From: Josh Cryer (firstname.lastname@example.org)
Date: Tue Jun 29 2004 - 20:20:12 MDT
Long time reader (off again on again), just wanted to post about this.
I do not think a transhuman superintelligent mind could necessarily
*know* it was inside a box. You invent a universe for it, you plop it
in there, volia.
You can still get useful stuff out of that universe by observing it
and essentially reverse engineering the things the AI did to meet its
technological achievements. If the AI invents some super matter
discombobulator that can figure out things about its underlying
universe, you could reasonably create a similar machine in the
physical universe. And for those who would think that the technology
an AI would create would be too complex for a mere human mind (which I
don't think is possible, personally), we could simply coerce someone
within its universe to ask it.
We just say "hello verse AI, ask the super-AI what's going on." How
could we reasonably believe that a superintelligent AI would believe
some crackpot human talking about the "higher verse," etc? (This is
assuming we even talk about a higher verse! We could simply manipulate
these verse AI's into doing anything we wanted, disguise it as genetic
coding and the super-AI wouldn't know squat and believe that Genetic
Type Z's are Curious and Ask Lots of Technological Questions.)
Of course, a transhuman super-AI could quite easily, on a whim, tell
the humans to build some technology they don't understand, but I think
that since we're taking this box AI approach, by our very nature we
would only work on technologies we do understand. So an AI couldn't go
(for some metaphysical reason) "build this fancy thing which is in
reality a copy of me" and actually be successful.
So, I would take Eliezer's little bet, but I will admit that I
wouldn't even talk to him as someone outside the box throughout the
whole chat. :) The whole premise is based on the assumption that a
super-AI could even know it was inside a box, while I think a
sufficiently built box makes that impossible.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:47 MDT