Path: utzoo!attcan!uunet!snorkelwacker!apple!amdahl!kp
From: kp@uts.amdahl.com (Ken Presting)
Newsgroups: comp.ai
Subject: Re: Why the Chinese Room doesn't convince
Summary: Inferring meaning from use does not distinguish ismorphic models
Keywords: Searle, Chinese Room, semantics, Quine
Message-ID: <85g302iO94GB01@amdahl.uts.amdahl.com>
Date: 21 Mar 90 19:46:07 GMT
References: <2080@skye.ed.ac.uk> <352@ntpdvp1.UUCP>
Reply-To: kp@amdahl.uts.amdahl.com (Ken Presting)
Organization: Amdahl Corporation, Sunnyvale CA
Lines: 82

In article <352@ntpdvp1.UUCP> sandyz@ntpdvp1.UUCP (Sandy Zinn) writes:
>In article <2080@skye.ed.ac.uk>, jeff@aiai.ed.ac.uk (Jeff Dalton) writes:
>> If you start with meaningless, uninterpreted symbols it's hard to see
>> how they end up referring to what we think we do.
> 
>Consider Wittgenstein's assertions in this context:  there is no logical
>reference between the formal system (uninterpreted symbols) and "reality".
>But reference (albeit an alogical one) does occur:  we establish reference
>by Usage.  If you and I agree that "cat" refers to a cherry, then "cat"
>attains meaning.

Quine's "indeterminacy of translation" prevents us from getting a firm
grip on the denotations of each other's terms.  Cats and cherries are not
too hard to distinguish, but suppose you say "Cat" and point to Felix.
Quine's point is that Felix is identifiable both as a cat, and as a set
of connected cat parts, and a succession of cat time-slices, and an
unlimited number of other descriptive expressions.  So any conscious
logical attempt to align our terminology will stop short of definitively
pairing a unique expression in my vocabulary with a term from yours.

>                  As the Witt said, "It's a question of who's the Master,
>the Man or the Words"

This sounds like something Humpty Dumpty would say ...

>Verbal symbols are interpreted according to past associations, which is
>to say they achieve meaning by usage.  It is the SHARING of associations
>which makes human communication possible.  These shared associations are
>the context for the verbal symbol.  Within that context, the symbol has
>meaning, i.e., it has usage.  In a different context, the symbol may have
>no meaning, i.e., no usage, or a different meaning.

What has always bothered me about Wittgenstein's appeal to usage is that
it seems to stop with the linguistic behavior of illiterates.  Ong, in
_Orality and Literacy_, cited an example from Luria's research which
showed how much trouble the illiterate peasants had with *definitions*.
Breifly, illiterates had no idea why anybody would care about the
definition of words, even practical words like "car" and "bus".  For the
illiterate peasants, words do seem to function according to W's theory,
but there is a lot more to language than just evoking each other's
associations.

That must certainly be the *foundation* of human language behavior.  But
there is a large superstructure of abstractions built on top of those
associations.  Nobody has ever seen a perfect right angle, but the
abstract concept is no problem for most of us.  Right angles are things
that we can *imagine*, but never experience.  Recall Quine's point, and
add the uncertainty we all have regarding each other's fantasy life, and
the indeterminacy of reference for abstract terms seems to me much worse
than for practical terms.

But it is much easier to reach agreement on mathematical issues than on
practical issues.  I conclude from considerations such as these that
reference has less to do with meaning than Quine or Putnam (or Russell,
for that matter) would have us believe.  I would give up denotational
semantics altogether, if it weren't for Tarski's solution to the Liar
paradox.  I don't think there's a non-denotational solution to that.

>A lot of effort has to be expended (as we see in our postings here) to
>correlate our associations and therefore our interpretations of words.
>Just think of the long and tortuous thread on "Intelligence".

Most of the torture here is based on formal maneuvers and disagreements
about the meaning of abstract terms.  The discussion of syntax vs.
semantics, and emergence are good examples.

I suppose we all here are ultimately interested in the practical matter of
how to make machines.  I think the problems are in the relations between
abstractions and practical reality, which are very confusing.

>
>Consider my pet notion:  we are most likely to achieve "Artificial
>Intelligence" by focusing on getting computers to negotiate relationships
>with one another.

This is a good suggestion.  Negotiations involve some very abstract
notions, such as "fairness" (which is *never* observed in nature!).  And
negotiations have very practical purposes and consequences.  Hume thought
that language evolved when humans needed to make promises to each other.
This can't be the whole story, but I think it's a suggestive idea.

Ken Presting