Test; Fuzzy; Errorloess?; Demo speed

[From Bill Powers (951116.0600 MST)]

Martin Taylor 951115 13:45 --

        2. Predict what will happen if the person is NOT maintaining
           the variable at a preferred level.

     Stick will not move, cat will not escape.

        3. Apply various amounts and directions of disturbances
           directly to the variable.

     Change the location of stick, and pressure required for escape,
     etc.

        4. Measure the actual effects of the disturbances.

     Cat moves stick and escapes.

Ah, I finally see what you're getting at. If the cat is reorganizing, it
will always end up, somehow, brushing against the stick and escaping. So
when you disturb the position of the stick, the cat appears to
compensate for the disturbance by varying its action so contact with the
stick is restored, even though it may not be controlling any perception
related to the stick. The reorganization masquerades as control of some
perception associated with the stick, simply because moving the stick is
the only thing that will open the box and eventually terminate
reorganization.

The Test obviously needs a way to distinguish reorganization from
systematic control. A worthy problem! Let's solve it.

The first question that comes to mind is what the cat will do if the
disturbance _increases_ the pressure of the stick against the cat --
i.e., if the stick is moved so the door is released before the cat
finishes its stereotyped movement or achieves the stereotyped position.
This would be easier to see if we put a short delay between movement of
the stick and the falling open of the door. If the cat is controlling
for pressure against the stick, the movement will stop when the pressure
is at the reference level; if not, the movement will be completed and
the increase in stick pressure will not be corrected.

Your turn.

···

-----------------------------------------------------------------------
Shannon Williams (951115) --

How else can the system doing the calculating know what the
position of the pendulum is right now? By telepathy?

     I can attempt to determine the position of the pendulum by
     attempting to calculate its current position from initial
     conditions.

This will work only when there are no independent disturbances of the
position of the pendulum. And even without disturbances, it will work
only as long as it takes for integration errors to accumulate to a
significant degree (the future position of the pendulum depends on a
nonlinear double integral).

     I can calculate how I respond to the pendulum's position by solving
     for the position desired, given the pendulum's current conditions.

Assuming that you also know the state of all independent forces acting
on the pendulum.

     And in programming these calculations, my mind need never outline
     rules that look/act like the perceptions of a biological control
     unit.

What matters is how the program you're writing knows about the position
of the pendulum. The real pendulum position would be known to the
computer through analog-to-digital converters which provide a real-time
representation of the angular position of the rod. This is what is meant
by a "perception" in PCT: an internal signal representing the state of
an external variable. Of course if you don't have a real pendulum, you
have to simulate one, as well as the A/D converter etc..

     I say what I think 'f' logic allows, and you respond with what you
     have seen done with it... There is no arguing against facts like
     that.

I don't think I understand fuzzy logic as well as I thought I did. We
should probably start over on this.
-------------------------------------------
You to Peter J. Burke 11/13/95 12:52--

     Peter:It's too bad that there is not more discussion of
     (re)organization. Where do these control loops and their parts
     come from?

     You: And how do you simulate them all?

I can't really reply to such comments and questions until I know what it
was about Chapter 14 of _Behavior: The control of perception_ that you
didn't understand or that you disagree with.
----------------------------------------------------------------------
Martin Taylor 951115 14:40 --
Chris Cherpas (951115.0918--

I'm not sure what you-all mean by errorless learning. If there's no
error to be corrected, why should there be any learning? If a person
isn't bothered by not understanding algebra, why should that lead to
understanding alogebra?

I wonder whether you aren't talking about learning without making any
incorrect moves -- i.e., the problems presented are so easy that the
right move is made right away by using an already-existing control
process. But if there is no error, why should even the correct move be
made? Somehow I don't think that the "error" you're talking about is the
same as "error" in a control system.

If learning is truly errorless, how can superstitious learning be
avoided?

I see that Kent McClelland has some problems with the idea of errorless
learning, too.
-----------------------------------------------------------------------

Martin Taylor 951115 14:55--
i.kurtzer (951110.1430)--

Isaac: >Does Reorganization preclude differential rates of reorganizing

at different levels in a posited hierarchy?

MMT: No. It is ordinarily assumed that reorganization occurs at
different rates in different levels.

There's also the consideration that the rate of reorganization must
decline as the amount of intrinsic error driving it gets smaller. That's
what makes the E. coli method work.
-----------------------------------------------------------------------
Bill Leach 951115.22:05 U.S. Eastern Time Zone --

     I suppose that there are some people running Pentium systems out
     there but has anyone run any of the PCT Demos on such a box? Last
     night I ran the various demos from Dag Forssell PCT Demo disk. The
     demos literally run too fast to really be able to watch!

This represents a problem, not a feature. All the demos are synchronized
to the vertical frame rate of the display, through reading a bit in a
hardware register. In Demo1, Demo2, and others there is a setup phase
which asks, among other things, whether you have a fast or a slow PC. If
you pick the "slow" option, sychronization is NOT used, and the demos
will run too fast on a fast PC. After this setup is done, a file called
ADCONFIG is created (in the same directory as the program) containing
the setup information, and you are not asked again. To set up again,
delete this file and run the program. You will have to do this in each
directory if you're running the demos from different directories.

Under OS/2, there is a display mode option which I don't understand but
which someone else pointed out a few years ago; if set wrong, it
prevents access to the hardware registers (or something); anyway,
correcting this mode setting allows the synchronization to work. You'll
have to ask an OS/2 expert for details.

Let me know if this solves the problem. If it doesn't solve it we hve to
look further.
-----------------------------------------------------------------------
Best to all,

Bill P.

<[Bill Leach 951116.22:11 U.S. Eastern Time Zone]

[Bill Powers (951116.0600 MST)]

... All the demos are synchronized to the vertical frame rate of the
display, through reading a bit in a hardware register.

Ok, that explains what is happening then, the vert. frame rate is 90 Hz
on this machine I believe. That would be pretty fast I should think.

-bill

[Martin Taylor 951120 14:10]

Bill Powers (951116.0600 MST)
       Martin Taylor 951115 13:45

...
    Stick will not move, cat will not escape.

       3. Apply various amounts and directions of disturbances
          directly to the variable.

    Change the location of stick, and pressure required for escape,
    etc.

       4. Measure the actual effects of the disturbances.

    Cat moves stick and escapes.

Ah, I finally see what you're getting at. If the cat is reorganizing, it
will always end up, somehow, brushing against the stick and escaping. So
when you disturb the position of the stick, the cat appears to
compensate for the disturbance by varying its action so contact with the
stick is restored, even though it may not be controlling any perception
related to the stick. The reorganization masquerades as control of some
perception associated with the stick, simply because moving the stick is
the only thing that will open the box and eventually terminate
reorganization.

The Test obviously needs a way to distinguish reorganization from
systematic control. A worthy problem! Let's solve it.

When I started this, I had a vague feeling there was a puzzle in this
situation. As the discussion evolved, I clarified this in my head as two
distinct problems. Now you have pointed out a third! Here are the
three problems--I wonder how many more will emerge from this old study?

(1) How can the cat-scientist determine that the appropriate controlled
perception is that of the stick moving?

(2) How can the god-experimenter determine what perceptions the cat is
controlling when (by hypothesis) the cat is continnually reorganizing until
a side effect of some perceptual control results in the stick being moved?

And now the new question (3) How can the experimenter determine whether the
cat is controlling any kind of perception of the stick, or whether the
success of the Test is a consequence of the environmental constraints?

You address (3), and so shall I.

The first question that comes to mind is what the cat will do if the
disturbance _increases_ the pressure of the stick against the cat --
i.e., if the stick is moved so the door is released before the cat
finishes its stereotyped movement or achieves the stereotyped position.

I'm not sure what you envisage, but I'm assuming that the experimenter
does two things while observing the cat. (1) When the cat minimally brushes
the stick, the experimenter pushes the stick against the cat, and (2)
at the same time the experimenter opens the door.

If this is the right interpretation, would it not be the case that the cat's
reorganization would stop at the "brush the stick lightly" state? Indeed,
the cat may not be reorganizing at this point, if it has already learned
to escape by the stereotyped procedure, so that the change of conditions
would not be expected to affect the stereotyped behaviour if the previous
requirements for "stick pressure" were later to be reinstituted by the
experimenter.

If it isn't the right interpretation, you'll have to explain further. At
the moment I don't see what the experiment would tell you.

This would be easier to see if we put a short delay between movement of
the stick and the falling open of the door. If the cat is controlling
for pressure against the stick, the movement will stop when the pressure
is at the reference level; if not, the movement will be completed and
the increase in stick pressure will not be corrected.

I think I must be missing either the point or the description. Suppose
a human person rather than a cat person were thge subject of this experiment.
And suppose the human person had correctly deduced that moving the stick
was the key to getting out of the box. If the stick pressure were to be
increased, wouldn't the human person push harder, to achieve the desired
movement of the stick, and yet harder if the door opening were to be
delayed? If the cat's stereotyped behaviour doesn't open the door, it
will continue to reorganize, but if the door soon opens, not much of the
old behaviour will have reorganized away, and future openings would presumably
mostly follow the same old stereotyped behaviour. Or am I way off beam, here?

Your turn.

I think the problem, from the experimenter's viewpoint, is that there is an
environmental constraint that is likely to be satisfied by random behaviour.
If the cat had two or three ways of escaping, that constraint wouldn't be
there, and if it continued to use the stick despite changes in the stick's
position and/or sensitivity, rather than by chance using one of the other
escape possibilities, then the experimenter might be more justified in
believing that the cat was controlling some kind of stick perception.

Imagine, for example, that the cat could escape by walking onto a particular
floor switch, or by rubbing on the wall away from the stick, as well as by
moving the stick. The different means could be designed so that frantic cats
might be about as quick to hit one as another, by chance. When one method
is used to escape, change it slightly, so that the same stereotyped actions
will not work. See if the cat uses the same method anyway, because if it
escaped by side-effect, the other methods should be as likely to be used.
Keep doing this, disturbing whatever escape method was used last time.

I don't know if this solves question (3), but I don't think it addresses
problems (1) or (2).

Problem (1)--How the cat can determine what gets it out of the box--is a
problem that is faced by every scientist, science being only a formalized
version of developing perceptual control structures. How do you know that
what you think you have found out about the way the universe works is right,
and that you haven't simply found something that correlates with an
environmental constraint?

Less puzzlement, more problem.

Martin