From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rutgers!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!caen!nic.umass.edu!dime!orourke Tue Mar 24 09:56:36 EST 1992
Article 4534 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rutgers!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!caen!nic.umass.edu!dime!orourke
>From: orourke@unix1.cs.umass.edu (Joseph O'Rourke)
Newsgroups: comp.ai.philosophy
Subject: Re: The Systems Reply I
Keywords: syntax vs. semantics
Message-ID: <45020@dime.cs.umass.edu>
Date: 18 Mar 92 03:51:33 GMT
References: <6374@skye.ed.ac.uk> <1992Mar11.201637.21875@psych.toronto.edu> <44765@dime.cs.umass.edu> <6422@skye.ed.ac.uk>
Sender: news@dime.cs.umass.edu
Reply-To: orourke@sophia.smith.edu (Joseph O'Rourke)
Organization: Smith College, Northampton, MA, US
Lines: 62

In article <6422@skye.ed.ac.uk> jeff@aiai.ed.ac.uk (Jeff Dalton) writes:
 >In article <44765@dime.cs.umass.edu> 
  orourke@sophia.smith.edu (Joseph O'Rourke) writes:

    >>	It seems to me that the way in which a program manipulates
    >>its symbols shows that it has attached some type of meaning to them:
    >>
    >>(a) As a crude example, if a program passes a double to a function to
    >>    compute the arctangent, it "knows" in some primitive sense that
    >>    the bits it is moving around represent a real number, and that
    >>    the library arctangent function expects such.  [...]

 >I have a suspicion that this is just our old dispute about whether
 >there can be different, equally good interpretations of, say, the
 >inputs and outputs to the Chinese Room.

Yes, but it may be worthwhile to renew the debate in another context.

 >Suppose I represent floats as strings of letters (in base 26, say)
 >and do adds and subtracts on them.  In some sense the machine "knows"
 >these strings represent real numbers?  Well, maybe so.  But why is
 >that sense a relevant one?

I probably shouldn't have said the machine "knows" anything, even in
scare quotes.  The point is that the machine manipulates some bits/strings/
symbols in a manner differently from other symbols.  It partitions the
universe of symbols into primitive classes: these strings are appropriate
to be passed to this function, and these strings are not.  And it uses
the symbols in a manner consistent with their meaning (to us).  Surely
the machine doesn't "know" that these bits represent a "real number,"
in the sense that we know real numbers.  But it *is* making distinctions
among its symbols, and using them appropriately.
	You said: "I still haven't seen a satisfactory answer to the point 
that the Room manipulates meaningless symbols ([...]) without any way to 
attach meaning to them." What I am claiming is that the symbols cannot 
be *completely* meaningless; for if they were, they would be manipulated 
without distinction.  Since the machine is discriminating amongst its 
symbols, they must have some meaning to it, the machine. 
	And furthermore, it is quite possible for the machine itself to 
have made this discrimination: it doesn't always have to come from the 
programmers.  The machine can create a function (e.g. "is-prime") and
learn empirically to discriminate between prime and composite integers
(although it would not thereby understand fully our notion of "prime").
In this sense, it comes to discriminate amongst its symbols, and
therefore to attach a primitive meaning.
	If you feel such discrimination is not a type of primitive
meaning, perhaps you should sketch the key requirements of what constitutes
a meaningful symbol in your theory of meaning.

 >[...]Now, the Geometry Room, for example, can
 >answer questions about geometry (even if the person in the Room hasn't
 >a clue).  But is that because this system understands geometry or
 >because the programmers (or the mathematicians they consulted)
 >understand geometry?

A good portion of what we learn is taught to us.  Do I understand
geometry because my teachers did?  Well, yes, in part.  But I have
demonstrated mastery by conjecturing and proving theorems myself, 
going beyond direct teaching.  We should apply similar criteria to a 
machine.  And such capability has already been demonstrated to some 
degree, e.g., with Lenat's AM program for making conjectures in 
elementary mathematics.


