NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/comp/ai/philosop/7394

Internet Message Format | 1993-01-25 | 6.7 KB

Path: sparky!uunet!cs.utexas.edu!asuvax!asuacad!icjga
Organization: Arizona State University
Date: Monday, 25 Jan 1993 17:30:45 MST
From: <[e-mail]>
Message-ID: <[e-mail]>
Newsgroups: comp.ai.philosophy
Subject:    Re: Searle on animal consciousness
Lines: 93

     I have spent considerable time looking for different tests people have
offered for determining consciousness.  Of all those found, I have had to
translate them for any practical use, as they really aren't measurable.  They
tend to be of the nature of defining consciousness with other undefined and
not measurable terms.  But reading the context of whomsoever is proposing
some test allows such a translation if you work at it.
     Anyway, I have a list of about 14 such tests, and most of them require
the ability to language.  However, there are some that don't, but that seems
to arise from non-measurable definitions also, such as "emotions".  But if
one takes a position for modeling and testing that emotions have a primary
function related to learning, things seem to fall into place.  For example,
if there is an emotion that arises that is painful (arousal level going up
fast, i.e. arousal chemicals being generated quickly) and is associated with
a certain situation such as an inability to find food, then there is a
"trigger" for not only some general action pattern for locating food, but
when food is found, the de-arousal (ceasing the production of the arousal
chemicals) "locks in" the "success", the "goal reached", and what happened
just before (retained in short-term memory) can be linked to that.
     If the finding of food is a many-step plan that exceeds short-term
memory, the plan will have to be learned backwards one step at a time,
each step being what was in short-term memory before the last one learned.
     Now what does this have to do with consciousness.  Some include an
ability to learn in a definition of consciousness.  That would mean an
ability of have emotions.  There evidence that so-called "lower" species
learn just isn't there.  Ants can follow a scent trail laid down by other
ants, but that can be explained as genetics.  Rapidly-producing "lower"
species seem to "learn" by mutation, not by linking "what-came-just-before
that" sequences.
    So if emotions are part of your consciousness definition, learning
probably belongs there too.There are only a few systems that can learn.
One can learn by experiment (the cat playing with a new thing) which is
cognitive learning (learning to predict the behavior of the thing).  One
can learn from experience such as that brown thing with the long nose ran
at me and bit me, so I avoid the brown thing with the long nose.  The next
time I see one I will feel a "pain" that compels me to run away from it.
That's what-came-just-before conditioning.  Just before I was bitten last
time I saw that brown, long-nosed thing running at me.  That's conditioning.
The cat doesn't have to remember a thing about the first episode regarding the
brown dog attacking it.  It now has a cognitive predictability that one in
the proximity produces a pain that compels flight.  It has become automatic.
One may not remember that uncle Joe beat the living daylights out of one
while wearing a red plaid shirt to be afraid of large 50-year-olds with
beards who are wearing red plaid shirts.  The mammalian pain/emotion
and learning system will set up the urge to flee just fine thank you.
     To actually remember a history one must use the kind of learning
anyone uses to learn any recallable action.  To bat a ball well you must
repeat the action many times.  To learn a history of an event you do the
same thing but with words.  Your repeat your story about something many
times.  If you repeat it enough times you have it memorized.  If not, it
may differ each time you recall it, as you will have to fill in some
blanks because you didn't memorize it.  Actions strings must be repeated
to perfect them.  Like batting a ball, remembering an event historically
is an action string.
     If mammalian functional brain structures are modeled one finds that
languaging can be added with no new fundamental structures.  That is why
languaging can be viewed as a new use of the action string and that is
why language can be modeled and language learning and "story" learning
can be added to the action system.
     Most lists of consciousness tests require languaging ability as they
require the ability to store stories about onesself and recall those stories.
But consciousness isn't a single state of a brain.  It's a collection of
brain functions, and you can draw the line wherever you want.
     Those who study species holistitically such as Jane Goodall studying
chimps understands this about language.  Mammals communicate.  But they
can only communicate about "now", now being no further back that short-term
memory, and no further forward than the current instants prediction of the
alternatives that might derive from the current situation (e.g. if a chimp
sees a leopard there is a prediction that the leopard could kill it, and
therefore, from that possible immediate future the chimp will sound a
warning).  But Goodall knows that chimps have no ability to communicate
personal histories or to make plans.  That requires a full language ability.
     Also there are evolutionary reasons to be offered regarding why only
one species has languaging ability, but I have gone on long enough.  As with
every known evolutionary change that sticks, either is is something that
aids in survival (maybe not the best way but the first one that worked fine)
or it does no harm to survival.  Language appears necessary for humans to
survive in the wild in a primitive state.
     One more comment.  Your can learn one thing experientally (mammalian
conditioning) and come up with a story about why you learned that thing
(e.g. avoiding men with plaid shirts et al) and the two can have nothing
to do with each other.  One is a modification of the mammalian cognition and
decision systems and the other is a story that is a learning in the action
system where episodic memory resides.
     You don't have to buy all this, but tell me what it does NOT explain and
either I'll come back with an explanation within my model or I'll adjust my
model.  But maybe you don't buy the value of modeling.  There certainly are
arguments against it.  But believe me, if you insist on making your model
programmable, you are insisting it be measurable and structured and cocrete,
and that prevents your defining anything with mush or by other words whose
definitions are mush.  And AI's early problem that sent original "strong AI"
down the drain was discovering that it couldn't program the behavioral
definitions that were current because they were not concrete, structured and
mesurable.  So maybe a measurable model is useful.