Newsgroups: comp.ai.nat-lang,alt.cyberspace,alt.internet,alt.net-scandal,comp.ai,comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news3.near.net!noc.near.net!bigboote.WPI.EDU!news.mathworks.com!udel!gatech!howland.reston.ans.net!ix.netcom.com!netcom.com!jqb
From: jqb@netcom.com (Jim Balter)
Subject: Re: Eliza (was Re: Are there non-humans lurking on Internet/Usenet?)
Message-ID: <jqbD3r784.KE4@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <3h2qas$m4f@percy.cs.bham.ac.uk> <3hb12k$1le@mp.cs.niu.edu> <jqbD3pB6w.94K@netcom.com> <3hc1uf$53l@mp.cs.niu.edu>
Date: Thu, 9 Feb 1995 22:16:52 GMT
Lines: 106
Sender: jqb@netcom8.netcom.com
Xref: glinda.oz.cs.cmu.edu comp.ai.nat-lang:2846 comp.ai:27298 comp.ai.philosophy:25388

In article <3hc1uf$53l@mp.cs.niu.edu>, Neil Rickert <rickert@cs.niu.edu> wrote:
>In <jqbD3pB6w.94K@netcom.com> jqb@netcom.com (Jim Balter) writes:
>>In article <3hb12k$1le@mp.cs.niu.edu>, Neil Rickert <rickert@cs.niu.edu> wrote:
>
>>>If asked to argue against Eliza understanding, I would be
>>>nonplussed.  It is so obvious that there is no understanding, I
>>>cannot easily comprehend what it is that I am supposed to be arguing
>>>against.
>
>>Given this definition of understanding from Random House 2nd ed.:
>>
>>	knowledge of or familiarity with a particular thing; skill in dealing
>>	with or handling something
>
>>I would say that Eliza displays *some* understanding.  Perhaps if you or Jeff
>>could offer your definitions of the word, any disagreement over the issue
>>could be resolved.
>
>We cannot always take dictionary definitions very seriously.

Of course.  My point is that we cannot take unstated "intuitive" notions of
what words mean, and any conclusions drawn from them, very seriously either.
If you won't tell me what "understanding" means in terms by which I can
determine whether something understands, I will find all your conclusions
about whether something does or does not understand to be mysterious and
unsupported.  In particular, I find any categorical statement about whether
something does or does not understand to suffer from the fallacy of the beard.

>This
>one is somewhat question begging, given that "knowledge" and
>"familiarity" are not easy to define either.  Skill is a measure of
>understanding, although short of a definition.  I suppose that
>understanding has two meanings.  The skill/knowledge/familiarity is
>more-or-less what we mean when we ascribe understanding to someone.

If one is non-randomly successful at something, it is reasonable to say
that one is skilled.  Why is Eliza not skilled at eliciting personal
commentary (from the sorts of people that Weizenbaum has success with)?
Can you give an answer that doesn't include anthropomorphic bias?

>But when somebody claims to understand they mean something highly
>subjective and not externally visible.

Taking one possible meaning of "understand", when I say I understand you I
mean that I do not notice any ambiguities or unknowns in your utterance; my
semantic handler doesn't raise any flags.  This is necessarily
process-relative.  Eliza may also understand your utterances at this level;
some utterances are "understood" by matching against certain patterns; others
may fail to match, and Eliza may actually say "I don't understand you".  Now
you can say that's not what you meant by "understanding".  Of course, because
you haven't said what you meant, you haven't deconstructed the notion, you
haven't laid out all the hidden assumptions and biases in this "subjective"
notion of understanding.

When I say I understand something, I also often mean that I have a familiarity
with it that has reached a point that "understand" seems appropriate.  So
there is a certain emotional component of an internal "sense" of
understanding.  But is that what we mean by the word?  There again needs to be
a bit of deconstruction here.  If I think I understand something, I can still
wonder if I really do.  Thus the sense and the fact are different within my
semantics.  To show that I *really* understand something, I need to test the
understanding.  I think it is a mistake to think that there is some sort of
holistic "understanding" that occurs in my head, that I can sense when it
occurs, and that I then say "I understand".  Again, beware of the fallacy of
the beard.  "Understanding" is not either/or; it is not even linear.

>We do test student understanding with examinations, and that fits
>with the "skill" portion of the dictionary definition.  Sometimes a
>student didn't study, doesn't understand the question, but tries to
>be creative.  I think the term for that type of answer is
>"bullshitting".

Ok, so by "understanding" a question, you mean that the recipient has grasped
the intended meaning.  I think of that as having created a semantic mapping
that to some degree parallels that of the creator of the question.  We can
talk about degrees of understanding and the accuracy of the mapping, but in
the case of Eliza we can probably agree that, for that meaning of
"understand", Eliza has not created much of a map.  But there are certainly
other programs that do create such maps, yet some people (such as Searle)
would deny that that constitutes "understanding".  I suspect that they are
really saying that the subjective emotional state mentioned above has not
occurred.  That's why it is important to do some deconstruction and untangle
these different aspects of the word.

BTW, note that one can understand a question perfectly well without being able
to answer it.  In my experience, bullshitting is usually a matter of
understandning the question well enough to provide an answer, just not the
correct answer.

>Perhaps I asked Eliza questions that were too hard, but when I first
>tried it, I only saw  an absence of understanding.  It either
>repeated a keyword from my sentence, "Tell me more about computers",
>or got into canned responses "Tell me about yourself".  In short,
>Eliza was bullshitting, but not very creatively.

Eliza "understands" sentence forms (well, sometimes; the versions I've
encountered make lots of trivial grammatical errors) but not sentence
content.  If Eliza did not "understand" the forms then the responses
would seem inappropriate even as bullshit (as when it makes grammatical
errors).  Without a careful definition of "understand" that does not
exclude many things that we mean by it when done by humans, and that
avoids either/or thinking that leads to fallacies of the beard, it is hard
to argue whether we can drop the scare quotes.
-- 
<J Q B>

