From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!wupost!darwin.sura.net!gatech!rutgers!banana!mips!smsc.sony.com!markc Fri Jan 31 10:26:46 EST 1992
Article 3242 of comp.ai.philosophy:
Xref: newshub.ccs.yorku.ca comp.ai.philosophy:3242 sci.philosophy.tech:1987
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!cs.utexas.edu!wupost!darwin.sura.net!gatech!rutgers!banana!mips!smsc.sony.com!markc
>From: markc@smsc.sony.com (Mark Corscadden)
Newsgroups: comp.ai.philosophy,sci.philosophy.tech
Subject: Re: Is understanding algorithmic?
Message-ID: <1992Jan29.004154.6552@smsc.sony.com>
Date: 29 Jan 92 00:41:54 GMT
References: <DIRISH.92Jan18155827@jeeves.math.utah.edu> <1992Jan26.010642.24883@smsc.sony.com> <1992Jan26.014607.8073@husc3.harvard.edu>
Organization: Sony Microsystems Corp, San Jose, CA
Lines: 112

In article <1992Jan26.014607.8073@husc3.harvard.edu> zeleny@zariski.harvard.edu (Mikhail Zeleny) writes:
> In article <1992Jan26.010642.24883@smsc.sony.com> 
> markc@smsc.sony.com (Mark Corscadden) writes:
> 
> MC> In article <DIRISH.92Jan18155827@jeeves.math.utah.edu> 
> MC> dirish@math.utah.edu (Dudley Irish) writes:
> MC> DI>... attempt to convince us (I have admitted it now, I am one of those
> MC> DI>rabid skeptics) that a Turing machine can refer.
> MC> DI>Dudley Irish
> 
> MC> Well, I don't know whether I can convince you, but I can give you the
> MC> following scenario in which two machines use a symbol to refer to an
> MC> object in the real world.
> MC> [ description of two simple machines which use an O/X symbol ]
> MC> [ drawn in the dirt to refer to one of two fixed locations   ]
> MC> [                     *deleted*                              ]
> MC> ...  Analogously, I'd like to suggest that
> MC> someone who believes it possible to prove that Turing machines cannot
> MC> refer help me to apply such a proof, step by step, to the scenario above.
> 
> It's a pleasure.  Your machines merely succeed in matching an internal
> representation of the laboratory location, as well as of some objects
> likely to be found there with a preprogrammed description, perhaps through
> the use of a visual pattern matching algorithm.

By "preprogrammed description" I assume you are referring to the X/O
symbol drawn in the dirt; please correct me if I'm wrong.  The sum
total of what you are saying above seems to be:

1)  Each machine has an internal representation of an elementary
    aspect of the external world.

2)  Part of each machine's programming is to match this internal
    representation against the X/O symbol.

If I understand you correctly, what you say is accurate and I agree
completely with it.  However you are simply restating part of the
conditions for this scenario, not adding anything new.

Can you give an *explicit* description of how the facts you single out
above demonstrate that the machines are not using the X/O symbol, drawn
in the dirt, to refer to one of two locations within the lab?


> Should you like to argue
> that human referential capabilities could be uncharitably characterized in
> the same fashion, I would like to point out the following facts:

I agree with you completely that it would be quite uncharitable to
characterize human referential capabilities as being anywhere near
as limited as those of the two machines described.  In fact, I designed
the scenario to be as minimalistic as possible.  The machines use only
two symbols, an X and an O drawn in the dirt, to refer to the external
world.  There are only two possible real world states which they are
capable of referring to!

Can you quote for me the specific portion of my original article which
you interpretided as my having argued that such extremely simple machines
somehow characterize the full spectrum of human capabilities?  This is
an astounding misunderstanding, though until I have some idea how you
came by it I won't deny that it may somehow have been my writing which
was at fault.  Would you care to help me out on this point?  What,
specifically, made you think that I presented these two machines as
representing the range of human referential capability?


> (a) It is intuitively evident that we are capable of referring to objects
> with which we have no direct perceptual acquaintance, in particular to
> abstract objects like numbers and sets.

Yes!  I agree completely.  My own belief is that reference to abstract
objects like numbers and sets is an immensely sophisticated and advanced
human capability.  Even reference to such "simple" abstract objects as
the numbers two and three.  However, since I never meant to argue that
the simple machines I presented had capabilities comparable to those
of a person, I don't see how what you state is relevant.  I agree completely
with the things you assert above concerning a person's ability to refer to
a vast diversity of things.  Can you explicitly describe why this implies
that the two machines in the scenario I present cannot be correctly said
to be using the X/O symbol in the dirt to refer to one of two locations
within the lab?

All you seem to have pointed out is that human beings are capable of
referring to things which these two specific machines are not capable
of referring to.  I would add that these two machines are incapable of
referring to *anything* except the two positions mentioned in the scenario.
They are also incapable of making references using *any* symbols other than
the dirt-drawn X or O.

Yes, they are indeed limited!  But having a severly limited set of
capabilities is absolutely distinct from having no capabilities at
all, yes?


> (b) No existing machine is capable of solving the general problem of visual
> pattern recognition, e.g. matching a three-year old infant's capacity to
> recognize a human face.  There seems to be no reason to suppose that this
> problem is effectively computable.

The machines I present certainly cannot match a three-year old infant's
capacity in any area, except perhaps in the one area of using one of the
symbols { "X" , "O" } to refer to positions within a fixed lab setting.

How do you move from the quite valid assertion that the two machines
have immensely limited capabilities of referral to your assertion that
the machines have no capabilities of referral whatsoever?

> : Mikhail Zeleny                                                     :

Mark Corscadden
markc@smsc.sony.com
work: (408)944-4086


