From newshub.ccs.yorku.ca!torn!csd.unb.ca!morgan.ucs.mun.ca!nstn.ns.ca!news.cs.indiana.edu!spool.mu.edu!olivea!uunet!usc!rpi!newsserver.pixel.kodak.com!psinntp!psinntp!dg-rtp!sheol!throopw Thu Oct  8 10:10:28 EDT 1992
Article 7047 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!torn!csd.unb.ca!morgan.ucs.mun.ca!nstn.ns.ca!news.cs.indiana.edu!spool.mu.edu!olivea!uunet!usc!rpi!newsserver.pixel.kodak.com!psinntp!psinntp!dg-rtp!sheol!throopw
>From: throopw@sheol.UUCP (Wayne Throop)
Newsgroups: comp.ai.philosophy
Subject: Re: Don't try to "define" intelligence (or flight)
Summary: still puzzled by some details
Message-ID: <717643608@sheol.UUCP>
Date: 28 Sep 92 04:50:32 GMT
References: <1992Sep5.023018.23734@news.media.mit.edu> <1992Sep6.195000.3465@Princeton.EDU> <716954131@sheol.UUCP> <1992Sep21.141208.18029@Princeton.EDU> <1992Sep19.164232.24652@Princeton.EDU>
Lines: 49

: From: harnad@phoenix.Princeton.EDU (Stevan Harnad)
: Message-ID: <1992Sep21.141208.18029@Princeton.EDU>
: Wayne Throop asks how to distinguish relevant from irrelevant aspects
: of performance for the purposes of TTT-indistinguishability. As a first
: approximation, get into the region worth discussing: A robot that can
: see, hear, manipulate, discriminate, categorize, and discuss the objects
: in the world roughly as we can.

Certainly sensible, but this doesn't seem to fit with earlier advice:

: From: harnad@phoenix.Princeton.EDU (Stevan Harnad)
: Message-ID: <1992Sep19.164232.24652@Princeton.EDU>
: "Intelligence" is an arbitrary word denoting clever performance that
: ordinarily requires "X" (where "X" is the mental state that allows humans
: and animals to do those same clever things).  [...]
: So don't bother trying to define intelligence; it's an empty, arbitrary
: exercise. Focus instead on what it takes to generate the clever performance
: that ordinarily requires X.

The point that doesn't fit for me here is this:  The "clever
performance that ordinarily requires [..mental states that allow humans
and animals to do those same clever things..]" does NOT (to me) seem to
require the "see, hear, manipulate" part of Stevan's TTT criteria.  The
"clever performance" that in humans requires intelligence is (it seems
to me) entirely contained in the "discriminate, categorize, and discuss
the objects in the world". 

It is for this reason that I find Stevan's predeliction for dragging 
seeing, hearing and manipulating into it most puzzling.  The invariants
in human "X-performance" don't seem to me to depend upon them, or even
be very closely related to them.

Another point I find puzzling is the "indistinguishable from human" part
of the TTT...  it seems to me that "clever performance that ordinarily,
if done by a human or animal, would require mental states" would better
serve to cover the ground by itself.  Note that many "clever
performances" by animals do NOT provoke us to suppose mental states,
even if it *would* tend to indicate mental states if a human were doing
it, so the problems of algorithmic behavior are in essence encapsulated
in this simple formulation. 

And finally, the "be strict and then relax the requirements when we know
more" seems pointless, at least in the case of requirements that we
*know* vary among beings considered intelligent.  I see no motive
whatsoever for requiring, say, sight, or two hands and five digits each,
or other "average" human capacities, when it is clear from the outset
that these are not intelligence-covariant. 
--
Wayne Throop  ...!mcnc!dg-rtp!sheol!throopw


