[From Bill Powers (930925.0800 MDT)]
RE: visit to U of Illinois in late '70s. I awoke this morning
remembering that some time around then I gave a seminar at the
psych department, I think sponsored by Brad Keeney (?). I don't
specifically remember meeting Chris Wickens there, but in that
context the name resonates. They didn't hire me.
Avery Andrews (930925.1215) --
Your example of fixing the lawnmower brings up a lot of
fascinating problems. One of them is the answer to the question,
"Why did I take the top off the lawnmower?" The answer could not
have been "to grasp the frayed end of the cord and stuff it
through the hole," because you couldn't see either where the end
of the cord was, or that it was frayed, or what "stuffing" would
entail. Before the top is off, all you have is an imagined model
of what might be under the cover (assuming you've never done it
or seen it done before). This model tells you that if you try to
reach the cord without taking the top off, the top will get in
the way. You don't need to try this in actuality -- reach for the
imagined position of the cord and try to grasp it. You can see
directly, in imagination, that your model won't work that way. So
you can see that the top must be removed.
Then your model tells you that the top is held on in some way (if
the casing is all in one piece you'd have to hacksaw it off,
which doesn't seem a reasonable requirement -- how, you wonder,
would they have assembled the thing?). Anyway, this leads you to
look for something to unfasten. A bolt head or slotted screw head
would fill the bill. So you search for something in that
category.
You can't actually see the part of the bolt or screw that is
under the cover, but you can imagine it. Your model includes the
properties of bolts and screws, namely the fact that twisting is
involved; also your model contains the correct direction of twist
for unfastening. So you find the unfastening tool and use it, and
remove the cover.
Now this really gets interesting. Removing the cover unveils a
new world of things you've never seen before. The first job,
before you can even try to do anything, is to make sense of it.
This means getting everything you see categorized. There's the
"pulley". That must be the "cord" wrapped around it. And that
fuzzy-looking thing, alas, must be the "end of the cord." Looking
at the inside of the cover, which you have never seen before, you
search for something that is the "hole" as seen from the inside
-- the same "hole" that you see from the outside.
Having identified everything in terms of the category to which it
belongs, you can now proceed in several ways.
You could refer to a gigantic knowledge base in which words like
"hole" and "end of cord" and "stuff" occur, and search for
sentences describing operations that contain these elements,
until you find a set of sentences like "pick up the end of the
cord and stuff it through the hole." Then you would search for
nonverbal actions that would be correctly represented by this
sentence, turn them into reference signals for specific actions,
and execute.
Or, you could simply imagine sequences of things happening, like
your fingers holding the end of the cord, pulling it toward the
hole, and pushing it through the opening. This method would have
the advantage of being cast directly in terms of the relevant
reference signals; the images being generated would serve without
translation as reference signals.
In fact, I think the second method has to be used even when you
elect to use the knowledge base approach. Right from the start,
there has to be some imagined condition that you wish to create:
the cord is threaded through the hole, and the handle is
reattached. If you haven't already gone through creating this
imagined condition, how will you recognize the sentence that will
achieve it, as opposed, say, to the sentence "The cord is lying
in a tangle around the pulley"? The _very first_ thing you must
do is imagine the perceptual state of affairs that will
constitute a solution to the general problem of repairing the
lawnmower.
ยทยทยท
-------------------------------
Another fascinating thing that your example brings up is the role
of reorganization. Clearly, reorganization is always on tap,
ready to be used. Equally clearly, it is brought to bear on
precisely the control problem that is in consciousness. This
seems to support the idea that reorganization follows awareness.
When you're trying to imagine an action that will correct an
error, you're reorganizing. When you're trying to recognize the
elements of an unfamiliar part of the world, you're reorganizing.
You look at the bunched-up cord stuck in the hole, and think "I
want a smooth cord going right through the hole." How to make it
smooth? Reorganize, reorganize, reorganize ... and finally you
imagine it wrapped in something that will keep it smooth. What
kind of something? Glue? Glue dripping into motor, me sitting
around waiting for it to dry. Thread? Tedious wrapping, knot
coming loose, weak thread, can't find thread. Scotch tape? Ah,
Scotch tape works fine (in imagination), it's in the kitchen
drawer, it sticks to itself. Do it. While we describe all this in
words, the actual imaginings are in images. _Image_-ination.
----------------------------------------------------------------
I'm on this tack because of having been reading BBS -- Mary
subscribed to it. I'm amazed at how some people try to make words
and symbolic operations accomplish everything. I don't doubt that
symbolic operations are used, but there are many control
problems, even some rather complex ones, that can be solved
without them. It seems to me that the "program" level would
operate in terms of pure symbols without meanings, and purely
logical or at least rule-driven operations. When you start with a
word-problem like "A farmer has a silo 17 feet in inside diameter
and wishes to store 110,000 bushels of corn in it ...", the first
thing you do at the program level is to convert it to a symbolic
program in which it doesn't matter what the specific symbols
mean. When you apply the formula for the area of a circle, you
don't care whose circle it is, or what color it is, or whether it
represents the shape of an object or of a hole. You just go
through the routine of applying the rule A = pi * r * r and
coming up with a number for A. At this level, it doesn't matter
what problem you're trying to solve by evaluating this
expression. The program (or maybe sequence, here) level doesn't
have to know any of the lower-level details, nor could it
perceive them even if given them.
When you repair the lawnmower, you achieve the goal using the
levels of control that are actually required. You don't use
sequence control unless sequence matters, and you don't use
symbol-processing unless the particular skills involved in
program-like symbol manipulation are required. You would probably
use the program level if the problem were to figure out how to
execute the steps involved in repair with the least expenditure
of energy or the lowest cost or the greatest chance of impressing
your wife with your dedication to household labors. Program-type
thinking is good for achieving program-like goals; it isn't
needed for anything else.
Sorry to free-associate like this, but it was a provocative
example.
-------------------------------------------------------------
Best,
Bill P.