to mean "nondeliberative", when there is not
> necessarily anything about it that is incomprehensible to a Bayesian
> reasoner. As I said before, if you lump all nondeliberative processes
> together with warped deliberative processes as "irrationality", then you
> lose your ability to distinguish the true path in deliberation and
> intuition. Philip K. Dick's thought processes are unconscious and they
> may, perhaps, involve the creation of deliberative thoughts that would,
> interpreted purely as deliberative thoughts, be irrational. But if the
> system as a whole works to produce nonaccidental successes, there is a
> rational reason behind it. What, specifically, is happening here that
> is outside the domain of evolved intuitions and trained intuitions?
> What is happening here that is outside the domain of genetic learning
> and neural learning of regularities in reality? Where does the
> irrationality come in?

I don't know. But if you were in the midst of one of these
experiences I think you would laugh yourself silly attempting to
see it as somehow "Bayesian". :-)

> No, some sorts of creative achievement seem to go along with rampant
> internal nondeliberation, perhaps even nondeliberative activities that
> create warped deliberative thought processes as side effects. It
> doesn't mean that the nonaccidental successes are rationally
> incomprehensible - that it is not possible for a Bayesian observer to
> understand how they work, and why they work, and conceivably employ a
> process of that type deliberately in order to arrive at the same answer.

OK. That is better. Yes, sometimes some of it can be
"stepped-down" to the level of a Bayesian observer. But I don't
think all of it can and I do not believe the parts that can't
are thereby without value.

- samantha

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:40 MDT