From newshub.ccs.yorku.ca!torn!cs.utexas.edu!uunet!psinntp!dg-rtp!sheol!throopw Wed Sep 23 16:54:35 EDT 1992
Article 6992 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!torn!cs.utexas.edu!uunet!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: what are relevent aspects of performance?
Message-ID: <716954131@sheol.UUCP>
Date: 20 Sep 92 02:53:10 GMT
References: <1992Sep5.023018.23734@news.media.mit.edu> <1992Sep6.195000.3465@Princeton.EDU>
Lines: 89

: From: minsky@media.mit.edu (Marvin Minsky)
:: throopw@sheol.UUCP (Wayne Throop)
:: And, since no artificial so-called "flying machine" has really flapped
:: its wings as the primary method of propulsion, they are all
:: distinguishable from natural flyers, [...]
: Gosh, I thought everyone on the planet had by now heard of Paul MacReady's
: successful construction and flight of a genuine ornithopter.  Three or
: four yars ago, at least.  Flapped its wings fine.  Look like a
: pterodactly, pretty much.

Yes, but the one I saw had to be towed aloft.  It could not initiate
flight independently.  It also had problems similar to those trying to
get quadruped robots to work as I recall, in that the flight control
software could only deal with very limited conditions, and was easily
provoked into various horrible failure modes.  Nowhere near as robust as
a practical flyer would need to be.  If MacReady has done better than I
last knew, I'd be delighted to hear of it. 

But even given wing-flapping, it was still distinguishable from natural
flyers, since it didn't eat and excrete, and its tendons were metalic,
and motive power wasn't contractile material driven by ATP cycle and so
on and on. 

Which leads to the next point.

: From: harnad@phoenix.Princeton.EDU (Stevan Harnad)
: Message-ID: <1992Sep6.195000.3465@Princeton.EDU>
: Unfortunately, this misses the point of my suggestion not to try to
: define intelligence but to concentrate instead on generating
: Turing-indistinguishable performance capacity. 

I don't think so.  My implied point was that unless one specifies just
what counts as "[...]indistinguishable performance", then the task of
"generating [...]indistinguishable performance capacity" is extremely
ill-defined, to the point of uselessness. 

Obviously, (at least, it seems obvious to me, even tautological) without
limiting what counts as part of the "performance", it is *impossible* to
be a non-X with "indistinguishable performance" to an X.  After all,
with no limit as what is part of the performance, this becomes just a
"non-X indistinguishable from an X" and is ill-formed, or at least
useless. 

Now, if by "Turing-indistinguishable" is meant Turing's notions of what
belongs in the performance vs what does not (that is, roughly, the
performance is the generation of a "meaningful" conversation with
strings of symbols), then we're getting somewhere.  But without the
notion of what is part of the performance, advice to "concentrate
instead on generating" capacity for such a performance is puzzling at
best. 

: From: harnad@phoenix.Princeton.EDU (Stevan Harnad)
: Message-ID: <1992Sep6.195000.3465@Princeton.EDU>
: Of course if
: (1) there were nothing special about mental states, or
: (2) performance capacity were identical with mental states, or
: (3) performance capacity picked out an interesting natural category
:     of its own (say, staying "cognitively aloft," which would be
:     analogous to the common property of both biological and artificial
:     flight)
: then Wayne's implied objection would be well taken.  But none of the
: above is the case. 

Huh? I'd say all three of the above are (at least possibly) the case. 
Depends on what is meant by "special about mental states" I suppose. 
But regardless of that, my objection certainly doesn't depend on any of
the three.  I'm merely pointing out that one can't even *try* to
generate a capacity for a performance without knowing what performance
one is talking about. 

: From: harnad@phoenix.Princeton.EDU (Stevan Harnad)
: Message-ID: <1992Sep6.195000.3465@Princeton.EDU>
: So with flight, you can shoot for flight,
: simpliciter, i.e., anything that gives a system the performance
: capacity to get and stay aloft on its own (excluding the obvious forms
: of cheating, like rocketing, orbiting, etc.). Or you can shoot for
: biological flight, in which case you WOULD have to capture all the
: relevant performance capacities of organisms (including flapping).

My point is reflected in the "relevant performance" phrase.  Stevan has
never specified what's relevant, nor adequately motivated why Turing's
notion of what's relevant for being "indistinguishable" is wrong-headed. 

Which leads into the "grounding" thread, of course (which is indeed
motivation for why Turing's notion is wrong-headed... I just don't
think it's *adequate* motivation, because Stevan's account of why
humans *are* grounded and computers *aren't* seems incoherent to me.)
--
Wayne Throop  ...!mcnc!dg-rtp!sheol!throopw


