stevet (12/09/82)
If we approach the idea of "intelligence" with purely mechanical
notions of explanation, then, of course, we find it quite natural
to speak of "machine intelligence." Given the assumptions
prevalent in all those sciences which have chosen to imitate the
methods of physics, it almost becomes a matter of definition that
machines can, in principle, exhibit what we call
"intelligence." And, so long as our ends are wholly mechanical,
it will prove quite fruitful to employ models which reduce
intelligence to its mechanical dimensions. There is nothing
illegitimate about this. It is worthwhile, however, to keep in
mind the connection between our philosophical conclusions and
our initial assumptions. I have not observed that a great deal of
meaningful discussion can occur between persons whose thoughts on
these issues diverge at the starting point.
I find it intriguing to note the ease with which we can detach
the idea of intelligence from that of consciousness. Our
awareness of the "interior" world of consciousness seems to be
dimming--at least it seems so if we can take as more than
intellectual play the whole drift of conventional scientific thought.
The awareness I speak of implies a distinction between the
brain as a physical, observed object (lying on the operating table over
there) and my (brain-mediated) BEHOLDING of that brain.
One would think that, given a universe in which this "beholding"
is the rock-bottom fact from which the very possibility of
"explanation" arises, any models of explanation would reckon
with it at a rather fundamental level. Ironically,
it is by all accounts the discipline of physics which is moving
in this direction, and away from mechanical models. It is the
"softer" sciences, thinking to imitate physics, which are rushing
to deny the category of consciousness altogether. (Reductionist
models of consciousness easily become simple denials of it. They
tend to explain various correlates of the beholding, while leaving
out of account the beholding itself.)
In sum, it is perfectly legitimate to develop, and implement on
machines, our purely mechanical models of such human functions as
intelligence. But the interesting question is why we feel so
driven to do this with the conventional biases--that is, to
develop models of human functions which not only leave out, but
positively deny, what cannot be modeled upon mechanical and
reductionist principles.
Steve Talbott
UUCP: ...!{ucbvax or decvax}!teklabs!tekmdp!stevet
CSNET: tekmdp!stevet@tektronix
ARPA: tekmdp!stevet.tektronix@rand-relay
TEKTRONIX, Inc.
Design Automation Division
PO Box 4600, MS 92-551
Beaverton, Oregon 97075
503-629-1670