Newt

[From Bill Powers (921215.1530)]

Gary Csizo? or Jim McIntosh?

I got a message with a header saying it was from Jim McIntosh and
signed -Gary, and containing a message to and from Linda
Littleton. TOTAL CONFUSION.

According to Linda Littleton's reply, the cause of my mail not
getting out to NetNews is a bug in her gateway program, which
evidently she will now proceed to fix. I have not changed
anything in the way I send things, and anyway what goes out on
the net by way of header information is completely out of my
control. I asked at Fort Lewis' computer center, and they say
they have used the same software for three years. If all Linda
has to do is collapse consecutive blanks (that's IBM for spaces)
to a single space, this shouldn't take long -- it's a nobrainer.

ยทยทยท

--------------------------------------------------------
Bill Haley --

Thanks for the encouragement. After all this time I'm not about
to give up. None of this is new to me. In fact it's kind of nice
to see others experiencing the problem -- now they won't think
I'm a whiner.
----------------------------------------------------------
Dennis Delprato (921215) --

Very very nice analysis of The Problem. I keep thinking that if
we could just make the explanations clearer, the referees would
get the point. But every time we make it clearer, they see more
clearly why they don't want that stuff in THEIR journal.
------------------------------------------------------------
Gary Cziko (921215.1956 GMT) --

The vestibulo-ocular reflex (VOR) seems to be a true open-loop
system. It moves the eyes counter to the direction the head
moves.

Like all open-loop systems it is crude and approximate. I believe
that the mean error for a head-turn of 40 degrees or so (in the
dark) is something like 15 degrees. Wayne Hershberger probably
has more truthful numbers. If you fixate on something and turn
your head as rapidly as you can to one side, you'll see the image
jump, and then a saccade or two to correct the error. All this
system can do is reduce the amount of correction the control
systems have to apply after a very sudden head-turn.

For slow head turns the eyes simply stay locked on target. The
amount of reflex signal must be small, because if there were any
such reflex signal it would cause an error in tracking. The
detectors in the macula report angular velocity, which gives the
eyes a jolt to the side that is large for larger peak velocities
and then an opposite jolt when the movement stops abruptly. Or
something like that. This is effective for large head movements
that occur within the normal saccadic delay.
The amount of this reflex adapts over a period of perhaps 20
minutes, implying a higher level of (very slow) control of the
correction. When the target is made to move with or against the
eye movement, there is at first a very large error that gradually
becomes smaller on successive trials. The final accuracy is never
very good, as I understand it. About what you'd expect of an
open-loop system. It's nothing compared with the arc-minute
precision of optical tracking.

There are probably other such open-loop systems, but they aren't
very important for normal control situations.
----------------------------------------------------------------
Rick Marken -- I got your test message
----------------------------------------------------------------
Martin Taylor (921210. 1710)--

The problem with the Little Man disk was at the receiving end.
Glad to know yours is working.

As to the planning question, as I was reading the problem I was
thinking how I'd go about solving it. I realized first that I
would do it in imagination, as you suggest. But then I realized
that this would enable me to circumvent some of the more
restrictive rules about the stacking order of the items in the
basket. I can do this in imagination because I can run the
simulation as many times as I like; if items come out in the
wrong order I can back up and change part of the route to get
them in the right order. In fact I can run the procedure forward
and backward, revising as many times as I like, until I get lucky
and find a solution.

Another realization was that I would visualize the locations of
the items, and then run paths from one item to the next in
imagination or on a piece of paper, like drawing construction
lines. I'd first locate the items on the list, then start with
the ones that had to go under others and go zip-zip-zip from one
item to the next, drawing the lines. Then I'd follow the lines,
picking up the items in order. Of course I realize that in really
solving the problem I'd have to get both carts into the act to
handle access to the items that were on odd squares that one cart
couldn't reach; the easy way is to bring both robots along and
fill the basket of the robot that has to skip squares first,
making transfers as necessary, and then let the one-square robot
finish up the list. I'm sure this wouldn't be a minimum-time
solution, but I guarantee that if I found any solution I would
lose interest in the problem immediately. In fact, I already
have. The only places in real life where the rules get that
complicated are in games. One has to be an intellectual-game
aficionado to put any real effort into meeting all the
requirements.

The same thing goes on in chess, doesn't it? You can't actually
touch the pieces, but in imagination you can say "If the knight
goes there, it can attack there, there, there ... and now thebishop or the rook
can reach it so that's no good ..." Or am I
betraying the reason that I'm such a poor chess player?

I'm very much an analog problem-solver, and I rely heavily on
trying all the combinations until I hit one that works (even when
there are, as I find out later, after giving up, 100E100 possible
combinations). That's pretty stupid, but it does answer one of
your questions about "reorganization" that isn't part of the
reorganizing system. I think we learn search strategies as
program-level algorithms, and use them to find ways of affecting
controlled variables. The algorithms we learn can be crude and
inefficient, like mine, or very clever and efficient like good
chess-playing programs. We have to reorganize to learn or invent
these algorithms, but once they're learned they can substitute
for true reorganization if they work fast enough to prevent
intrinsic error. This applies to other things we learn like
reducing algebraic expressions to canonical forms. This isn't
learning a specific set of moves, but learning how to search for
moves that will get us closer to the final form we want. The
initial learning of these strategies, in algebra class, requires
some painful reorganization, but once they're learned even a
duffer can fumble through to the answer just by trying everything
possible and legal in imagination (supplemented by that wonderful
invention, writing).

One more note. When I solve problems in imagination, I don't
usually see all the environmental "barriers" that makes some
solutions unworkable. But in the course of running the problem in
imagination, the barriers show up all by themselves and get in
the way. It isn't that I anticipated them; it's just that the
mental model of how things work naturally spits up those barriers
when I try to make the world work in a way that violates my model
of it.

Here's an example. Imagine a golf ball. Now imagine a coke
bottle. Now, without stopping to analyze anything, imagine
dropping the golf-ball through the opening into the bottle. Does
it go through? Mine doesn't; it bounces off. The hole is too
small.

I just checked this out with Mary, and she had no trouble at all
dropping the golf ball into the coke bottle. I asked "Well, what
about a beach ball?" Oh, no -- it wouldn't fit inside the bottle!
But what about getting it through the neck of the bottle? Oh,
well, when you're imagining, it just sort of squeezes through --
you didn't say how real-world this was supposed to be. Then we
tried threading a needle with a piece of knitting yarn -- sure,
no problem, you just get an embroidery needle... thus erasing my
model with an ordinary needle in it, which of course has an eye
too small to get the yarn through.

As to the coke bottle revisited, we then decided that when you
really mean it, the golf ball refuses to go through. A BB-shot is
easy -- it drops right in. But a marble? Well, that's borderline.
The mental models aren't that accurate. You'd have to go get a
coke bottle and a marble and try it.
The upshot is that we both decide things of this sort in
imagination just by trying them and seeing what happens. We don't
refer to symbolism, like "a golf ball is 1.5 inches in diameter
and the opening in a coke bottle is about 3/4 inch, and 3/4 is
less than 1.5 so it won't work." We just look, in imagination,
and see if it works.

This doesn't work for everything; I can't see the square root of
7.5 just by looking at a length of 7.5 in imagination. I have to
go the symbol-manipulation route, on paper.

Obviously, explaining how these mental models work in such
shocking detail isn't going to be easy. The imagination
connection is probably remotely related to what is really going
on. Furthermore, there's no reason to think that everyone plans
in this way; there's probably no one method of planning. One
person draws columns on a piece of paper and lists everything;
another just starts pushing possibilities around. "Planning" is
really a catch-all word, that covers lots of things people can do
at different levels.
---------------------------------------------------------------
Best to all,

Bill P.