From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!samsung!daemon Mon Mar  9 18:36:00 EST 1992
Article 4334 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!samsung!daemon
>From: smoliar@iss.nus.sg (stephen smoliar)
Newsgroups: comp.ai.philosophy
Subject: Re: Definition of understanding
Message-ID: <31046@samsung.samsung.com>
Date: 7 Mar 92 00:40:51 GMT
Sender: daemon@samsung.COM
Lines: 64

In article <1992Mar6.001859.1518@memstvx1.memst.edu>
langston@memstvx1.memst.edu writes:
>
>  Perhaps it is not the case that 'understanding' is inherent in memory
>structure.  I would argue that for an agent (read agent however you like...
>whatever makes this most interenting...) to 'understand' a concept (or what-
>have-you) the agent must be able to MAKE USE of that item in some meaningful
>way.  One could say "I understand golf" without ever actually playing golf.
>Does this person really understand golf?  I could learn the procedure for
>baking a cake, but without ever actually baking said cake, I would be at a
>loss
>as far as true, useful understanding of the complexities of the process.
>  This has implications for AI.  Does this mean that computers and such
>'understand' their internal data representations?  Not any more than the rock
>'understands' heat or molecular physics as it expands in the sun.  To
>understand, the agent must act on the information in some meaningful way that
>would integrate both semantic and episodic knowledge of the information at a
>deep level.

There is a lot of Wittgenstein in this point of view, and I am generally
sympathetic with it.  However, phrases like "in some meaningful way" are
sufficiently slippery that they really do not allow us to make very much
progress.  Also, even the aspect of use is not quite as simple as it would
first appear (or may have appeared to Wittgenstein).  This is because use,
itself, may vary radically as the context varies.  There is a nice example
of this in music.

There is now an extensive body of scholarship in an area known as "music
theory."  The largest amount of this scholarship tends to be concerned with
the study of artifacts of music notation (musical scores) and various syntactic
relationships which may be inferred from the surface structures of the
notational patterns.  Over the last twenty years, this scholarship has
drawn heavily on tools provided by mathematics, such as group theory and
number theory.  As a result, a fair amount of the mechanics of such analysis
may now be performed by computer programs.

What is particularly interesting about this approach is the behavior of some of
the scholars who practice it.  Some of the most distinguished members of the
community have openly boasted that they not longer listen to music or bother
with such obsolete practices as going to performances!  For them the music
exists only as symbols on a printed page mediated by a mind which tries to
find patters in those symbols.  (Did I hear someone mutter something about
glass bead games?)  From my point of view, this raises a significant question
about whether or not such individuals "understand" (to use the word we have
been picking on) music.

Within their own community (that of music theorists), the "understanding" of
such individuals is unquestioned.  Indeed, other scholars will flock from
far-off sites to attend conferences and receive the fruits of such
"understanding."  However, what happens when you drop one of these
scholars into a community of music PERFORMERS?  More often than not,
a performer may see no connection between all this theory and the music
knowledge which he actually uses.  He may even listen patiently to a
theoretical explanation of a work he has been playing for ten years
and conclude that the theorists does not understand the work at all!

I think Marvin's original article was trying to get at some of the critical
issues in this matter of variation of context.  Use is definitely important;
but it is also highly context-dependent.  Neither the Imitation Game nor the
Chinese Room really take such context-dependency into account.  The first was
not really intended to, since Turing made it quite clear that he was not
interested in defining intelligence in the first place.  Do we need to make
up another game?



