From: Ben Goertzel (firstname.lastname@example.org)
Date: Fri May 03 2002 - 18:56:02 MDT
Max, I think there is some truth to this approach. But it's also true that
the requirements of, say, a standalone NLP system are very different from
the requirements for an NLP component of a general intelligence.
So I don't think that the "piecing together separately functional
subsystems" approach to general intelligence can work. However, I do think
that we can learn a lot from narrow AI systems that is useful in building
general intelligence systems.
For instance Peter Voss's AI system is certainly inspired to some extent by
narrow-focused NN systems, among other things. My system is inspired by a
lot of narrow AI work but incorporates none of it exactly...
From: email@example.com [mailto:firstname.lastname@example.org]On Behalf Of
Sent: Friday, May 03, 2002 3:11 PM
To: email@example.com; Kevin Bermeister; firstname.lastname@example.org
Subject: Top down vs. bottom up AI synthesis, a(n attempt at)
clarification of my previous posts
To me, it seems like the problem of mind building can be viewed from two
perspectives, top down and bottom up, just like nanotech. Top down
approaches consist of trying to form an entire mind structure; e.g. Friendly
AI, Novamente, Adaptive intelligence. Bottom up synthesis represents more
fine grained approaches working on creating a small slice of intelligence in
more limited settings; e.g. cyc, natural language, 3d vision, pattern
recognition, mind mapping, etc. Both approaches are important!
Just because what we all really want is true AI doesnt mean that we should
discount the importance of dumb IA subsystems. What should happen is that
these IA subsystems should be incorporated into the development of AI. They
should be used both as development platforms, as the tools run the
computer(s) that the AI coders use. And, where possible, they should also be
used in the design of the AI itself. These IA systems are like tiny pieces
of a dynamic AI jigsaw puzzle, and as they are found and placed they can
"fill in" much of the gaps in our knowledge about AI, and eventually, in the
AI's knowledge about us. This saves us from having to design an entire AI
mind algorithm from scratch that can fill in all the gaps of it's knowledge
on it's own. As more "pieces" are created the islands of intelligence and
capability will grow larger. All this could be incorporated into the
development of any top down AI system to make development occur faster.
Or, said another way, why not use general top down AI principles to sculpt
and synthesize many of these dumb IA subsystems into the real working AI
that we all want. AI will never be able to help in it's own development
until it becomes aware, but IA should allow us to help in the development of
AI in the same accelerated fasion, because it provides so many of the same
forward feedback processes. Only difference is that right now, humans have
to act as the will and the purpose for the system.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:38 MDT