RE: Metarationality (was: JOIN: Alden Streeter)

From: Ben Goertzel (ben@goertzel.org)
Date: Sun Aug 25 2002 - 23:31:46 MDT


> Minds are physical systems.

No, they are most certainly not!

A mind is a fuzzy set of patterns that are

-- emergent in a certain system
-- emergent between a certain system and its environment

The system in question need not be physical, it could be virtual (e.g. a set
of particle-like entities in a computer simulated reality running on your
future PC).

Human minds are currently closely associated with physical systems (human
bodies), and they mostly consist of patterns emergent from and between human
brains. But this does not mean minds are physical systems -- not even human
minds, let alone virtual minds....

> If a mind has a particular U, that U
> must have come from *somewhere*. That particular U has a physical cause;
> something shaped it. Those shapers will be either rational or irrational
> depending on what kind of reliable correspondence they bear to reality.
>
> The Bayesian Probability Theorem is inescapable - it can be applied to
> *every* causal system that exists in our universe to determine whether a
> given pattern is likely to correlate to any other causes or effects that
> exist in outside reality. *Any* causal system, *any* physical process,
> not just the ones that we usually consider as "rational" or even
> "mindful".

Yes, Bayes' Theorem can be applied in any circumstance, just as can, for
instance the theorem that "1+1=2".

But the theoreom "1+1=2" does not tell us how to identify whole entities (to
count as "1" ... is a person a single entity or a conglomerate, etc.), nor
how to map a complex entity like a planet into a numerical vector or tensor.

Similarly, Bayes' Theorem and associated probabilistic results do not tell
us how to construct the universal set U they need in order to be applied.

Mathematical theorems are all universally applicable and inescapable
(assuming one accepts the axiom systems from which they're derived!). This
is just as true of Bayes' Theorem as of any other mathematical theorem.

I don't disagree on this, but I don't see how it implies all that you say it
does.

The real question is whether a given math theorem's application in practice
tells us anything interesting.

I am not seeing how the observation that Bayes' Theorem applies to the
universe tells us anything interesting.

I think that Bayes' Theorem applies to the mind as a whole tells us
something sort of interesting. For instance, Solomonoff Induction, one of
my favorite abstract approaches to intelligence, is closely tied to Bayesian
learning, with a special prior distribution based on algorithmic
information. Great stuff! However, it doesn't do you much good in terms of
practical AI design, nor in terms of the analysis of human psychology. In
these cases it seems to me more useful to think of explicit Bayesian
probabilistic inference as one aspect of intelligence among many -- and to
effectively ignore the fact that, yes, it is possible to model other aspects
of the mind using Bayes' Theorem as well, in various senses.

> If your U is a physical thing, it exists for physical reasons,
> and the BPT
> will govern whether those reasons are such that U is likely to bear any
> given kind of correlation to external reality.

A mathematical set like the universal set U in probability theory is NOT a
physical thing, any more than a mind is.

A set like U may be a pattern in a certain mind, that this mind "acts as if"
it were carrying out probabilistic inference with respect to a certain
universal set U.

In this case, the set U is a part of the pattern-set that is the system's
mind.

>
> You cannot flee from the Tao; you cannot run from the laws of physics;
> they govern the very process of your flight...

heh...

Baby, I can flee from the Tao if I want to!

I just don't feel like it tonight...

I feel like slinking upstairs, resting my head on my wife's lap, and
watching the Amazing Wildlife of the Serengeti on the Discovery Channel...

See ya later!!

Ben



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