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!uunet!olivea!news.hal.COM!decwrl!netcomsv!netcom.com!jqb
From: jqb@netcom.com (Jim Balter)
Subject: Re: Penrose and Searle (was Re: Roger Penrose's fixed ideas)
Message-ID: <jqbD0Kqxq.6M4@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <D03L02.J5B@gpu.utcc.utoronto.ca> <D0Cwn2.1t9@cogsci.ed.ac.uk> <jqbD0DDvB.MHy@netcom.com> <D0Eqo9.5B0@cogsci.ed.ac.uk>
Distribution: inet
Date: Sat, 10 Dec 1994 02:58:38 GMT
Lines: 292
Xref: glinda.oz.cs.cmu.edu sci.skeptic:97597 comp.ai.philosophy:23485 sci.philosophy.meta:15458

In article <D0Eqo9.5B0@cogsci.ed.ac.uk>,
Jeff Dalton <jeff@aiai.ed.ac.uk> wrote:
>In article <jqbD0DDvB.MHy@netcom.com> jqb@netcom.com (Jim Balter) writes:
>>In article <D0Cwn2.1t9@cogsci.ed.ac.uk>,
>>Jeff Dalton <jeff@aiai.ed.ac.uk> wrote:
>>>In article <D03L02.J5B@gpu.utcc.utoronto.ca> pindor@gpu.utcc.utoronto.ca (Andrzej Pindor) writes:
>>>>In article <D01FA6.DuK@cogsci.ed.ac.uk>,
>>>>Jeff Dalton <jeff@aiai.ed.ac.uk> wrote:
>>>>>In article <Czzrvs.A1u@gpu.utcc.utoronto.ca> pindor@gpu.utcc.utoronto.ca (Andrzej Pindor) writes:
>>>>>>In article <3b5d05$d2o@news-rocq.inria.fr>,
>>>>>>Mikal Ziane (Univ. Paris 5 and INRIA)  <ziane@monica.inria.fr> wrote:
>>>>>>......
>>>>>>>
>>>>>>>My point was precisely that I do not think TT is a very good definition
>>>>>>>of intelligence and I think that this is what CR suggests albeit clumsily.
>>>>>>
>>>>>>It probably is not, but Turing thought that it was the best we could do and 
>>>>>>not much chaged since then. Or perhaps you have a better definition?
>>>>>>I can't see how CR suggests anything of the sorts. In fact, being methodolo-
>>>>>>gically wrong, it does not suggest anything.
>>>>>
>>>>>Andrzej -- can I tell Ozan Yigit that you defend the TT?  From this,
>>>>>but more from other articles, it seems to me that you do.
>>>>>
>>>>Depending what you mean by "defending". In a sense - yes. However, it seems
>>>>to me that you are trying to box me into a silly position which I do not
>>>>hold. 
>>>
>>>I am happy to be corrected.  If I seem to be saying you hold that P
>>>when you actually hold otherwise, for any P, please let me know.
>>
>>How can one say when P is poorly formed?
>
>In that case, say that P is poorly formed.

I said pretty clearly "you defend the TT" is poorly formed.

>>>As for my dispute with Oz, I hope I turn out to be wrong.
>>>If, in fact, many disagree with the TT I'd regard that as
>>>an excellent result.
>>
>>It would help if you defined your shorthand "the TT" (I take it to be
>>shorthand, since noun phrases don't make good propositions).  You might mean
>>anything from "the TT is a useful tool" to "ability to pass the TT is the one
>>and only possible definition of consciousness".
>
>By "the TT" I mean the teletype-based Turing Test.  Perhaps you wanted
>to know what I'd count as agreeing or disagreeing with the TT?

Hmm, must be a reading comprehension problem.  Jeff says "many disagree with
the TT".  I ask him to define "the TT" since one normally disagrees with a
proposition.  I of course didn't ask him to define "many disagree with".  So
now we have "many disagree with the teletype-based Turing Test".  Of course,
this still isn't a proposition, so we still don't know what P it is being
claimed that persons do or do not disagree with.  Now Jeff wants to know if I
wanted to know what he'd count as agreeing or disagreeing with this
non-proposition.  Well, no, I want to know what the proposition is, just as I
said.  But perhaps we will get there, even though Jeff doesn't seem to know
what a proposition is.

>If so,
>I'm interested in the whole range, including (on the agree/defend side)
>"the TT is an interesting goal", "the TT is a good starting point",
>"the TT is a useful tool", "the TT defines `consciousness'", your
>"ability to pass the TT is the one and only possible definition of 
>consciousness", and so on.

Ah, well, now we get propositions, when we were expecting something that would
or would not count as agreeing with a proposition.  Well, we try to make
allowances.  So "the TT" could be short hand for any one of those, I guess.
Well, no, Jeff said "the TT" is shorthand for "the teletype-based Turing Test".
So perhaps we are just supposed to guess which proposition Jeff has in mind at
any moment.  Perhaps it's a multiple-world sort of thing, and we are supposed
to defend or attack all propositions simultaneously.  For instance this one:
"Ability to pass the teletype-based Turing Test is the one and only possible
definition of consciousness".  But isn't the teletype-based Turing Test a
replacement for a definition of intelligence?  If so, that proposition doesn't
seem to make much sense; it certainly doesn't seem to be the sort of
proposition that intelligent people here would try to defend, and even if it
were, it hardly seems like a position that an intelligent person would bother
to challenge.  But then we are dealing with someone who thinks that "the TT"
is a proposition and "the TT is an interesting goal" "count[s] as agreeing or
disagreeing with the TT".  Or something.  I don't know, my parents only gave
me a nickel for my allowance ('twas a long time ago).

Of course, Jeff says it isn't the meanings of words that matter, just what is
case.  Perhaps that explains it.

>>>>Mark Rosenfelder presented a position which basically is the same as 
>>>>mine.
>>>
>>>Really?  I agree with most of what he said, but I often feel I
>>>disaagree with you.  (Not always, of course!)
>>
>>Could be a reading comprehension problem.
>
>Could be.  Does Mark Rosenfelder agree with Andrzej Pindor's view
>of the TT?  Perhaps he will say.

I suppose it depends upon where he's standing.  I think it is best to get
as high as one can.

>>>I think "understand" is an unfortunate choice.  I don't think the
>>>sense of "understand" at issue w/ Searle is the quite same as that in,
>>>say, "Bill understands Godel's theorem".  The Searlean sense is
>>>more along the lines of whether words mean anything to Y (or does
>>>Y treat them as meaningless symbols).
>>
>>Well, one of Searle's (and his cohorts in intellectual crime) tricks is
>>to use a word with multiple, broad, or vague meanings and switch between them
>>as serves the argument.
>>
>>It seem to me that you have simply moved the vagueness from "understanding"
>>to "meaning".  Do Chinese words mean anything to the CR?  They certainly
>>seem to.  By some possible accounts of meaning, it is meaningless to speak
>>of symbols that are used in a process to be "meaningless".
>
>So what?  The important issue is not what words mean but what is
>the case.  We may need to look at a number of different accounts
>of meaning.  But the important thing is what's so, not which words
>we use to talk about it.

Perhaps now Jeff is advocating telepathy, since we need not worry what words
mean.  We will just communicate "what is the case" mind to mind (or algorithm
to algorithm).

>>How do we determine whether various sounds mean anything to a fellow
>>human being?  How would you determine whether the Chinese words mean anything
>>to me?  Would you look at my program?  Examine my interior dialog?
>
>If I had your program, I'd look at it.

I'm sure you would, but of course this still leaves me wondering how you
would determine whether Chinese words mean anything to me, with or without
my program.  I suspect I'll be wondering for quite a while.  Not that
I really care all that much what methods an obstructivist like you
(ooh, that ad hominem should be good for a few rounds, but let's call a spade
a spade) might use.

>Interior dialogue is not something I say is necessary for
>understanding, for consciousness, or for words to mean something.
>I introduce it as another, perhaps less vague and otherwise
>difficult, thing to test for.  

Yes, let's come up with as many tests for specific artifacts of human
mentation as we can.  That way we can be sure that, if we ever do conclude
that something is conscious, we will be dealing strictly with what is the case
and won't have to bother with any nastiness about what words mean.

>>>In any case, it does seem to be possible to pass tests without
>>>understanding the subject or at least w/o understanding it as well
>>>as the test results suggest.
>>
>>Amazing how induction can fail, isn't it?  The question is, how do we make
>>the determinations we do, not how accurate they are.
>
>The question is not how we make the determinations we do but how
>we should make determinations when it comes to programs/computers.

Ah, yes, of course, because what words mean isn't relevant to how 
we make determinations; there is simply what's so, what is the case.
So of course we cannot apply what methods we use now, because
what is the case for programs is or may be different from what is the case
for humans.  What we actually *mean* by "understanding" or "consciousness"
or whatever not being intimately connected with *how* we actually determine
whehter it is present.

And of course, we don't want to go back and explain what possible relevance
there was to pointing out that tests have false positives.  After all,
that would lend some sort of coherence to the discussion, but discussions are
just matters of meanings of words and may have little to do with what is the
case.

>How we (now) make the determinations we do is just one thing we
>might consider.

Ah, yes, it is just one thing to consider, but not "the question".
"The question" was how students can pass tests but not understand.
Or was it?  Oh, who cares; that's just a matter of what words mean.
Or is it a matter of what is the case?

>>  Andrzej is wrong to
>>say that we *define* "understanding" as "passing a test"; we define it as
>>"grasping the subject", or some such; but our (necessarily fallible) *test*
>>to *determine* whether the subject has been grasped is a TT of sorts.  In the
>>same way, the TT is not a *definition* of intelligence, but it is a test of
>>whether intelligence, however that attribute is defined, applies to a 
>>particular subject.
>
>Fair enough.  I merely say it's not already known that the TT is a
>reliable test of, say, consciousness.  Or whether internbal dialogue
>takes place, or whether the subject has any subjective experience
>at all.

Well, let's be sure not to say just what we mean by "the TT", and to
introduce "internal dialog" and "subjective experience" as new qualities
to test for, new things that we expect the "defenders of TT" to defend,
despite the fact that no one has ever so defended it.  After it, it doesn't
matter what words mean, simply what is the case.

>>>>>It may be that we will eventually establish that the TT is a
>>>>>reliable test.  But that's not the only possible outcome.
>>>>>
>>>>Considering the vague notion of "understanding" we have now, basically based
>>>>on passing sort of a TT, TT is good as it is. If we at some point require
>>>>something "more", it will mean that have additional criteria what it means 
>>>>to "understand" and hence that our notion will be different, as rightly
>>>>pointed out by Neil Rickert.
>>>
>>>Well, if _that_'s the way you want to run it, I might even agree.
>>
>>Er, um, isn't this what everyone has been saying all along?
>
>It might be when it comes to understanding, but not understanding
>in Searle's sense.

But of course we don't have to explain what that sense is, because there
is simply what is the case and it doesn't matter what the words mean.

>>>At least I've often agreed with (and even made) that kind of argument
>>>(e.g. a discussion in c.a.phil a fair while back about whether future
>>>or alien maths could be radically different from ours.  I argued that
>>>if it were sufficiently different, we wouldn't think it was maths at
>>>all.)
>>
>>I don't think Andrzej was talking about radical differences of this
>>sort. 
>
>It seems to me a related issue.
>
>>He was talking about the sort of difference between "gold is a shiny
>>malleable yellow substance" and "gold is the element with atomic number 79".
>>We still think that gold is gold, but we have a different notion of what
>>it means to be gold, because we have a more refined understanding of the
>>underlying mechanism that gives rise to the characteristics of gold.
>>Rather than defining gold by its characteristics, we define it in terms of the
>>underlying model.  Such may some day become the case for "understanding".
>>
>>It's interesting how many of these discussions seem to be based upon confusion
>>about meaning and language.
>
>But they're not addressed as such.

Oh, no, neither Aaron Sloman nor I have ever addressed them as such.  Oh no.

>The arguments for the TT are
>usuallky along the lines of: we use the TT for other humans so
>it's arbitrary (or otherwise unjustified) to do anything different
>for machines.

*One* line of the defense is that additional tests need to be justified
as being appropriate or relevant to whatever quality is being tested.
Whether they are appropriate or relevant depends upon the *definition* of
the quality, the *case of the matter* as to what quality is being discussed.
That's why the meaning of words matter.

>Moreover, we seldom find: my "understanding" I mean this, and here's
>how the TT is a reliable test for it.

Yeah, especially if you edit that out every time it comes up.
What we really have is Dalton saying "the TT is not enough for understanding"
with  the response "what do you mean by understanding" or "what is it about
understanding that you think the TT doesn't capture" or, as I said in this very
thread, that "understanding" has multiple  menaings which get shifted around
to suit the argument.  The fact is that without absurd attacks on "the TT",
there would be no defense because, as Rosenfelder and Rickert and others
have pointed out numerous times, the TT has little or no practical role in AI.

>Instead, TT-defenders try to
>present a choice like this:
>
>  -- the TT and the associated verificationist/behaviorist
>     attitide to questions of this sort.
>
>  -- some unacceptable prejudice (e.g. judging by "looks").
>
>  -- dualism, mysticism, vitalism, being unscientific, etc.

Ok, so there are a couple of other possibilities, such as the one usually put
forth by me, Sloman, Rosenfelder, Rickert, and others, in one form or another,
that good answers to the questions require a clear understanding of the terms,
which can probably only come with better models of mind.  And then there's the
position put forth by Dalton, which is that there a matter of what is the case
as to whether an entity understands or is intelligent or conscious, or what
is gold, independent of the *meanings* of those words.





-- 
<J Q B>
