Newsgroups: sci.skeptic,alt.consciousness,comp.ai.philosophy,sci.philosophy.meta,rec.arts.books
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!pipex!uknet!festival!castle.ed.ac.uk!cam
From: cam@castle.ed.ac.uk (Chris Malcolm)
Subject: Re: CR denies computer addition? (was Re: Penrose and Searle)
References: <Czzosp.Br5@cogsci.ed.ac.uk> <D062F0.FCK@festival.ed.ac.uk> <D0EGs0.167@cogsci.ed.ac.uk>
Message-ID: <D0IJ1J.LzL@festival.ed.ac.uk>
Sender: news@festival.ed.ac.uk (remote news read deamon)
Organization: University of Edinburgh
Distribution: inet
Date: Thu, 8 Dec 1994 22:12:55 GMT
Lines: 187
Xref: glinda.oz.cs.cmu.edu sci.skeptic:97503 comp.ai.philosophy:23418 sci.philosophy.meta:15426

In article <D0EGs0.167@cogsci.ed.ac.uk> jeff@aiai.ed.ac.uk (Jeff Dalton) writes:
>In article <D062F0.FCK@festival.ed.ac.uk> cam@castle.ed.ac.uk (Chris Malcolm) writes:
>>In article <Czzosp.Br5@cogsci.ed.ac.uk> jeff@aiai.ed.ac.uk (Jeff Dalton) writes:
>>>In article <Czu5zD.Dto@festival.ed.ac.uk> cam@castle.ed.ac.uk (Chris Malcolm) writes:
>>>>In article <JMC.94Nov22011226@white.wisdom.weizmann.ac.il> jmc@white.wisdom.weizmann.ac.il (McCarthy John) writes:

>>>>>Imagine that the procedure that the man in the Chinese room were performing
>>>>>unbeknownst to him were simulating an abacus rather than translating Chinese.
>>>>>Would the Searle argument then prove that a machine can't do arithmetic?

>>>>Yes. The interesting point is that we are (nearly all of us) happy to
>>>>extend the meaing of "add" to include what calculators do, but not (so
>>>>many of us) to extend "understanding" to computers which get the
>>>>behaviour right.

>>>Bull.

>Chris, you're distorting my point by leaving out the part of your
>article that did the most to get me to say "Bull". 

Sorry Jeff. You were a bit vague.

>Here it is back:

>  "Adding" is a low level menial activity that we are quite happy to
>  consign to serfs such as clerks and calculators. On the other hand,
>  "understanding" is something we all feel rather proud of, one of our
>  unique distinguishing characteristics, so we are more defensive about
>  it.

Ah, so this is what you think is "bull"? This may suprise you, but I
posted it because I thought it was true. I'd hate to waste your time
by defending the wrong thing again, so can I ask you to elaborate?

>>There is a strong interpretation of "understand", adopted by Searle,
>>which necessarily includes the subjective feel of understanding. The
>>CR argues that a computer running a program wouldn't have this. There
>>is a weak interpretation of "understand", in which the possession of
>>understanding is certified by the right behaviour. This is often (not
>>quite correctly) considered to the Turing Test view. Searle denies
>>that this is a valid interpretation of "understand". There is a
>>middling interpretation which views understanding as getting the
>>behaviour right by means of the right kind of internal machinery. This
>>is (for example) the interpretation supposed by Brian Smith's
>>Knowledge Representation Hypothesis.

>I don't think this all turns on how we interpret "understand".  Let's
>use 3 separate terms U-1 (Searle's), U-2 (B Smith's), and U-3 (Turing
>Test).  No matter which camp gets to own "understand", we can ask how
>various entities fare w.r.t. the several U-i.

Exactly, and I think that when we do this, we will find that a lot --
not all but a lot -- of the published arguments under debate here will
evaporate into clear cases of changing interpretations in mid-stream,
misunderstandings, and crossed purposes. For example, Searle argues
that computers won't be able to understand1, whereas the Systems Reply
(most versions) argues that computers will be able to understand2. So
there is no dispute going on there. The Systems Repliers haven't
addressed Searle's argument, and Searle doesn't counter the Systems
Reply, but an "understanding1" re-interpretation of it.

>>There are three corresponding interpretations of "semantics",
>>"intentionality", and so on.

>Can you provide a bit more description of the versions of semantics?

Your own cartoon (Searle, Smith, and Turing) is good. S1 asserts that
it depends on a clear apprehension of the meaning in a conscious mind.
S2 that it depends on the right behaviour (apparently knowledgeable
behaviour) being produced by the right machinery (the represented
knowledge and its associated access and inference mechanisms). S3
asserts that if the behaviour looks knowledgeable then it is, even if
there is no "knowing" machinery behind it; a sophisticated variant
asserts that the knowledge and reasoning we think we employ as part of
our decision-making processes are in fact a rational reconstruction
for social purposes.

>>Although Searle originally intended his CR argument to show that
>>running the right program couldn't cause strong understanding, in
>>counter to Schank's claim that script systems displayed a minimal form
>>of understanding, it is not clear that Schank meant to claim strong
>>understanding in the first place. Searle offers as a corollary of the
>>CR argument that you can't get semantics from (nothing but) syntax. Of
>>course all that the CR justifies is the denial of strong semantics.

>Do you think it justifies that?  That it works in that case?

Yes. I don't think a Schankian script discussing hamburger joints has
any conscious awareness, and I don't think that any further
elaboration of that kind of machinery (more knowledge, bigger faster
computers) will provide any glimmer of consciousness. On the other
hand, I don't think this is in the slightest suprising, since it is
(IMHO) one of the fundamental if inexplicit assumptions of symbolic
functionalism and GOFAI that you don't need consciousness (or
emotions) in order to solve puzzles. It's not there because they
didn't program it in, and they didn't try to program it in because
they didn't think they needed to. It is true that some naive optimists
did hope that it might emerge spontaneously, and some of the more
cantankerous among them argue "well you can't prove it isn't
conscious, can you?", but Occam's Razor deals with that.

>>Similarly there is a strong, middling, and weak interpretation of
>>addition. My point was that many people who take the strong
>>interpretation of understanding are happy to adopt the weak
>>interpretation of addition, in which case there is no problem with
>>computers adding, but there is one with computers understanding (by
>>means of running programs). But if one adopts the strong intepretation
>>of addition, then a suitably modified CR proves that running the right
>>program won't do it.
>
>Well, there I disagree.  It can be argued that computers don't
>perform additions.  They work with numerals or perhaps with 
>"uninterpreted symbols" or something, whereas proper addition
>involves numbers.  There are, presumably, argumants along those
>lines.  But I don't agree that modified versions of the CR
>show any such thing.

When you say "there are, presumably, arguments along those lines"
that's exactly what I mean by "modified versions of the CR". For
example, the punchline might be "but see, now we look inside the
calculator and all we see is Searle rattling the beads back and forth
on the abacus according to the rule-book. He has no idea whether you
are interpreting the results as a square root or as the next move in a
game of NIM, and being himself totally innumerate, he will never even
be able to make an educated guess about what is going on. But since he
is the only thing inside the calculator capable of adding (adding3 of
course), we can see that calculators don't add -- or if they do, it is
certainly not as a result of following some internal microprogram."

>>I can't see any logical grounds for adopting the strong interpretation
>>of "understand" and the weak of "addition" as the "proper"
>>interpretations. It seems as arbitrary a choice as that which allows
>>planes to fly but not submarines to swim. It is certainly sanctioned
>>by popular usage in philosophical writing that "understanding" be
>>interpreted strongly, "semantics" middlingly, and "addition" weakly.

>I think "understand" is an especially poor choice, because it
>offers endless opportunities for misunderstanding; but unfortunately
>that's what Searle used.

Agreed. It raises so many red herrings that the real issues are well
camouflaged.

>I find that I don't care which camp gets to call their interpretation
>of "understands" the proper one.  Rather than fight over this, why not
>just consider the several cases?

That is exactly what I am advocating.

>>But I don't think this contingent linguistic fact constitutes a
>>consistent philosophical view. If one adopts the strong interpretation
>>of understanding then I can see no good reason for failing to adopt
>>the strong interpretation of addition, and so arguing Searlishly that
>>although getting the bit strings to come out right is the right
>>behaviour, it crucially misses the subjective perception of number.

>Is that all you mean by a modified CR argument, "it crucially misses
>the subjective perception of number"?

Yes. It does seem trivial and silly, doesn't it? If we are agreed on
that point, then just what is it about the "understanding" CR argument
that makes it less trivial and silly than the "addition" one? Apart
from the shoal of kippers of course!

>I think subjective perception of number is a red herring, BTW.
>So is "subjective feel of understanding".  Set aside the TT debate,
>source of so much distortion.  Imagine something that miserably
>fails to exhibit a behavioral understanding of what "dog" means.
>Indeed, if you ask them they say "Yeah, you've got me.  It's
>just a meaningless symbol to me, but I've picked up some of the
>ways it's used in English and I thought I could get by."  Is
>what this guy lacks the "subjective feel of understanding"?  

U1, U2, and U3 are a hierarchy. If you fail to get the behaviour right
you don't understand in any sense. If you get the behaviour right by
the wrong method (e.g. parroting) then you don't understand-2 or -1.
Searle, in a long established philosophical tradition that goes back
well beyond Brentano, holds that the quintessential defining property
of understanding is the conscious awareness of understanding1. Of
course understanding is no less understanding for being liable to
error. Your example is of someone trying for understanding2 but not
getting it right.
-- 
Chris Malcolm    cam@uk.ac.ed.aifh          +44 (0)31 650 3085
Department of Artificial Intelligence,    Edinburgh University
5 Forrest Hill, Edinburgh, EH1 2QL, UK                DoD #205
"The mind reigns, but does not govern" -- Paul Valery
