This topic needs a title

CSGnet Archive CSG2014
81

[From Bill Powers (921017.0900)]

Greg Williams (921017) --

you're adding lines in the diagram that come from the output of a
learned reorganizing system and go to the reference inputs of my
proposed system. Is that what you mean?

I think so. I'm not saying I've got it all mapped out in detail,
but I'm saying it looks like something of this sort is needed to
account for situations such as someone "gladly" killing oneself
"for one's country," (some) "heroes" getting into (perceived by
them as) perilous straits to save other persons, and someone
"choosing death before dishonor." Still, this point isn't on the
mainline of our argument, but a sidetrack which I don't judge of
utmost importance.

This "sidetrack" may be more important than you think. It indicates to
me that you haven't understood my proposal about a reorganizing system
-- that what you're arguing against isn't even what I'm proposing.

My reorganizing system has no reference levels concerning "life" or
"death," or "peril" or "danger" or "survival." Those concepts belong
in the learned systems. I can understand that there can be conflict
between learned goals, such as "safety" and "patriotism," and that
the control system with the highest gain and output capacity will win
the conflict (if any side wins it). But these goals do not conflict
with critical reference levels. They themselves have nothing to do
with critical reference levels. They are learned cognitive goals in
the upper levels of the hierarchy.

Critical variables are related only to actual present-time states of
the organism itself. The critical reference signals likewise have
nothing specific to do with events external to the organism or
relations of the organism to external events or things. The
reorganizing system has no fear of death or hope for survival. It
recognizes neither danger nor safety. It does not even know that food
is a good thing to eat, or that water is a good thing to drink. It is
concerned strictly with the current state of the organism itself, in
terms that can have meaning before any organized hierarchy exists,
before the organism has any perceptions of an "outside world" or a
"body." It will just as readily put the organism in danger as save it
from danger, because it knows nothing of danger in the world outside
the organism.

Consider the soldier going into combat. This soldier wishes to defend
his country, and he also wishes to stay alive. Both of these goals are
learned; they are cognitive goals. The soldier has also learned that
combat could end his life, that bravery is something to be sought, and
that desertion in the face of the enemy is severely punished, that
valor is copiously rewarded. So the soldier wants to go into combat
and he wants not to go into combat. He feels a desire to flee, and a
desire to stay, with the result that he experiences fear and other
emotions. If the degree of conflict is sufficient to cause critical
error, the soldier will begin reorganizing. He may resolve the
conflict by learning to perceive an honorable death as glorious (like
Worf), or to consider it non-threatening, or to imagine that something
protects him from death, or to believe that he will simply awaken into
a better world after death. That would remove the conflict and the
critical error caused by being in conflict. The soldier would then
march to his death without disturbing any critical variables. Of
course the outcome might go the other way; the soldier might
reorganize so that the concept of patriotism is modified or abolished,
leaving the self-preservation side in charge. He will then gladly
desert and suffer the symbolic punishment, and again the critical
error will be corrected.

You can see that I would wonder why you consider it necessary for a
learned system to change a critical reference signal.

I also think that the models one builds must be fully informed by
genuine data on what is being modeled, in addition to meeting
criteria of internal consistency, elegance, and being informed by
data on what might or might not be analogous to what is being
modeled.

What you consider "genuine data" depends on the model you already
believe in. If you come into the discussion thinking that the outside
world can "facilitate" control, then you will interpret the
relationship of the outside world to the organism as demonstrating
"facilitation," something done to the organism in a purposeful way.
If, on the other hand, you think that all changes are internally
motivated and accomplished, you will see the very same actions in the
outside world as having a different meaning in relation to the
organism. You will see the "facilitator" as doing nothing more than
applying disturbances and rearranging the environment, without any
special effect on the organism. You will see all changes in behavior
as resulting from the natural adjustments of the organism to
disturbances and changes in feedback parameters, with the causes of
these changes being entirely internal, beyond influence by anything
external.

This is similar to the difference in the way S-R psychologists and
CTers see "reinforcement." To the S-R psychologist, a reinforcer has
some effect on the organism that alters the way it responds to
stimuli. The underlying model is causal; reinforcers have a special
kind of influence on organisms. So the S-R psychologist can produce
mountains of "genuine data" showing how reinforcers have these effects
on the responses of organisms (defined by their outcomes). The data
seem to support the concept of reinforcement because they are
interpreted in exactly the way needed to make them seem to do that.

The CTer, of course, sees reinforcements as controlled variables. When
the CT theorist looks at the same genuine data, it does not at all
support the idea of a special kind of effect on behavior. Now the data
simply show that the organism produces whatever behavior is required
to keep the so-called reinforcer at a reference level determined by
the organism. The reinforcer is no longer seen as having any special
effect on the organism, other than the effects that any sensory input
has.

Data are "genuine" only when described at a sufficiently low level of
abstraction that there is no disagreement on what is being observed --
where theory doesn't enter into the description. To say anything more
than what was physically done is to bring in theories. So I can say
that I present an organism with a certain environmental situation. I
cannot say, without aid of a theory, that doing this amounted to
"facilitation" or "teaching" or any of those abstract terms that carry
within them a theory of causality, that imply an arrow reaching from
the outside world into the organism and changing something inside the
organism. All descriptions that imply an effect inside an organism
controlled by an external agency are based on the S-R concept of
behavior and the causal model that supports it. All data concerning
such effects, however genuine, can also be interpreted a different way
under control theory.

···

--------------------------------------------------------------------
Best,

Bill P.

82

I ran across an interesting review article on opto-motor
control in the fly, and the research program which is
outlined (of Werner Reichardt) really seems to fit
quite nicely with PCT:

"A look into the cockpit of the fly: visual
orientation, algorithms, and identified neurons."
Egelhaaf & Borst, J. Neuroscience November 1993, 13(11)4563-4574.

I was wondering whether anyone has seen the article or
has comments about the work of the Reichardt group over the
years. It was initially interesting to me because Reichardt's
work (and following it, Bialek's) point to a temporal cross-
correlation theory of motion detection which I think
might also support a similar mechanism for visual forms.

···

_________________

Some comments on falsification. First, I agree with Martin Taylor that
"The whole notion of the "truth" of a theory seems vacuous"
and that no theory is ever proven true or false, in black-and-white
terms. Nevertheless some theories are so
strong and so clearly stated that they lend themselves to clear tests.
I'm not sure that anyone would dispute the notion that we should try
to construct our theories as clearly as possible
and as amenable to probing and empirical testing as we can,
and that we should be fairly relentless in constantly checking our
assumptions. That having been said, there are developmental
problems with a rigid, judgemental interpretation of "falisifiability":
younger, less developed theories often cannot explain everything at
once, so one shouldn't automatically discount (or defund) them because
they don't have the same total explanatory power as an older one. Taken
in this less dogmatic way, with allowances made for power differentials,
I think Popper's call to test theories is basically a healthy one.
It becomes a tool for suppression of new ideas when it is
used to judge which theory should receive funding, publication,
and/or other kinds of support. But these are fundamentally matters
of power anyway -- if power is dispersed and the community values
intellectual honesty over the maintenance of hierarchies, then
criticality will prevail, if not, no amount of gentle urging will
suddenly change the situation. Science, unfortunately, is no
different from any other human institution......

Martin says:
<It isn't the "principle of falsification" that makes critical appraisal
<valuable. In fact, that principle is quite irrelevant since its
<application can provide no information. What critical
<appraisal does, in science as in everyday life,
<is to help define the limits of "reasonable validity"
<for a theory. Every theory is false,
<but likewise every theory has some
<range of validity, even if it be only
<the single set of data that encouraged
<the initial proposal of the theory

What falsification does is it brings the phenomena which are unexplained
to the forefront and forces the scientist to deal with them (by
altering the theory in some way). The information it provides is
an error correction signal, "change what you're doing," when the
results are not as predicted.
I think if you pay attention to those areas where the theory
does not work very well and change the theory accordingly, then you
are more likely to generate better theories in the end than if you
pay attention only to those areas where it does work. How much one
wants to improve the predictive power in any one area is also a
function of how important (economically, in terms of utility) it is
to predict some particular aspect of the world.

There are also differences between those trying to do "normal" science
(looking for incremental increases in predictability) and those trying
to do "revolutionary" science (looking for qualitatively different
explanatory paradigms). Usually those working in areas of "normal"
science favor incremental, accretionary and descriptive models of
scientific progress, while those who believe qualitative change
to be necessary in a given area favor sharp, punctuated, conceptual
models of scientific change (a la Kuhn or Feyerabend). In my
experience, it is the accretionists who adopt gradations of
statistical significance as their criteria, and the "radicals" who
want the new way of looking at things to be obviously right, for the
explanation to "jump out at you".

A new theory can almost never compete with an established one for
descriptive breadth (there simply aren't the resources to do as
many experiments or show as many implications),
but it often can compete conceptually on a more
equal footing. My guess is that a relative newcomer like perceptual
control theory will do better in an environment where the scientific
community has the second conception of scientific progress.

Peter Cariani

83

[From Bruce Abbott (951008.1230 EST)]

Bill Powers (951008.0815 MDT) --

Dammit, why do you always have to be so @#$%&*)! reasonable?

Personality flaw. Gets me in trouble all the time: I argue that the "other
side" (whomever that may be at the moment) may have a point and people say
"dammit, we though you were on OUR side!" Makes it hard to keep friends.

Now, are you going to say, "yes, but it's still USEFUL to gather these
empirical facts"?

But of course! This despite the fact that I agree with your analysis, which
elaborates my brief statement about MSH possibly affecting the control
system involved at just about every point from perceptual input to gain.
Yet because of this study we now at least know that MSH does _something_ to
this system so as to produce DVs when normally the visual feedback would
have reduced or eliminated them, we know the time course of this effect, and
we know its dependency on dose. That's still not much, but this was only an
initial "screening" study to determine whether MSH had any detectable effect
on the system. By itself it's not much, but this information does not
exhist in a vacuum. MSH is somehow modulating the control system involved
in keeping the chick in proximity to its mother and broodmates. Next we do
an anatomical mapping study to determine where those MSH receptors are
located in the brain. Perhaps we also have information about the
neuroanatomical subtrate of the system that produces DVs, gained through
electrode stimulation and lesioning studies. Our mapping study shows that
the MSH receptors concentrate heavily in just those areas where our "wiring"
studies show some of the important neural components of the system to lie.
We still don't have much, but I contend that we've made progress. We're
starting to nail down the "circuit diagram," the anatomical map of the
control system whose actions typically produce DVs (and other outputs) in
young chicks when separated from their mothers and the brood.

Thousands of different brain operations employ similar chemistry. I
would guess that the main evolutionary reason for the existence of
different neurotransmitters is to reduce crosstalk between separated
systems, or, conversely, to provide information links between closely-
associated systems. But the particular brain chemicals, I strongly
suspect, have very little to do with the kind of function that is being
carried out. Brain chemicals are simply carriers of signals; the
particular chemical used as such a carrier is optional. So the fact the
MSH seems to be associated with a particular carrier is no more
significant than the fact that boron happens to be used to dope a
particular transistor to make it a semiconductor, or the fact that in
one transistor electrons carry the signal while in another, holes carry
the signal.

Actually, there is good anatomical and functional evidence that certain
systems have evolved to perform certain definite functions. For example,
endorphines (endogenous morhpines) found in neural terminals in the central
gray of the midbrain and in portions of the spinal cord reduce sensitivity
to pain when released onto their target neurons. These same biochemicals
appear to serve a different but possibly related function in the "old
cortex" where their release into target neurons there may promote a
subjective state of relative tranquility related to social-emotional
processes like bonding and relief from the "pain" of separation from loved
ones. Because these biochemically coded systems emerged in the context of
specific functional mechanisms, their specific roles may have been conserved
and only gradually elaborated over the course of evolution. Thus one would
expect that neuronal tracts differing in their neurotransmitters would be
distinguished by their functional role, although certainly the use of
different neurotransmitters would have the additional benefit of minimizing
cross-talk.

Because MSH modulates the chick's separation/proximity system, we can
hypothesize that MSH is one of the "carriers" of the signals that modulate
this system. With appropriate histological methods, we can visualize those
neurons that participate in this signal and thus map a part of the circuit
involved in the chick's control of proximity to its mother.

I strongly agree with you that mere knowledge of the effect of one
biochemical or another on some global behavior like DVs is not very useful.
Analyzing the situation from a control system perspective shows why, and
without the guidance provided by that analysis it is unlikely that one would
arrive at a correct understanding of the underlying mechanism, even given a
fairly detailed anatomical map of the brain circuits involved. The PCT
perspective tells you what to look for, what to expect. On the other hand,
simply performing the Test for controlled variables is not going to identify
the brain systems through which control of those perceptual variables is
achieved. The two lines of research are complementary, not antithetical.
Each has something to contribute to the other.

Regards,

Bruce

84

[from Joel Judd 951024.0750 CST]

(Bruce Abbott 951021)

<< I apologize for taking a couple of days in response--since this thread
was to be "dropped," but I just got the last four days' messages all at
once this morning >>

This statement reopened an offending wound I received a couple years ago
when submitting a discussion of PCT and Second Language Acquisition (SLA)
for publication:

Science
is concerned with establishing general principles (within which specific
cases can be understood, to be sure).

Two of the three reviewers responded to my paper in part:

"The main problem with this paper is it criticizes SLA research for
something most of it is not intended to do and does not generally try to
do--explain the individual."

and

"I do not think it is the basic aim of SLA research to study how a
particular person learns a language. I rather think it is the aim of our
field to specifiy the general laws/[rinciples/regularities of SLA."

The second commentator goes on to suggest that individual behavior is a
particular instantiation of "SLA principles," but then sinks this thought by
saying "we may know the general principles without being able to predict
their application in the specific case."

I found it extremely depressing to realize that I was in a field where
many--particularly the gatekeepers--were content to believe in a science
that applied to EVERYBODY and NOBODY. This reminded me of the catechism
that God is everywhere and nowhere.

These comments I shared with Philip Runkel, and one thing he said makes me
smile every time I remember it. He asked, rhetorically, how I would like to
be in a plane at 30,000 feet with a pilot who knew general principles of
aviation but not how to apply them to our particular plane in our particular
circumstances? Or take my car to a mechanic who knew general principles of
combustion but not how they applied to my particular make and model? It
seems to me a waste of time to knowingly spend effort develping principles
that don't lead to specific applications.

The catch, I think, is that when we talk of "general principles" we must
still be clear about what KIND of general principles we're talking about.
We will only be able to understand "specific cases" if we develop
principles that arise from study of individual, specific cases to begin
with. If our general principles are basically statistical in nature, then
application to specific cases may only be a poor probability, and even then
will provide no answer to the "why" of the behavior.

Part of the rub in developing general *control* principles is that
particular BEHAVIORS may not be the understanding or explanation we glean
from study of specific cases. The psychological explanation implied by PCT
is CONTROL OF PERCEPTION, not BEHAVIORAL CONTROL. Control principles are
what PCT provides. Particular behaviors will only be predicted inasmuch as
we know the effects of the disturbances on controlled variables and the
avenues for reducing error available to the organism. If the only avenue
available to me to get food is bashing my head on a lever, and I'm hungry,
you can feel safe wagering I'm going to hit the lever. There is always the
chance, however, that I can say "Screw it" and go sit in a corner. Then
what happens to the probability I will hit a lever to obtain food?

Those in a position to benefit from control of behavior (politicians,
lawyers, teachers, businessmen, psychiatrists, and so on) are often those
least impressed by a science of autonomous control. If your livelihood is
tied to academic test outcomes, which research program would seem attractive
to you: one which promotes the idea that we're getting better at creating
environments where student outcomes will be more guaranteed, or one which
suggests that students learn to reduce perceptual error, and that rewarding
particular overt behaviors does not address the actual mechanisms of
learning?

Joel Judd

85

[From Bruce Abbott (951026.0945 EST)]

Hans Blom, 951026 --

We, educated in control theory, have a deeper insight into
differences between these two modes of "control". Many discussions
have been fought about terminology, e.g. "control" vs. "influence".
But my point was that in a "causal loop" it is not fruitful to view
one variable as independent and another one as dependent. But if you
do so, it is not important _which_ one you take as the independent
variable.

And by only focussing on PCT's "causality" one runs the risk of
missing the great extent to which people's actions are determined by
changes in the environment. Seen this way, reinforcement theory is
PCT's other side of the coin. Both formally and factually :-).

Well put, Hans. Well put.

Regards,

Bruce

86

[From Dag Forssell (961006 0915)

I received the following from Clark McPhail. Thought it belongs on the net:

···

---------
Dear Dag:

Thanks for your prompt delivery of the Van De Rijtt/Plooij volume. I look
forward to reading it. From a quick "thumbing" of the book it appears they
have managed the most unobtrusive and non-threatening introduction yet in
print to the formerly formidable "levels of perception". I'm also struck
by the total absence of footnotes, references to their own scientific
journal publications or to the work of WTP. Now if someone can just
figure out the corresponding trojan horse for introducing HPCT into the
behavioral science journals. . . .

Clark
----------

I have given copies of "Why They Cry" to a few people. As I do, I make a
point of suggesting that it is very easy to read (and thus not grasp the
full significance of) the book as a description of successive stages of
"growth." I suggest that a person should read the book not as a time-line,
but as an outline of a layer cake -- yourself. The book tells you about how
your own mind is organized at this time, not just about the time phazes of
some baby's development.

Best, Dag

87

[From Bill Powers (961020.0430 MDT)]

An interesting domestic-affairs phenomenon. I used some of that earwax gunk
and succeeded in totally plugging up my ears, drastically attentuating all
the high frequencies above about middle C. Then I emptied the dishwasher,
and discovered that all the saucers and plates had acquired some sort of
soft rubber coating. The way they felt when they bumped into each other as I
stacked them was completely different. The reason? No clink! I suppose I
would have adapted my perceptions sooner or later, but this sudden change
left me with a different set of perceptions. I can't think of a neater
illustration of how perceptions at higher levels make use of all inputs,
from any sensory modality, that are relevant to controlling them.
Unfortunately I can't think of a computer demo that will convey this experience.

Erling Jorgensen (961020.0115 CDT --

I appreciate the discussions that have been going on regarding emotion,
and found this post a really helpful synthesization. I, too, have had
the sense that the various positions of Bruce and Martin and you don't
have to be at odds. Even Hans' notion of the ionic requirements of
cellular control systems is not that far away from the "intrinsic
variables" of a reorganizing system.

The BIG difference between Hans' position and mine is that he wanted the
cellular level to be superordinate to the behavioral systems.

Martin Taylor and I have more or less converged to an agreement ...
What is needed is _local_ reorganization that is specific to the
control system that is having trouble

Thanks for reiterating this. I wasn't sure what the current thinking
was about locus of action.

Actually it's not "the" locus, but "a" locus. We still need the basic
non-specific reorganization idea. The whole concept of reorganization as a
function of physiological error signals came up because I needed a way for
somatic error signals to lead to learning. Why will a pigeon learn to walk
in a figure eight just because it's hungry? We can't assume that evolution
could have prepared pigeons for the possibility that B. F. Skinner would
make eating contingent on walking in some particular pattern for one or two
randomly-selected members of that species. Organisms have to be able to
learn to control complex variables which have an _arbitrary_ relationship to
their well-being. The power of Ashby's concept of superstability, which is
the basis of my reorganization principle, is that it can lead to control
without the need for ANY knowledge of the properties of the environment. Why
does it hurt when I push this button? Who cares? I can still learn not to
push it.
The only alternative to this generalized kind of reorganization that I can
see is for organisms to be born with a complete knowledge of classical
physics, chemistry, geology, astronomy, and so on, as well as the ability to
predict what B. F. Skinner would want them to do.

So we need the basic kind of reorganization as well as the more specific
kind. Obviously, there's a lot we don't know about how these various forms
of reorganization work together.

Perceptual reorganization could be the back-up system, triggered when
output reorganization wasn't doing the job, and thus when error signals
were sizable and prolonged.

That makes considerable sense, considering that in the PCT style of control,
the exact organization of the output function is not important, while the
organization of the input function critically determines what is being
controlled. In my computer experiments with multiple-system reorganization,
I either picked a random set of output connections to the environment and
reorganized the input function, or a random set of input weightings and
reorganized the output function. Either approach works pretty well. But I
wasn't able to make any sense of what happened when I reorganized BOTH the
input and output functions. My cut-and-try methods have some pretty serious
limitations; what we need here is someone who can take this whole problem in
hand and do a systematic mathematical analysis of what's required, as has
been done for perceptrons and neural nets. Once upon a time I could get
myself into a strange mental state in which I could just slog through an
analysis step by step until it was done, keeping track of subscripts by
writing everything out in detail and so on, but those days are past. We need
someone who can stay in that strange mental state all the time -- AND who is
interested in PCT.

Either you or Bill Leach brought up at one point that the lack of a
reference signal was functionally equivalent to a reference signal of
zero. The first consequence of such an arrangement would be that new
perceptions would be driven back toward a zero reference, and thus would
tend not to be seen. This seems to be the difficulty we have in
perceiving things in novel or unfamiliar ways.

This is an important aspect of acquiring new levels of control. When a new
level is being developed, there are no higher systems to give them nonzero
reference signals. So any reference signals must come from memories of
recent experiences (imitation) or random noise (experimentation), or they
must be zero. A reference signal of zero tells the system to behave in such
a way as to bring any nonzero perception of the associated kind to zero. We
call that "avoidance" and think of the source of the perception as "aversive."

I think you can tell exactly when a baby becomes able to distinguish faces.
The baby starts trying to prevent perceptions of faces. All the baby is
trying to do is to restore the perception of the face with which it is
familiar, or reject perception of a face it can recognize but for which it
has no non-zero reference level. Of course it can do little to accomplish
this but look away and send out signals that something is happening that it
doesn't want to happen. The folk interpretation is that the baby has
developed "fear of strangers." I believe the Plooijs identify this as the
stage of learning to control configurations. After the baby has experienced
the "stranger" for a while, there are memories from which some non-zero
reference signals can be derived, so the "fear" goes away. Provided nothing
bad has happened when the stranger is repeatedly perceived.

So the first requirement for perceptually reorganizing and "seeing things
in a new light" (which just might result in intrinsic error levels
decreasing) would seem to be a _non-zero_ reference, to allow new
perceptions to be sustainably seen in the first place.

The very first thing that has to happen is that the perceptual function
appears. Without a perceptual function and a perceptual signal there is
nothing to control. According to my memory postulate, reference signals are
actually derived from memories (at least in some higher systems), and before
there can be any memories there must be perceptions to be remembered. Could
reference signals come first? It would be hard to see how, since they don't
have any meaning until the input function and comparator exist. And a
comparator has nothing to do until a reference signal and a perceptual
signal exist. The output function has to come last, because it needs an
error signal to drive it. Of course anywhere in this process there could be
built-in connections to get the process started or make it converge faster.
Lots of unanswered questions here.

I don't know
enough about neuroanatomy and plasticity to know how quickly neurons
generate (or atrophy) their connections with neighboring cells, but
from what I recall it's common among developing organisms.

I read not too long ago about adult animals showing microscopically
observable changes in synaptic connections in the brain over intervals of
five minutes.

Is this consistent (or not inconsistent) with what you've been thinking?

If you're good enough to call it thinking.

Best,

Bill P.

88

Bill Powers (961020.0430 MDT) sez:

Perceptual reorganization could be the back-up system, triggered when
output reorganization wasn't doing the job, and thus when error signals
were sizable and prolonged.

That makes considerable sense, considering that in the PCT style of control,
the exact organization of the output function is not important, while the
organization of the input function critically determines what is being
controlled.

1. Well Bill, I really can't resist. If the input function is so critical,
how can you possibly justify your black box attitude about how those
functions operate? You can tell me about how obscure the brain is, etc.
But I'm not buying it. Yes, the literature is a mess and the brain is
complex etc. But you know perfectly well that that literature is written
by people who are professionally blind to robust systems thinking.
Expecting those folks to figure out what the brain is up to is like
expecting an auto-mechanic, even a very good one, to design a new
automobile. If you or others really go at that literature as it exists,
you might be able to see order in it which is invisible to those creating
the literature.

2. Consider the notion that hunger is a learned perception. It is on the
input side of the systems which regulate food-seeking and eating (FSE).
The hunger perception gets reorganized into the system so that FSE can be
initiated before intrinsic error happens and thus eliminate it. So, over
time, FSE shifts from happening subsequent to intrinsic error to happeneing
subsequent to hunger.

This is an important aspect of acquiring new levels of control. When a new
level is being developed, there are no higher systems to give them nonzero
reference signals. So any reference signals must come from memories of
recent experiences (imitation) or random noise (experimentation), or they
must be zero. A reference signal of zero tells the system to behave in such
a way as to bring any nonzero perception of the associated kind to zero. We
call that "avoidance" and think of the source of the perception as "aversive."

Gerald Edelman made a suggestion as long ago as 1978 which might be recast
and put to use here. Imagine we have a bunch of control units which are
innately wired to respond to a wide range of arbitrary inputs. It doesn't
matter just what those inputs are. What's important is that we have a
large range of them and that each unit has a strong preference for one or a
few such patterns while be only weaking or not at all responsive to the
rest. When a sheet of such units is exposed to real inputs it will be able
to make some valid discriminations among those inputs and those initial
discriminations can be finely tuned under reorganization.

Note that one of the more stunning findings of neuroscience in the last 20
years is the massive "over-production" of connections during development
and their subsequent "pruning" under the influence of environmental input.
It seems that the genes wire all the connections they can and those axons
and dendrites whose synapses aren't reinforced (through Hebbian learning?)
simply degenerate.

See BP immediately below:

The very first thing that has to happen is that the perceptual function
appears. Without a perceptual function and a perceptual signal there is
nothing to control. According to my memory postulate, reference signals are
actually derived from memories (at least in some higher systems), and before
there can be any memories there must be perceptions to be remembered. Could
reference signals come first? It would be hard to see how, since they don't
have any meaning until the input function and comparator exist. And a
comparator has nothing to do until a reference signal and a perceptual
signal exist. The output function has to come last, because it needs an
error signal to drive it. Of course anywhere in this process there could be
built-in connections to get the process started or make it converge faster.
Lots of unanswered questions here.

Bill Benzon

···

********************************************************
William L. Benzon 518.272.4733
161 2nd Street bbenzon@global2000.net
Troy, NY 12180 Account Suspended
USA
********************************************************
What color would you be if you didn't know what you was?
That's what color I am.
********************************************************

89

Whew!

“The Occupy movement is powerful, not because it is fighting for the rights of a few hundred people to sleep outdoors, but because it is fighting for the right of millions of Americans to sleep indoors.” - Van Jones

image00145.jpg

···

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Bill Powers
Sent: Wednesday, November 30, 2011 2:10 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject:

Hello, everybody –

You will all be relieved to know that Bill, his daughter Allie, and his granddaughter Sarah, though terrified, safely escaped from a giant control system while on a recent trip to downtown Denver.

            <img width="512" height="366" id="_x0000_i1025" src="cid:image001.jpg@01CCAF6C.BCC35B30" alt="Emacs!">

Best,

Bill P.

90

You are indeed quite fortunate. Several behavior modification specialists were eaten while trying to condition that rascal.

Fred Nickols

image00145.jpg

···

From: Bill Powers [mailto:powers_w@frontier.net]
Sent: Wednesday, November 30, 2011 2:10 PM
To: CSGNET@listserv.uiuc.edu; warren.mansell@manchester.ac.uk; wmansell@gmail.com; sara.tai@manchester.ac.uk; jrk@cmp.uea.ac.uk; hy43@duke.edu; Sergio.VerduzcoFlores@colorado.edu; Brian.Mingus@colorado.edu; randy.oreilly@colorado.edu; Lewis.Harvey@Colorado.EDU; Tim.Carey@flinders.edu.au; steve.scott@queensu.ca; mcclel@grinnell.edu; marken@mindreadings.com; dag@livingcontrolsystems.com; fred@nickols.us; mmt@mmtaylor.net
Cc: songsofmu@gmail.com
Subject:

Hello, everybody –

You will all be relieved to know that Bill, his daughter Allie, and his granddaughter Sarah, though terrified, safely escaped from a giant control system while on a recent trip to downtown Denver.

            <img width="512" height="366" id="_x0000_i1025" src="cid:image001.jpg@01CCAF6D.DD6FF5E0" alt="Emacs!">

Best,

Bill P.

91

[From Bill Williams UMKC 12 November 2002 7:30 AM CST]

Several days ago I mentioned a relationship between price changes and changes
in quantity for a Giffen good. When a succession of equal price increases are
applied to the Giffen good the result is not an equal change in the quantity
purchases. Instead as the price increases go on there is an increase in the
ratio between the price increase and the change in quantity of the giffen good
which is purchased. Further, using the graphic depiction of the Giffen
situation it can be predicted that this effect is more extreme the closer the
consumer is to being able to survive on the more expensive source of calories.
The effect is illustrated in the attached program PGS.EXE ( Price Giffen Shift )

On the first page of the illustration the consumer is close to being able to
survive on the more preferred good-- and the change in reaction to sequence of
price increases is quite marked. In the second case the consumer has a much
lower income and the increase in response to a succession of price increases,
while it still occurs, is less dramatic. Does anyone think this could be
experimentally confirmed by an animal study?

You'd mentioned the relationship between poor nutrician and obesity several
years ago, but I'd forgotten it. There's a paper Sclarfani, A. 1973 "Feeding
Inhibition and Death Produced by Glusose Ingestion in the Rat" Physiology and
Behaivior vol 11 p 595-602. that may describe a related phemoena. According
to the study Rats over estimate how many calories glusose contains and
consequently starve to death.

The next few days I will try to concentrate on preparing a talk on how control
theory can be applied to problems in economic theory, rather than CSGnet stuff.
Something of the problem I'm facing can be illustrated by a minor conflict I've
been having with a guest. THe guy is a German fulbright visiting professor. In
the past I've thought of Germans in terms of the steriotype of a
technologically proficient people. However, this guy has what I think of as
some strange ideas about what a thermostat is for. He sets the thermostat to
either fifty or eighty-five-- that is full scale on or off. I've tried to
explain to him how a thermostat functions, but he considers what I'm trying to
tell him to be "excessively technical." He knows there is something more to a
thermostat than an on-off switch, but as far as he is concerned it obviously
works as an on/off switch, so what could be the point of making things more
complicated? My objection is that when the apartment temperture goes to 85
degrees my room seems to go over a hundred. As an example of one time learning
after I woke up in the middle of dream about north africa I blocked the heater
duct in my room and now my room temperature is very livable. Any, suggestions,
that is _constructive_ suggestions as to how to go about explaining control
theory to an audience part of which considers a thermostat "excessively
technical."

Best

Bill Williams

PS. Rick, I thought we were engaged in a SCIENTIFIC discussion on our brand new
thread. But, if you want to flick off one of your scabs, and you want some help
from me in doing so ( even though I'd just as soon not go there) we can resume
play a bit later. But, this week I've got to pay attention for the time being
to trying to figure out how to explain what control theory could do for
economic theory. With guys like my German guest, it isn't easy, I'm not sure
its even possible.

PQS.EXE (36.7 KB)

···

______________________________________________________________________
Do you want a free e-mail for life ? Get it at http://www.email.ro/

92

[From Rick Marken (2002.11.12.0920)]

Bill Williams (UMKC 12 November 2002 7:30 AM CST) --

Several days ago I mentioned a relationship between price changes and changes
in quantity for a Giffen good. When a succession of equal price increases are
applied to the Giffen good the result is not an equal change in the quantity
purchases.

If you run my spreadsheet model you will see that this effect is, indeed, produced. So
your prediction, based on the graphical analysis, is correct. You can also see this
non linearity in the experimental demo of the Giffen effect
(http://home.earthlink.net/~rmarken/demos.html). Consumption of bread rises
exponentially as the price of bread increases, at least when you are acting as the
"Poor Man".

Does anyone think this could be
experimentally confirmed by an animal study?

Yes. And it already has been experimentally confirmed with humans in the demo described
above.

PS. Rick, I thought we were engaged in a SCIENTIFIC discussion on our brand new
thread.

I think we are. Don't you?

But, if you want to flick off one of your scabs, and you want some help
from me in doing so ( even though I'd just as soon not go there) we can resume
play a bit later.

Say what? Where did this come from?

Best regards

Rick

···

--
Richard S. Marken, Ph.D.
The RAND Corporation
PO Box 2138
1700 Main Street
Santa Monica, CA 90407-2138
Tel: 310-393-0411 x7971
Fax: 310-451-7018
E-mail: rmarken@rand.org

93

[From Rick Marken (2002.11.12 1100)]

Bill Powers (2002.11.12.1044 MST)

Bill Williams UMKC 12 November 2002 7:30 AM CST --

>Several days ago I mentioned a relationship between price changes and changes
>in quantity for a Giffen good. When a succession of equal price increases are
>applied to the Giffen good the result is not an equal change in the quantity
>purchases. Instead as the price increases go on there is an increase in the
>ratio between the price increase and the change in quantity of the giffen
>good which is purchased.

I did a quick check with the Giffen2 program I just distributed.

Meat price = 5.00:
  A 33% increase in bread price ($1,00 to $1.33) results in an 8.2% rise in
bread consumption.

Meat price = 2.57:
  A 33% increase in bread price results in a 26% rise in bread consumption.

I thought Bill W. was referring to the change in consumption of bread as a
function of bread price with meat price held constant. In my model I get these
results:

Meat price = $1.00:

A 50% increase in the price of bread, from $.10 to $.20, results in a 10% increase
in consumption of bread (from 223 to 248). A 50% increase in the price of bread,
now from $.20 to $.40, results in a 25% increase in consumption of bread (from 248
to 331).

I think Bill W. may have been referring to both effects; and both are, indeed,
produced by a control model.

You said:

This leads me to formulate the Basic Principle of Capitalism:

Sell the most goods possible,
Of the lowest possible quality,
At the highest possible price.

I offer one empirical counter to the second principle: CostCo! They carry the
highest quality stuff and their chicken and ribs are to die for! Several very
rich people I know are willing to endure the hordes of hoi polloi to get their
stuff from CostCo. Once we get the Giffen effect nailed down I suggest we work on
the CostCo effect; that will be the _real_ challenge for control theory;-)

Best

Rick

···

--
Richard S. Marken, Ph.D.
The RAND Corporation
PO Box 2138
1700 Main Street
Santa Monica, CA 90407-2138
Tel: 310-393-0411 x7971
Fax: 310-451-7018
E-mail: rmarken@rand.org

94

[From Bill Powers (2002.11.12.1044 MST)]

Bill Williams UMKC 12 November 2002 7:30 AM CST --

Several days ago I mentioned a relationship between price changes and changes
in quantity for a Giffen good. When a succession of equal price increases are
applied to the Giffen good the result is not an equal change in the quantity
purchases. Instead as the price increases go on there is an increase in the
ratio between the price increase and the change in quantity of the giffen
good which is purchased.

I did a quick check with the Giffen2 program I just distributed.

Meat price = 5.00:
  A 33% increase in bread price ($1,00 to $1.33) results in an 8.2% rise in
bread consumption.

Meat price = 2.57:
  A 33% increase in bread price results in a 26% rise in bread consumption.

Don't know where that would show up on your graphical plot, but the change
is there. Is the direction right?

Best,

Bill P.

···

Further, using the graphic depiction of the Giffen
situation it can be predicted that this effect is more extreme the closer the
consumer is to being able to survive on the more expensive source of calories.
The effect is illustrated in the attached program PGS.EXE ( Price Giffen
Shift )

On the first page of the illustration the consumer is close to being able to
survive on the more preferred good-- and the change in reaction to sequence of
price increases is quite marked. In the second case the consumer has a much
lower income and the increase in response to a succession of price increases,
while it still occurs, is less dramatic. Does anyone think this could be
experimentally confirmed by an animal study?

You'd mentioned the relationship between poor nutrician and obesity several
years ago, but I'd forgotten it. There's a paper Sclarfani, A. 1973 "Feeding
Inhibition and Death Produced by Glusose Ingestion in the Rat" Physiology and
Behaivior vol 11 p 595-602. that may describe a related phemoena. According
to the study Rats over estimate how many calories glusose contains and
consequently starve to death.

The next few days I will try to concentrate on preparing a talk on how control
theory can be applied to problems in economic theory, rather than CSGnet
stuff.
Something of the problem I'm facing can be illustrated by a minor conflict
I've
been having with a guest. THe guy is a German fulbright visiting professor. In
the past I've thought of Germans in terms of the steriotype of a
technologically proficient people. However, this guy has what I think of as
some strange ideas about what a thermostat is for. He sets the thermostat to
either fifty or eighty-five-- that is full scale on or off. I've tried to
explain to him how a thermostat functions, but he considers what I'm trying to
tell him to be "excessively technical." He knows there is something more to a
thermostat than an on-off switch, but as far as he is concerned it obviously
works as an on/off switch, so what could be the point of making things more
complicated? My objection is that when the apartment temperture goes to 85
degrees my room seems to go over a hundred. As an example of one time
learning
after I woke up in the middle of dream about north africa I blocked the heater
duct in my room and now my room temperature is very livable. Any,
suggestions,
that is _constructive_ suggestions as to how to go about explaining control
theory to an audience part of which considers a thermostat "excessively
technical."

Best

Bill Williams

PS. Rick, I thought we were engaged in a SCIENTIFIC discussion on our
brand new
thread. But, if you want to flick off one of your scabs, and you want some
help
from me in doing so ( even though I'd just as soon not go there) we can
resume
play a bit later. But, this week I've got to pay attention for the time being
to trying to figure out how to explain what control theory could do for
economic theory. With guys like my German guest, it isn't easy, I'm not sure
its even possible.

______________________________________________________________________
Do you want a free e-mail for life ? Get it at http://www.email.ro/

95

[From Bill Powers (2002.11.12.1509 MST)]

Rick Marken (2002.11.12 1100) --

>I thought Bill W. was referring to the change in consumption of bread as
a >function of bread price with meat price held constant.

He was, but held constant at two different meat prices. When the meat price
is relatively low (so the person could almost afford to live on meat
alone), bread consumption is highly sensitive to bread price. When meat
price is higher, the sensitivity of bread consumption to bread price is
lower. At least that's how I interpreted his graphs. I think I saw the same
relationships in the program.

Best,

Bill P.

···

In my model I get these
results:

Meat price = $1.00:

A 50% increase in the price of bread, from $.10 to $.20, results in a 10%
increase
in consumption of bread (from 223 to 248). A 50% increase in the price of
bread,
now from $.20 to $.40, results in a 25% increase in consumption of bread
(from 248
to 331).

I think Bill W. may have been referring to both effects; and both are, indeed,
produced by a control model.

You said:

> This leads me to formulate the Basic Principle of Capitalism:

> Sell the most goods possible,
> Of the lowest possible quality,
> At the highest possible price.

I offer one empirical counter to the second principle: CostCo! They carry the
highest quality stuff and their chicken and ribs are to die for! Several very
rich people I know are willing to endure the hordes of hoi polloi to get their
stuff from CostCo. Once we get the Giffen effect nailed down I suggest we
work on
the CostCo effect; that will be the _real_ challenge for control theory;-)

Best

Rick
--
Richard S. Marken, Ph.D.
The RAND Corporation
PO Box 2138
1700 Main Street
Santa Monica, CA 90407-2138
Tel: 310-393-0411 x7971
Fax: 310-451-7018
E-mail: rmarken@rand.org

96

from Phil Runkel on 15 Nov 99
with a question about GAIN.

I have long felt that my understanding of gain in the loop is
inadequate.

I remember that old-time "public address" systems (amplifiers)
had a knob labeled "gain." Radio sets had a knob labeled
"volume" that seemed to do the same thing. But I thought the
function in the radio set came from circuitry something like a
transformer connected so that the radio signal could be copied
onto the audio signal, the latter being supplied by much higher
voltage (and amperage?) than the former.

I understand the asymmetry of control (so I claim), but I am
hesitant about gain. Are we using the term to label the ratio of
(a) the energy (?) exerted by a batter whaming a baseball to (b)
the energy (?) brought to the retina by the light from the
baseball as it approaches the batter?

97

Hi, Phil --

I really need to write something on the subject of gain in the loop, and on
the properties of negative feedback in general. It's clear to me that
nobody in other camps has the slightest understanding of how negative
feedback works, and the entire beauty and uniqueness of PCT lies in that
astonishing phenomenon. If I could only get that across the battle would be
won.

Gain is, as you say, simply amplification. In applications like the audio
amplifier you speak about, you would expect that turning up the
amplification would produce more audio output. But when that gain exists in
a negative feedback loop, that isn't what happens. As you raise the
amplification, everything remains more or less the same, but the error
signal gets smaller.
The increased amplification is seen only in the increased stiffness of the
system when disturbances arise: it takes less error to produce enough
action to counteract the disturbance. If the action was previously enough
to counteract 90 percent of the disturbance, now, after a tenfold increase
in the amplification, it is enough to counteract 99 percent of it. But the
action itself, from outside the system, hardly changes at all. If it was
counteracting 90 percent of the disturbance and is now counteracting 99
percent of it, the action has changed by only about 10%. You have to look
closely to realize that what was a somewhat lackadaisical act of control
has now tightened up and is humming with sensitivity to every little change
in the controlled variable.

Bill

98

[From: Bruce Nevin (Fri 931119 14:00:59 EST)]

Rick Marken (931119.0900) --

The perception of a category is present whenever I imagine controlling
some perception without imagining precisely which of diverse possible
means I use for the purpose. "Something to pry this cover up with."

I do not see a need for a separate category level.

What am I missing?

I think you are descibing the categorical nature of perception
(imagined perception in your example) -- not the perception of a
category. I think that's the difference. You can pick up a knife
and pry with it becuase you've pryed with knives before. But you
cannot look around for "something to pry with" until you can
perceive many lower level perceptions as the same thing -- "some-
thing to pry with".

I think that the appearance of "perception of a category" is due to
nothing more than the existence of a label for a category, and the
perception that the label (another perception) is associated with the
category and can "stand for" it in social communication, real or
imagined. The map is not the territory (Korzybsky, yes, Martin); the
label is not the category. It appears to me that the "categorical nature
of [the] perception" is the category, or constitutes it.

I see no evidence that there is any "perception of a category," in the
nicely distinguished sense that you and Bill have discussed, without
category labels or symbols. Work with primate-human communication has
shown that primates can learn and use symbols in social communication; it
is not clear to me that they do so in naturally occurring communication
with primate peers, much less in their asocial category perception. Is
there evidence of "perception of a category" apart from either
label-association (must be some better term than that) or the
"categorical nature of perception"?

You can pick up a knife
and pry with it becuase you've pryed with knives before. But you
cannot look around for "something to pry with" until you can
perceive many lower level perceptions as the same thing -- "some-
thing to pry with".

Au contraire, all I have to do is imagine prying (imagine doing something
that reduces error in the perception I am trying to control), and
specifically to imagine prying with this object or that as it comes to my
attention in my environment or in my imagination. I don't have to have a
pre-fab PIF such that I perceive many lower level perceptions as the same
category, "things that one can pry with". I find the notion of
accumulating so many ad hoc PIFs difficult to credit, and at variance
with the facility with which ad hoc new categories seem to come into
being--precisely as many as I have controlled perceptions, it seems.
Instead, I suggest that the many-one relation is established through
active imagining of using this or that to pry with.

Labels and especially words help us calibrate the many-one mappings in
the "categorical nature" of those of our perceptions where social
coordination is important for our control. As we know words and labels
by no means guarantee that we arrive at the same mappings, but they do
help, because in the course of communicating with words we continually
put our categorizations to the test: Are we in agreement? (Which we are
apt to perceive as "have I got it right?" -- ref here to pique post.)
The category labels are social realities of great importance to us, and
it is important to us to get the many-one mappings which they label
"right", that is, in agreement with those with whom we must cooperate in
order to accomplish our aims. Perhaps because of this existential
burden, people often identify category labels with categories, and feel
that the category exists independently of the label and the social
process of mutual calibrating of the "categorical nature" of our
perceptions. When we do that, it appears that there is a "perception of
category" apart from the "categorical nature of perceptions." But I
think that's all there is to it.

    Bruce
    bn@bbn.com

99

[From Bruce Abbott (951102.1505 EST)]

Rick Marken (951102.1100) --

Bruce Abbott (951102.1115 EST)

EXPLAINING THORNDIKE'S OBSERVATIONS

Slow down! We don't know what we're observing yet. That is, we don't know
what variables the cat is controlling. Before we start trying to explain
Thorndike's observations, we have to determine what those observations tell
us about the purposeful behavior of the cat. It looks to me like they tell
us almost nothing about it at all -- so PCT based explanations of these
observations are little more than just-so stories.

First, I agree that the empirical work identifying the controlled variables
in Thorndike's experiment needs to be done. But that shouldn't prevent me
from thinking about what might be happening there, and trying to develop
possible explanations that might be consistent with PCT. Your opinion--that
the observations tell us almost nothing at all--is just that, your opinion.
You're entitled to it, but I'm entitled to mine, too, and I disagree, I
think they tell us plenty. Of course, finding out which of us is right will
require research. It is, after all, an empirical question.

I heard a lot of speculation similar to mine coming from Bill Powers, who
began his post with this:

A nice essay on Thorndyke and his cats. Let me try my own interpretation
using the biases of PCT.

Funny, I didn't hear you telling Bill to slow down. Perhaps you can explain
why you didn't.

You seem to be rather stubbornly avoiding my question about how much space
you devote to The Test for the Controlled Variable in the new edition of your
Research Methods text.

It's irrelevant and you know it. Would you like a complete list of all the
methods designed to provide specific kinds of information that I don't cover
in my text? Well, I don't have time to cover them all (there's so MANY of
them!), so I'll just offer a short sampler:

Methods not covered

···

-------------------
How to determine the:
  absolute threshold for a sensation
  threshold for the difference between two sensations
  size of an illusion
  controlled perception
  D' and Beta in signal detection
  duration of short-term memory
  serial-position curve
  magnitude of the Stroop effect
  optimal interstimulus interval in classical conditioning
  I.Q.
  simple reaction-time
  flicker-fusion frequency
  onset and duration of REM sleep

I could keep this up all day. Now, what do all these methods have in
common, besides the fact that they aren't mentioned in my text?

Regards,

Bruce

100

(From Samuel Saunders [951108:17:30:43 EST])
Bill Powers (951108.middle of the night)
   responding to Chris Cherpas (951107.1734 PT)

Duh. Random responding on Concurrent VIs would not produce
matching, my friend. It would produce the worse case of
undermatching that anybody had ever seen.

Are you saying that it did, or that it would if anyone tried it? If it
has been tried, I'd like to see the results, and if not, shouldn't it be
done to get a baseline against which to evaluate the degree of matching
actually obtained? I can't tell whether you're speaking hypothetically
or reporting on actual results.

I have been trying to avoid this topic, since we reached the conclusion
last year that new data were needed to develop a PCT approach to matching.
Now I have to jump in. First, remember that speculation that matching
could be easily produced and modelled proved inaccurate in the attempts
Bill Powers, Rick Marken, Tom Bourbon, Bruce Abbott, and I made last year.
Second, at the time I mentioned work by Miller, Saunders, and Bourland
(1980), Animal Learning and Behavior, 8, 635-641. We used concurrent VI
with a changeover key procedure, so that the VI was available to the
pigeon on one key, while responses to a second key changed the VI. Using
this procedure, we manipulated the degree on discriminability between the
stimuli on the VI key associated with the two schedules. In one condition
of this experiment, the stimuli were identical line orientations (i.e.,
there was no difference in the stimuli associated with the different VIs).
This condition produced severe undermatching, although not complete
indifference. We assumed that the obtained rate of reinforcement itself
provided information to allow this residual control of allocation. In the
current context, this could be considered using the pigeon, rather than a
program, as a means of testing (and rejecting) Bill Powers' conjecture
about matching with random responding, since matching was severely reduced
by removing the pigeons' ability to determine the schedule in which they
were working.

//----------------------------------------------------------------------------
//Samuel Spence Saunders,Ph.D.