From: Charles D Hixson (charleshixsn@earthlink.net)
Date: Tue Aug 22 2006 - 19:23:35 MDT
Michael Vassar wrote:
> ...
> Well yes, but the sorts of advantage that an AI, even one that was not
> a superintelligence, would likely have relative to a human don't
> necessarily closely resemble those a smarter human would have unless
> the AI is
a recent
> upload, or at most a neuromorphic AI. IQ metaphors can be
> mis-leading. After all, much less intelligent parents routinely
> manage to keep control over their more intelligent children by virtue
> of superior knowledge, physical power, social support, etc, for quite
> a while, a situation entirely unlike that with a UFAI.
I would expect that a designed AI would have an "direct perception" of
the ratios of large numbers and "lightening calculator" features. Also
various other equivalent modules. I'm less certain that it would have
an intuitive recognition of human faces. It might...if the builder
chose for it to. (I can see advantages as well as disadvantages.) I
would expect an AI to have "direct perception" of well formatted
URLs...possibly including an automatic check to see whether they were
live. (This though, isn't certain as it speaks to the environment in
which it is presumed the AI will live.) I would be surprised if it
could easily learn to run and catch a ball.
I don't really find IQ to be a useful concept. I see intelligence as a
coordination between many specialized modules. I suspect that we aren't
yet aware of most important modules that go into "being human", but
certainly many of them have to do with maleness or femaleness. I don't
think that an AI will implement very much of that set of modules. OTOH,
it will need to understand easily many things that have no relation to
it's own lifestyle if it is to communicate easily with people, but most
of these will probably be handled with a shallow mapping to "stub
routines". The purpose will be to allow it to respond appropriately, or
at least sensibly, when someone anthropomorphizes it.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT