Re: The problem of cognitive closure

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Fri Mar 16 2001 - 13:08:40 MST


For me, it boils down to (a) Turing computability and (b) the
observability of noncomputability. I myself don't believe in HOPC, so I'm
substituting computability for that; any system or behavior that can be
understood by reference to computable rules. If our world is
noncomputable, I would expect that noncomputability to be, at the least,
observable.

The obvious disaster scenario is that a mind "programmed" to believe in
computability uploads humans who turn out to be non-uploadable. Given
that Godel's Theorem and Turing's exposition of the halting problem are
quite explicable to computable minds, I'd expect any mature AI to be able
to formulate the concepts "computable" and "noncomputable", and to
understand the importance of finally verifying the computability of the
human mind before uploading it into computable substrate. In fact, I'd
expect final verification to turn up in any case, as single neurons and
groups of neurons are completely simulated - or unsuccessfully simulated -
as a test case before going on to the larger mind.

I don't expect noncomputability to be subtle. In particular, any form of
internal mental life that I have - whether I'm monads or quarks or
something else entirely - which can affect these sentences I now type, is
either computable or not explicable by reference to computable forces. If
it's possible to explain away all my philosophizing by reference to
computable forces - or rather, if I'm opened up and nondestructively
scanned and all my philosophizing is the product of clearly *visible* and
computable forces - then I, me, the person now doing this philosophizing,
has been captured in toto. Insofar as monads exert observable forces on
these my sentences, their internal processes have visible effects, and the
monads can be categorized as effable or ineffable accordingly. If the
monads are monolithic objects whose internal states have no visible
effects on the flow of causality, then it is not possible that they
account for my qualia or my beliefs about qualia or these my typed
sentences about qualia.

What I want is some kind of Bayesian-useful account of how a monad-dualist
Chalmersian universe differs from this our Universe, or how a totally
computable universe would differ from this our monad-dualist universe. As
far as I'm concerned, the signature effect of qualia is a civilization in
which - not contrivedly, but naturally and emergently - philosophers arise
who start talking about qualia. If this quality is present in both
universes, then all the parts of myself that I value have been accounted
for. If coming up with a computable account of this Universe requires
deliberate tampering or very wide searches to find equivalent speculations
about qualia, then this Universe is most likely noncomputable. More
likely still is that simulating a single neuron would turn out to be
impossible.

-- -- -- -- --
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 - 04:00:36 MDT