From: Ben Goertzel (ben@webmind.com)
Date: Mon Mar 05 2001 - 07:12:06 MST
> So a priority for me is to see what concepts like
> seed AI, self-enhancement, and Friendly AI would
> look like given the philosophy of mind sketched
> in the last paragraph. [a quantum-computing-based philosophy of mind]
I certainly wouldn't want to dissuade anyone from thinking about
something as fascinating as this -- not in a world where most people
spend their time watching sit-coms and reading People Magazine...
But still, from my point of view, there's something a bit funny in
worrying about the implications for "real AI" of something as poorly
understood
as quantum computing. I guess my problem is this: real AI is badly
understood
and quantum computing is badly understood... so by putting them together
what
are you going to get? Either the two will resolve each others'
uncertainties
(not likely in my view) or you'll just get a doubly fuzzy mess (likely in
my view)
Personally I'm more inclined to work out real AI in as much detail as
possible
(not just theorizing but trying to build the damn thing), and to try to
understand
what the fuck is going on with quantum computing (which Mitchell is doing
quite nicely
in his own research work, although this work is clearly at a very early
stage).
Having said all that, I do have a somewhat more productive comment as well
;>
A lot of what's unique in the Webmind AI Engine is the fusion of techniques
we call "reasonable" and techniques we call "intuitive"
Roughly speaking, reasonable techniques are incremental. They combine two
pieces of
knowledge to form a third piece of knowledge. Different ways of doing this
constitute
different reasoning rules (induction, abduction, deduction, implication,
unification, etc.).
Intuitive techniques on the other hand are holistic. A node in Webmind can
form a halo, meaning
that it uses NN-like activation spreading techniques to suss out what other
things in Webmind's mind
are "in some way related to it", without necessarily figuring out what the
relationship is...
In the system's dynamics, intuitive techniques guide reasonable techniques.
The system reasons in the
direction that its intuition has guided it. Right now we have tested this
only in the context of first-order
inference, over the next 6 months the higher-order inference code we've
written will be integrated with intuitive
processes in this way.
If I had to make a guess about where quantum computing would help a mind,
then, it would be this:
FASTER AND BETTER INTUITIVE PROCESSING. If intuitive processsing involves
an element of the mind integrating
into itself a rough understanding of how it relates to the rest of the
mind -- this is the sort of thing quantum
nonlocality would presumably help out a lot with.
Self-enhancement on the part of AI, in my view, will just be an advanced
reasoning process which conducted
a lot like other advanced reasoning processes. The intuitive part of it
will come from introspection, which
may be enhanced by quantum nonlocality. The reasonable part of it will be
good old higher-order logic, which
due to its discrete and precise nature is fairly unlikely to be hugely
assisted by quantum phenomena (at a guess).
Finally I'll reiterate from previous messages my basic attitude towards
quantum computing and the mind.
My guess is that quantum computing is not necessary for consciousness and
real AI. It surely would be a useful
tool for mind-building, but is probably not a necessary one. Whether the
brain uses this particular tool or not is
unknown, and I don't have a strong intuition about it.
Ben
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT