From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!psych.toronto.edu!michael Tue Mar 24 09:56:00 EST 1992
Article 4480 of comp.ai.philosophy:
Newsgroups: comp.ai.philosophy
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!psych.toronto.edu!michael
>From: michael@psych.toronto.edu (Michael Gemar)
Subject: Re: The Systems Reply I
Organization: Department of Psychology, University of Toronto
References: <BL1p0D.6II@world.std.com> <1992Mar14.182737.15329@psych.toronto.edu> <1992Mar14.213045.21776@mp.cs.niu.edu>
Message-ID: <1992Mar16.224423.29809@psych.toronto.edu>
Date: Mon, 16 Mar 1992 22:44:23 GMT

In article <1992Mar14.213045.21776@mp.cs.niu.edu> rickert@mp.cs.niu.edu (Neil Rickert) writes:
>In article <1992Mar14.182737.15329@psych.toronto.edu> michael@psych.toronto.edu (Michael Gemar) writes:
>>
>>The first premise is simply that syntactic manipulation cannot on its own
>>yield semantics, which has been argued to be an analytic truth by many
>>philosophers who are not involved in the AI debate.
>>
>>The second premise comes from the foundational assumptions of AI, namely,
>>that all relevant human cognitive activity is computable, and can be 
>>replicated by the appropriate functional relations.  These relations are,
>                                                      ^^^^^^^^^^^^^^^^^^^
>>by their nature, at base syntactic.  This premise is adopted by Searle in
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>order to examine the implications of Strong AI.
>
>>The Chinese Room gedanken, as I have argued many times, is not in itself
>>the crucial argument.  It is merely an attempt to demonstrate the truth
>>of the claim that syntactic manipulations can't yield semantics.  Even
>
>  There has been much discussion of the problem that "understanding" is
>not well defined.  But an even more serious problem is that neither
>"syntax" nor "semantics" are well defined.
>
>  The claim "syntax can't yield semantics" is intuitively appealing, but
>this does not make it true, and indeed it is a statement without content
>until we have precise definitions of "syntax" and of "semantics".  In the
>intuitive sense, surely syntax cannot yield semantics.  But I have a bunch
>of compact disks which are encoded by by computers, yet can produce great
>symphonies when inserted in a CD player.  If you wish to say that everything
>a computer can do is merely syntactic manipulation, clearly some of those
>manipulations can yield semantics.

All you have shown with your CD example is that some syntactic manipulations
can be given semantic *interpretations* by humans (I have my doubts about
calling music "semantic,", but I'll let that pass for now).  The data on
your CD could instead by interpreted as machine-code for a computer, or
as pretty pictures viewable by some special machine (heck, some CD's
*do* have digital pictures as *well* as sound encoded on them).  There is
no *privileged* interpretation of the raw data - the bits encoded on the
CD do not *inherently* mean "music" or "machine-code" or "pictures".

There is in general a lot of confusion about this issue.  There is no
doubt that computers can produce output that can be *interpreted*
semantically (even ELIZA's output can be read as English).  The question
is whether such outputs *inherently* have semantics, or alternatively,
whether the computer *means* anything by the outputs. 

> The sentence I have underscored is at the heart of the debate.  If we
>restrict "syntax" to its intuitive sense, that statement is pure nonsense,
>and anybody who claims it to be true does not fully understand the capabilities
>of computers. 

I disagree.  See above.

> If, on the other hand, we treat the underscored statement as
>a definition of "syntax", then this is a much broader view of syntax, and
>one which the CR argument does not touch.  In effect the CR argument uses
>a broad interpretation of syntax to set the scene, but then resorts to the
>much narrower intuitive interpretation to get the contradiction.

Again, I think that this is wrong, and a conflation of derived and
inherent semantics.

> --------
>
> Please understand that I am not trying to initiate a tiresome debate about
>defining "syntax" and "semantics".  Much of the discussion of the CR
>argument is pointless.

Now THAT'S something that I CAN agree with :-)

>
> More to the point:
>
>	There can be no final convincing proof that strong AI is
>	possible until there is an actual implementation.

No.  This is wrong.  An implementation will *not* demonstrate that it has
semantics (or understanding, or qualia, or whatever).  This is *not* a
matter of empirical investigation, but of conceptual analysis. 

>
>	There can be no final convincing proof that strong AI is
>	impossible until all terms are precisely defined.  Roughly
>	speaking, any 'proof of impossibility' must be a purely
>	syntactic proof, and perhaps what the CR argument really
>	does is demonstrate the impossibility of producing
>	such a 'proof of impossibility'.
>
> Until there are final convincing proofs, the discussion amounts mostly to
>an exchange of opinions in which there is no common agreement as to the
>meaning of the terms.

While you may be right, I am convinced that the way in which the terms
are being used is not at all "special" or "non-standard".

- michael




