From: Jeff Medina (analyticphilosophy@gmail.com)
Date: Tue Oct 19 2004 - 16:24:16 MDT
Eliezer: "Lest any despair of AI, remember that humans can think about
this stuff
without going up in flames, ergo it must be possible somehow."
This implies it must be possible somehow for AI to think about this
stuff the way we do.
We haven't figured out how a self-rewriting AI might be able to prove
Theorem Prover X is an improvement to its current theorem prover. So
it may well not be possible for a self-rewriting system to improve
upon its own theorem prover.
Hence, how does this comment save one from despairing of AI, at least
as far as one had one's heart set on AI being capable of improving on
its own theorem prover?
I don't happen to think any of us should despair, given that
improvements in human reasoning almost never involve mathematical
proof that the new method of belief-acquisition is an improvement.
Reliance on proof-theoretic deduction alone as a method of
self-improvement will leave us toeing the waves on a rock near
Descartes's. AI shouldn't be restricted to such a limited method of
self-rewriting.
This archive was generated by hypermail 2.1.5 : Tue Feb 21 2006 - 04:22:47 MST