[From Bruce Abbott (2001.05.21.2000 EST)]
I have posted a synopsis of PCT on the web at
http://users.ipfw.edu/abbott/pct/pct.html
and would appreciate constructive comments -- inaccuracies, things omitted,
and so on -- especially from you, Bill. I'll correct anything that seems
problematic.
Best wishes,
Bruce
i.kurtzer (2001.05.21.2200)
[From Bruce Abbott (2001.05.21.2000 EST)]
I have posted a synopsis of PCT on the web at
http://users.ipfw.edu/abbott/pct/pct.html
and would appreciate constructive comments -- inaccuracies, things
omitted,
and so on -- especially from you, Bill. I'll correct anything that seems
problematic.
Bruce I have read your intro, and while I find the sentiement fine I think
1) enough intros have been written
2) its emphasises oblique issues such as Power's suggestion about memory, 11
levels, "intrinsic variables", etc
while not explaining what IS control, how control of input is unique, and
how do you test for control.
I would personally prefer more experiments with the rats. No knock to you
nor that we couldn't agree on an optimum presentation, its just
30+ years of intros haven't done a damn. I would suggest a line of research
would be a better use of your time and understanding.
i.
[From Bill Powers (2001.05.22.1451 MDT)]
Bruce Abbott (2001.05.21.2000 EST)--
I have posted a synopsis of PCT on the web at
http://users.ipfw.edu/abbott/pct/pct.html
and would appreciate constructive comments -- inaccuracies, things omitted,
and so on -- especially from you, Bill. I'll correct anything that seems
problematic.
I think you've done a fine job, although of course (as Isaac noted) all
synopses have to be selective if they're to be any shorter than the books
they summarize.
There is one thing I would like to see in any introductory material aimed
at a new audience (as I presume yours is). It is simply a discussion of the
nature and properties of a negative feedback control system, treated as an
abstract unit of organization in its own right. Many myths persist about
control systems, such as the "feedback is too slow" business. But myths
aside, it's hard to get across the differences between a closed-loop
input-controlling system and the other kinds that have been proposed: the
action-planning system, and the input-causes-output system. Everyone who
proposes to teach about PCT ought to have a try at explaining these
differences, because each new explainer will think of some way to put it
that might be uniquely illuminating for some people.
Do you have a target audience in mind, or is this more an exercise in
getting the materials organized?
Best,
Bill P.
[From Bruce Abbott (2001.05.23.1030 EST)]
Bill Powers (2001.05.22.1451 MDT) --
Bruce Abbott (2001.05.21.2000 EST)
I have posted a synopsis of PCT on the web at
http://users.ipfw.edu/abbott/pct/pct.html
and would appreciate constructive comments -- inaccuracies, things omitted,
and so on -- especially from you, Bill. I'll correct anything that seems
problematic.
I think you've done a fine job, although of course (as Isaac noted) all
synopses have to be selective if they're to be any shorter than the books
they summarize.
Thanks. Selective -- did I leave anything out that needs to be there?
There is one thing I would like to see in any introductory material aimed
at a new audience (as I presume yours is). It is simply a discussion of the
nature and properties of a negative feedback control system, treated as an
abstract unit of organization in its own right. Many myths persist about
control systems, such as the "feedback is too slow" business. But myths
aside, it's hard to get across the differences between a closed-loop
input-controlling system and the other kinds that have been proposed: the
action-planning system, and the input-causes-output system. Everyone who
proposes to teach about PCT ought to have a try at explaining these
differences, because each new explainer will think of some way to put it
that might be uniquely illuminating for some people.
I hope to be adding other papers to the collection as I find the time to
write them. I thought that the first thing to do was to get an accurate
overview of (H)PCT posted.
Do you have a target audience in mind, or is this more an exercise in
getting the materials organized?
More the latter -- I was going through B:CP again and thought it might be
nice to use the opportunity to create an outline of the proposed
mechanism(s). I was surprised to find details I had either forgotten or
misremembered (in particular, the proposal to have reference values
retrieved from memory) and thought others besides me might also benefit from
the "review." I think the next step may be to generate a list of
implications of the model, together with a description of the tests that
have been conducted to examine these implications and the results obtained.
I also plan to take up your (and Isaac's) suggestion to discuss "the nature
and properties of a negative feedback control system." With these and other
future additions in mind, I've started an index page (skeletal though it may
be at present), from which it eventually will be possible to access all of
this material as well as relevant off-site links. The URL is
http://users.ipfw.edu/abbott/pct/index.html
Best wishes,
Bruce A.
[From Bill Powers (2001.05.23.1030 MDT)]
Bruce Abbott (2001.05.23.1030 EST)--
I think the next step may be to generate a list of
implications of the model, together with a description of the tests that
have been conducted to examine these implications and the results obtained.
Excellent. Some discussion of the tests with me and others really ought to
precede the conclusions, however. For example, I don't think we agree
completely on the outcome of the attempts to model the rat experiments.
It's too bad there are so few of us -- a meaningful consensus is hard to get.
The biggest problem we had (or at least that bothered me), as I recall, was
that the rats didn't seem to increase their rate of eating in the
experimental cage when the food available in the living cage was decreased.
This implies that there was NO variable controlled by the rats' varying the
amount of food they ate in different places. That would make no sense in a
natural environment -- if the food runs out in one place, why would a rat
not seek and eat more food in another place? Did we do something to prevent
this obvious strategy from being learned? Did the fact that the rat did not
propel itself into the experimental cage in search of more food have
something to do with it? I can see how we might have done the experiment
differently -- now.
An additional possibility: At first the food given to the rats in the home
cage (after the initial total deprivation period) was artificially varied
to maintain their weight at a low level, so the experimenter was strongly
controlling the rats' weight in conflict with any control the rats may have
tried to exert. This may have had a profound effect on the young rats'
ability to control their own weights. Eating more in the experimental cage
would only result in less food being available in the living cage. Did the
rats learn that right away? But that's just speculation. I would really
like to see a "clean" PCT experiment done.
Also, when we started the experiment, I don't think either of us realized
that female rats increase their body weight fairly steadily during their
lives. Assuming a constant reference level for a weight control system
therefore was a mistake. I now think we should have _deduced_ the reference
signal from the data. Of course this would have assured a perfect fit, but
the question then would have been whether the changes in the model's
variables that resulted were clearly systematic.
I guess that episode left me pretty frustrated.
Best,
Bill P.
[From Bill Powers (2001.05.23.1125 MDT)]
Bruce Abbott (2001.05.23.1030 EST)]
Thinking some more about the rat experiments. Suppose we set the rats up so
each rat has two cages connected by a door or a passageway. In one cage,
food is freely available up to an experimenter-determined limit. In the
other cage, food is available on an FR1 schedule. The rats can move freely
from one cage to the other.
Phase 1: Food is available only in the home cage, ad libitum. Food delivery
in operant cage disabled, though you might go through a sham shaping
procedure (producing clicks but no food). Measure the intake and time spent
in each cage during this and all other phases.
Phase 2: Food is available only in the operant cage. Use shaping procedures
to help the rat learn how to get food by pressing a bar.
Phase 3: Limit the food in the home cage to a fraction F (0 < F < 1) of the
ad libitum amount per day, and enable food delivery in the operant cage.
Prediction: in the third phase, after learning, the rat will eat all the
food in the home cage and go to the operant cage to make up any shortfall.
There might be some tendency to get food in the operant cage even when food
is available in the home cage. In any case, there should be a mirror-image
relationship between the food obtained in the operant cage and the food
obtained in the home cage, as long as the total food available with
moderate effort is at least equal to the ad libitum consumption rate, and
neither source alone is sufficient to supply food at that rate. If a rat
seems to enjoy getting all its food in the operant cage, the schedule can
be changed to make that more difficult -- at some point the rat would begin
filling in from the free supply in the home cage.
The hypothesis, of course, is that the rat is controlling some variable
that depends on total food intake. If less food is available in one place,
more food will be sought wherever else it can be found, with the total
intake being about the same when other conditions are the same (exercise,
temperature, whatever).
A change in the amount of food obtainable in one place will be compensated
by an increase or decrease in consumption of food found in other places.
To help us understand what happened in our first experiment, we can subject
some rats to the same initial conditions: initial deprivation at a young
age followed by the experimenter's varying the home-cage food supply to
keep body weight constant for some period of time (was it 2 weeks?). If
this results in the rats' inability to control total food intake, our
previous results will be explained. If not....
Best,
Bill P.
Bill Powers (2001.05.23.1030 MDT) --
Bruce Abbott (2001.05.23.1030 EST)
I think the next step may be to generate a list of
implications of the model, together with a description of the tests that
have been conducted to examine these implications and the results obtained.
Excellent. Some discussion of the tests with me and others really ought to
precede the conclusions, however. For example, I don't think we agree
completely on the outcome of the attempts to model the rat experiments.
It's too bad there are so few of us -- a meaningful consensus is hard to get.
Yes, discussion would be good. My tentative plan is to begin by talking
about what I would call "proofs of principle," which mainly take the form of
computer simulations -- the "Little Man" demo, for example. I don't think
we would find too much disagreement there about conclusions.
I may not be aware of all the relevant studies, so perhaps we could start by
developing a list . . .
The biggest problem we had (or at least that bothered me), as I recall, was
that the rats didn't seem to increase their rate of eating in the
experimental cage when the food available in the living cage was decreased.
This implies that there was NO variable controlled by the rats' varying the
amount of food they ate in different places. That would make no sense in a
natural environment -- if the food runs out in one place, why would a rat
not seek and eat more food in another place? Did we do something to prevent
this obvious strategy from being learned? Did the fact that the rat did not
propel itself into the experimental cage in search of more food have
something to do with it? I can see how we might have done the experiment
differently -- now.An additional possibility: At first the food given to the rats in the home
cage (after the initial total deprivation period) was artificially varied
to maintain their weight at a low level, so the experimenter was strongly
controlling the rats' weight in conflict with any control the rats may have
tried to exert. This may have had a profound effect on the young rats'
ability to control their own weights. Eating more in the experimental cage
would only result in less food being available in the living cage. Did the
rats learn that right away? But that's just speculation. I would really
like to see a "clean" PCT experiment done.
In my view, there were two main problems with that experiment. The first
problem is that the rats had a "back-door" method for controlling body
weight that I was not equipped to monitor directly: adjustment of metabolic
rate. Thus it was possible that when food amount in the home cage was
restricted, the rat could simply reduce its metabolism and get by on less.
(I found a nice study showing that such a change does "kick in" when the
rat's body weight exceeds about a 10% reduction below normal values.)
The second and probably more important problem may have been the existence
of a short-term satiety mechanism (filling of the gut etc.) that probably
brought eating in the operant chamber to a halt before enought food had been
consumed to keep body weight near a reference value. By the time the gut
emptied enough to permit further consumption, the session was over and the
rat had no further opportunity to make up the deficit.
I did run a short test of this hypothesis toward the end of the study. This
involved placing the rat back into the operant chamber for a second operant
session in the afternoon (i.e., two operant sessions/day rather than one).
As I recall, the rats did return to the lever during the second session each
day and consume more food, which is consistent with the hypothesis. I don't
recall whether this led to better control over body weight, but I still have
the data so it shouldn't be much of a problem to find out.
Also, when we started the experiment, I don't think either of us realized
that female rats increase their body weight fairly steadily during their
lives. Assuming a constant reference level for a weight control system
therefore was a mistake. I now think we should have _deduced_ the reference
signal from the data. Of course this would have assured a perfect fit, but
the question then would have been whether the changes in the model's
variables that resulted were clearly systematic.
Yes -- in fact, I found data from one of the suppliers showing the
time-course of average weight over the lifespan. If the short-term satiety
problem did occur, as I believe it did, this would have made inferring the
reference body-weight from the data a bit problematic. However, we did have
three rats in that study that were kept in the metabolic cages (no operant
sessions) and given food ad libitum. Even there we had some problems
fitting model to data. A prime suspect there was the female estrus cycle.
I guess that episode left me pretty frustrated.
Me too, but I think that some good may have come from it -- we learned that
the system in question is more complicated than our simple model allowed
for, and have a better idea of what additional variables we need to be
measuring or taking account of in the experimental design. Moreover,
although the performance of the simulation was below the usual standards, it
actually did a fair job of accounting for much of the change in body weight
over the different conditions imposed.
Best wishes,
Bruce A.
[From Bill Powers (2001.05.23.1442 MDT)]
Bruce Abbott(2001.05.23b) --
I may not be aware of all the relevant studies, so perhaps we could start by
developing a list . . .
How about checking out Demo1 again? It is actually a series of studies in
which different controlled variables are tested, with predictions being
made about the form taken by a plot of the data, and about the correlations
that will be observed between disturbance and action, and between
controlled variable and action. Each experiment is accompanied by an
on-screen writeup. In most of the writeups, blanks are left for the actual
correlations, which are filled in automatically after the user does an
experimental run. An actual plot of each 30-second run is shown, with
disturbance, controlled variable, and action. The writeups correctly
describe (in advance) the numbers that will be produced as well as the
appearance of the experimental data plots.
I don't know if you ever saw the "squared circle" demo. In this one, the
participant's hand (that is, mouse) positions are plotted, and the position
of a controlled spot of light is plotted -- each in its own square window,
in x-y coordinates. After the run, it is seen that the mouse moved in a
circle (with lots of small wobbles but a clear overall circular form). The
spot of light controlled by these mouse movements leaves a trace that can
be a triangle, a square, or any other closed shape. The mouse side of the
screen can be covered during the run.
The secret is in the environmental connnection from the mouse to the spot.
The radial direction of the mouse from the center of its space determines
the direction of movement of the spot. The speed of movement is
proportional to the difference between the mouse's radial position and a
fixed radius. If the mouse is outside this fixed radius, the spot moves
radially outward; if the mouse is inside the radius, the spot moves
radially inward. So the combination of mouse radius and mouse direction in
the mouse's space can be used to place the spot anywhere within its space.
As a result, the participant can (with practice!) draw any desired figure.
But in doing so, the mouse will always be kept close to a particular radius
from the center of mouse space. Thus anything that is drawn will leave a
trace of mouse positions that lies in a band near the radius of a circle --
a complete circle or an arc of the circle.
The point is to show that patterns controlled by behavior do not
necessarily mimic the patterns of action that create them. In this case
there are gross differences; what the hand does does not resemble what the
hand draws. Contemplating this demonstration might help to break the
mind-sets that go with certain other theoretical systems.
I'll mention others as I think of them.
Best,
Bill P.