Ryan Vetter (physicalconstants@yahoo.ca) wrote:
> Well, the problem is isolated to computers having trouble with
> semantics: they really don't understand anything at all, they just
> "compute" (Searle's Chinese Room analogy/argument).
The Chinese Room argument (which can be traced back to Leibniz' equally
inconclusive Mill analogy) doesn't prove anything except several
misunderstandings about computer science in general and AI in
particular. And it is not like Searle's position attributing
intentionality to some mysterious "causal properties" of the brain would
answer any questions -- just more handwaving, basically.
> This, of course, adversely affects an AI driven computer's ability to
> predict things, and respond meaningfully. At best a computer operates
> in a very rudimentary ad hoc fashion.
This is a different issue. The problem with traditional approaches to AI
is that they try to solve the most complex problems first. Instead of
following the path of evolution, starting out with simple systems and
building upon these achievements, AI researchers wanted to tackle tasks
like natural language understanding or planning. In a way, they
succeeded, but only within extremely confined micro-worlds -- thus the
brittleness of AI systems.
On the other hand, the Subsumption Architecture developed by Rodney
Brooks et al. suggested a different route: building very simple robots
with simple, but robust behaviors, then layering progressively more
complex behaviors on top, retaining the robustness due to the lower
layers. Robots built according to the Subsumption Architecture act in
real-time and they don't just bump into unexpected obstacles; they are
dealing gracefully with anything out of the ordinary. But they don't
speak or understand English, and neither can they help with medical
diagnosis etc.. Not yet anyway. But AI is hard, and spectacular
successes are not to be expected just around the corner.
> I don't have the knowledge of different programming
> languages, but I wonder what languages would be able to recreate some of
> what the KN does.
The programming language wouldn't matter that much. I would choose LISP
(which is certainly better suited than some other languages), but that's
just a personal preference.
- Michael
Michael J. Hußmann
E-mail: michael@michael-hussmann.de
WWW (personal): http://michael-hussmann.de
WWW (professional): http://digicam-experts.de
====================================================================
The NewtonTalk Mailing List - http://www.newtontalk.net/
The Official Newton FAQ - http://www.splorp.com/newton/faq/
The Newton Glossary - http://www.splorp.com/newton/glossary/
WikiWikiNewt - http://tools.unna.org/wikiwikinewt/
====================================================================
Received on Sat Feb 21 10:23:28 2009
This archive was generated by hypermail 2.1.8 : Sat Feb 21 2009 - 15:30:00 EST