What is controlled?

[Bruce Nevin (2016.01.20.12:42 ET)]

Martin Taylor Nov 29, 2015 at 9:04 AM –

I never responded to this. My apology. It got lost in the post-Thanksgiving shuffle. You posed a multiple-choice quiz. My answers are interleaved.

I will ask you three questions. The first is about your use of language, the second is contingent on your answer to the first, and the third on your answer to the second.

  1. Under which of the following conditions, if any, do you consider a variable to be controlled:

(a) when disturbed from an apparent resting position it returns to or near that position.

Your (a) is a necessary but not sufficient condition.

The variable need not be resting as a precondition for observing resistance to disturbances, and it’s usual to talk of the value of a variable rather than its position–presumably a position on some scale of values.

(b) when disturbed from an apparent resting position, it returns to or near that position as the result of an observable effect from a source other than the observed disturbing influence.

The observer perceives the source of q.o as being distinct from the source of d. I don’t know why you’re being coy about identifying the “source other than the observed disturbing influence” as the subject’s behavioral outputs q.o. This is necessary for applying other criteria that are essential to the Test (q.v. below).

A caveat: In an internal conflict, the source of the disturbance is within the observed organism so that we have two loops within the hierarchy, each with its q.o which can eventuate in two successive values of q.o observed in the common environment. (The q.o of one provides a reference for effectors and then the q.o of the other provides a different reference for the same effectors.)

Again, your (b) is necessary but not sufficient.

(c) when disturbed from a resting position apparently determined by some other variable, it returns to or near that resting position as the result of an observable effect from a source other than the observed disturbing influence.

I do not understand what distinction you intend to make here.

The only significant difference between (b) and (c) is the phrase “apparently determined by some other variable”. Here’s a breakdown to show the parallels. (I’ve put two irrelevantly discrepant words in italics.)

(b) when disturbed from an apparent resting position,

(c) when disturbed from a resting position

(c) apparently determined by some other variable,

(b) it returns to or near that position

as the result of an observable effect

from a source other than the observed disturbing influence.

(c) it returns to or near that resting position

as the result of an observable effect

from a source other than the observed disturbing influence.

The most salient reading is that the “[apparent] resting position” of the variable appears to the observer to be “determined by some other variable”. On either reading, I don’t understand what you’re driving at.

First, other than what? Assuming you mean other than the disturbance, are you saying that some (unnamed) variable other than the disturbance has determined the (resting) position? Or is the other variable q.o but for some reason (as above under (b), perhaps) you’re not yet warranted to call it that? Surely, we presuppose that we are observing the subject organism’s activity?

Another possible reading is that it is the disturbance that “appears to be determined by some other variable”.

On that reading, it is even less clear “other than what”. Are you saying that some (unnamed) variable other than q.o has determined the (resting) position of the variable? Or (again) is q.o the “other variable”?

OK, assuming that reading, I’ll venture a paraphrase:

(c) when disturbed from a resting position that was apparently determined by
the subject’s prior actions (which we can’t yet call q.o for some reason),
it returns to or near that resting position as the result of an observable effect from
the subject’s present actions (which we can’t yet call q.o for some reason).

But no, some other agent might well have arranged the variable in a state that is no disturbance for the subject. So I can’t see the relevance of the distinction in (c). Surely no one would dispute that the socalled ‘resting position’ was determined by some prior cause or causes.

There are other necessary conditions in addition to (a) and (b), including:

(d) The subject organism must be able to perceive the putatively controlled variable.

(e) The subject must actually be perceiving the variable when observations of putative control are made.

(f) The subject’s activity must be able to affect the state of the variable.

(g) Actual subject outputs q.o must be observed to cancel the effects of d on the variable.

(h) Observer can and does predict the effect of d if q.o is not present.

(i) The disturbance d avoids side effects disturbing other variables that the subject might be controlling.

There may be more that I’m overlooking in a quick look through Phil Runkel’s summary statement in People as living things (2003:77-79), which in turn was based on Powers (1973:232-246; 2005:233-248; 1979:110, 112), and (I am sure) conversations with Bill, including the correspondence reprinted in Powers & Runkel (2011).

  1. If your answer includes (a) or (b), do you think it possible to have two independent controlled variables in a standard PCT control loop?

The framing of the question suggests to me that you see these choices as mutually exclusive. I do not say that (a) or (b) is sufficient. I do say that (b), which subsumes (a), is necessary as are also (d)-(i). I don’t understand (c)

But what is most striking to me about your question (2) is the word “independent”.

The question is not about independent controlled variables. p is certainly dependent upon q.i, and the chain of dependency goes all the way around the loop, q.i dependent on q.o + d, q.o dependent upon r - p. Perhaps the assertion that q.i and p are both controlled is a disturbance to you because you understand this to mean that they are independently controlled. They are two aspects of one thing. q.i is the controlled variable CV as perceived by an observer, and p is the same CV as perceived by the subject.

And that is my answer to your third question.

···

On Sun, Nov 29, 2015 at 9:04 AM, Martin Taylor mmt-csg@mmtaylor.net wrote:

Bruce,

I will ask you three questions. The first is about your use of

language, the second is contingent on your answer to the first, and
the third on your answer to the second.

1. Under which of the following conditions, if any, do you consider

a variable to be controlled:

(a) when disturbed from an apparent resting position it returns to

or near that position.

(b) when disturbed from an apparent resting position, it returns to

or near that position as the result of an observable effect from a
source other than the observed disturbing influence.

(c) when disturbed from a resting position apparently determined by

some other variable, it returns to or near that resting position as
the result of an observable effect from a source other than the
observed disturbing influence.

2. If your answer includes (a) or (b), do you think it possible to

have two independent controlled variables in a standard PCT control
loop?

3. If your answer to (2) is no, which variable in the standard PCT

control loop is controlled?

---------



My answers: 1-only c, 2- No, 3-the perception. (I answered 2 as a

tautology, given that 1c precludes a yes answer to 2).

Martin



  On 2015/11/28 10:54 PM, Bruce Nevin

wrote:

[Bruce Nevin (2015.11.28.22:53 ET)]

      Martin Taylor (2015.11.26.14.54)

            BN: This was in response

to a person who denies that Qi is controlled at all.

            MMT: I have read Boris

as simply pointing out that the output affects Qi in
order that
perception is controlled. I say the same.

              BN: if that were the

case, there would be no way for an observer to notice
the fact of control. No stabilization of the
environment against disturbances would be perceptible
to anyone except the organism that was doing the
controlling.

              MMT: Why not? I never perceived you as being of the

all-or-none Black or White persuasion, but here you
are saying that if (as must be the case) an observer
has a different set of inputs to the senses than the
person doing the controlling, no matter how similar
their inputs and perceptual functions may be, what the
observer sees must be totally unrelated to what the
controller sees. Sure, if the controller is
controlling the placement of a glass on a table, and
the observer is looking at the degree to which a door
is open, the observer will say there’s no control. But
that’s not what we are talking about, is it? The
observer sees the glass on the table, and if he wants
to know whether the controller cared where it was
placed, the observer can become an experimenter and
move it. The fact that they see it from different
angles may matter, but probably doesn’t.

        I am not at all saying that since inputs to the

observer’s senses are different from the inputs to the
senses of the subject, “what the observer sees must be
totally unrelated to what the controller sees”. Although as
I attempt in vain to relate that to what I said, it does
seem that you may be exemplifying what you said.

        I am saying that in your glass scenario or in the TCV the

perception that each of the participants controls is related
to the perception that the other controls by way of
that aspect of their common environment which they are controlling .
To talk about that relationship of the observer’s perception
to the subject’s perception, you prefer to say that the
perception that each of them controls is related to the
perception that the other controls by way of that aspect of
their common environment which they are influencing .
I assume you have a purpose for that choice of words, but
you have not stated it. I have a purpose in saying that * Qi*is controlled. I will explain that here.

        In the TCV, the tester controls variables until a

(gentle) conflict with the subject is confirmed. That
conflict affirms that they are both controlling the same
aspect of the environment. Or in your words, they are both
influencing the same aspect of the environment. That
controlled or influenced aspect of the environment is
quantified as Qi. The controlled perception p is a
transform of Qi from physical units measured in the
environment to (per the PCT model) a rate of firing in a
nerve or nerve bundle. The transformation by the input
function is quantified as a constant Ki . You have
objected that imperfections in the sensory apparatus make Ki a
noisy variable. My rejoinder was that if that has any
significant effect at all, and is not zeroed out as just
another disturbance in the loop, the effect is that Qi is
less well controlled than p is, but Qi is
nonetheless still controlled.

        As far as I can see, to say that the tester and the

subject are merely influencing Qi (or that aspect of
the environment which is quantified as Qi ) as means
of controlling their respective perceptions is sophistry, a
terminological distinction without a difference, serving no
purpose and confusing the issue. Or if you do have a purpose
in making that distinction, please do say what it is. But
even my astigmatism does not interfere with my ability to
put that glass back where I want it, so perfect me no run of
the mill sensory imperfections, please. Or, more politely,
let us say that I remain unconvinced.

        Perceptual control has environmental consequences that

are perceived (and can be controlled) by others. Your
position is that when a perception is controlled the
environmental consequences are not controlled. In my view,
environmental consequences that are not controlled are
called side effects.

        In your view, the environment is merely influenced by

control activities in order that the perception may be
controlled. The perceived influence is controlled, but the
influence that is perceived is not controlled. The intended
environmental consequences of that influence do not
constitute control of the affected aspect of the
environment. I say that there is evidence that the affected
aspect of the environment is controlled, and that the
environmental consequences of control, as perceived by
others, measured by instruments, etc., are controlled. The
effect is intentional. Indeed, the nature of that effect is
precisely, control. One kind of evidence is that it is
perceived by another as control. “What are you doing to that
glass?” Another is that the tester’s perception (from the
imagined point of view of the subject) is sufficient basis
from which successfully to deduce the subject’s internally
maintained reference value for p . Another is that
conflict often has environmental consequences (“Now see what
you’ve done! You’ve spilled the water!”) which may disturb
collectively controlled variables. Collective control is yet
another kind of evidence: stabilization of what? An * environmental*feedback path.

          Perhaps you are brought to your position in part by the

testimony of the physical sciences that the objects,
relations, and events that we perceive devolve to shifting
arrangements of subatomic particles and energy.

Did old Sam Johnson bruise his foot in vain?

          The assumption that perceptions are veridical, and that

control of a perception indicates control of that which is
perceived, is the converse of a sacrament. A sacrament, as
you may recall, is said to be an outward and visible sign
of an inward and spiritual reality. A controlled
perception is an inward and perceptible sign of an outward
reality which, aside from perceptions, is unknowable. The
latter is as much an article of faith as the former. Yet
it certainly seems not so, because of our existential
reliance on perceptions. Indeed, faith of the sacramental
sort is characterized by belief without evidence; and your
claim seems to amount to saying that the only evidence we
have, our perceptions, is no evidence at all. As Alice
would say, curiouser and curiouser.

I have two questions:

  1. How do you avoid solipsism?
  2.               What explanatory principles do you invoke to account
    

for how the Test for the controlled variable discloses
the subject’s CV on the basis of your perceptions?
(Let the TCV serve as first proxy for the other kinds
of evidence enumerated above.)

          I postulate only one explanatory principle: that an

aspect of the environment is controlled when a perception
is controlled.

          We derive our conviction as to the veridicality of

perception from the mutual consistency of many
perceptions, including our incessant informal testing of
what variables those around us are controlling.
Collaboration, collective control, conflict and its
resolution, all hinge upon a public actuality that is
commonly affected by the separate and private control of
perceptions by the participants, thereby confirming again
and again that control of perceptions is by means of
control of the perceived environment. Is that confidence
ill founded?

          The physical universe, whatever it is, is resistant to

our control activities. When you shift the alignment of a
dime in the coin game or a glass on the table it stays put
when you take your hand away. Presumably, that resistance
emerges from what seems to be an infinite plasticity of
subatomic phenomena somehow–collective control by
infinitesimal points of energy/consciousness?–but however
it comes about, a consequence is that control through the
environment is very different from control in imagination.
When we control our perceptions, we do so by overcoming
the inertial character of material things, by making
changes in the environment which are perceived as effects
of our control of perceptions. And a great many of those
effects endure in our absence until our return. The
furniture is where we left it. Ah, that’s where I left my
glasses, now I remember.

          I know he's a crotchety old fellow, but let Mr. Ockham

have a word. On offer is a single explanatory principle to
account for all this: an aspect of reality is controlled
when the perception of it is controlled. Please show us
how any other account avoids multiplying explanatory
principles.

/Bruce

        On Thu, Nov 26, 2015 at 3:21 PM,

Martin Taylor mmt-csg@mmtaylor.net
wrote:

            [Martin Taylor

2015.11.26.14.54]

[Bruce Nevin (2015.11.25.20:04 ET)]

Martin Taylor 2015.11.24.23.37 –

                      when [disturbances] appear between the

controlled variable p and the complex
environmental variable (the CEV) to which it
corresponds, all that means is that the CEV is
not controlled as precisely as the perception
is

                    Yes, but it is controlled, however

imperfectly that may be.

             I deny that.



            The appearance of control of teh Complex Environmental

Variable is, if I understood you correctly when you used
the term in another context, a spandrel. The appearance
that is is controlled is a consequence of something else
truly being controlled. It used to be quite obvious that
phlogiston flowed in and out of objects, and even now we
observe heat flowing in and out of objects, but there’s
“really” no flow of anything. All there is is a bunch of
molecules moving around and beating the hell out of each
other. It’s the same kind of thing. When we talk
casually, I have no objection to saying that the CEV is
controlled. I do it myself quite often. But when we want
to explain the theory to anyone in or out of CSGnet, one
of the very first things we have to explain is that the
CEV is not controlled, however much it looks as though
it is. The perception of it is controlled, and that is
the reason it looks as though it is controlled (as also
is everyone else’s perception of anything correlated
with the CEV, though that fact is never mentioned in
this discussion; why should the argument not be that the
controller is controlling what some undefined other
person is perceiving? The logic is the same.).

                    You're denying my assertion that a

disturbance at that point in the loop can be
resisted.

             No I most definitely am not!!! If it were not

resisted, how could the corresponding perception be
controlled?

                    In the case of a pathology, it is certainly

the case that control is impaired, as I said.
Example: before the invention of corrective
lenses, my astigmatism would require me to rely
on others to make out details of a scene and
report them, as would my relatively slight
myopia.

                      Look, all I'm trying to do is to emphasize

that PCT is about The Control of Perception,
something that seems in danger of being
forgotten even on CSGnet.

                    Yes, but consider the context. This was in

response to a person who denies that Qi is
controlled at all.

             I guess we bring different prior assumptions to

our reading of what always must be ambiguous, and that’s
foubly true of someone whose first language is no
variety of English. I often disagree with what Boris
says, but on this I have read Boris as simply pointing
out that the output affects Qi in order that
perception is controlled. I say the same, but I am not
usually told I don’t understand PCT – at least not in
the 20 or so years since Bill challenged Rick when he
made that claim, saying something along the lines of
“Who do you think you are saying doesn’t understand
PCT?”

                    I agree with Rick: if that were the case,

there would be no way for an observer to notice
the fact of control. No stabilization of the
environment against disturbances would be
perceptible to anyone except the organism that
was doing the controlling.

Why not? I never perceived you as being of the
all-or-none Black or White persuasion, but here you are
saying that if (as must be the case) an observer has a
different set of inputs to the senses than the person
doing the controlling, no matter how similar their
inputs and perceptual functions may be, what the
observer sees must be totally unrelated to what the
controller sees. Sure, if the controller is controlling
the placement of a glass on a table, and the observer is
looking at the degree to which a door is open, the
observer will say there’s no control. But that’s not
what we are talking about, is it? The observer sees the
glass on the table, and if he wants to know whether the
controller cared where it was placed, the observer can
become an experimenter and move it. The fact that they
see it from different angles may matter, but probably
doesn’t.

                Martin

/Bruce

                      On Tue, Nov 24, 2015 at

11:59 PM, Martin Taylor mmt-csg@mmtaylor.net
wrote:

[Martin Taylor 2015.11.24.23.37]

                            On 2015/11/24 10:37 PM, Bruce Nevin

wrote:

                              [Bruce Nevin (2015.11.24.

ET)]

                                      Martin

Taylor (2015.11.24.14.02) –

                                      "PCT"

doesn’t imply it. It’s simply
a fact of life (and of
engineering) that ONLY if the
connection from Qi to the
perceptual variable is
invertible, perfect, and
noise-free will there be no
difference between the ECV
(whatever that may be) and the
perception. The perception is
controlled, and as a
consequence, the environmental
variable appears to be. As an
approximation, it’s good
enough for most purposes, but
like Newtonian gravity, it’s
not a good foundation for
theoretical discussion or
precise analysis.

                                    In

the equations that I’m familiar
with the connection from * Qi*to p is
represented by a constant Ki .

                          So it is, but how realistic do you think

that is in the real world of live
organisms?

                                    Hasn't

that sufficed for implementing
simulations, or have I missed
something?

                           As I said: " As an

approximation, it’s good enough for most
purposes,". Does anyone claim that the
simulations actually represent what goes
on inside the organism? Even the concept
of a neural current has no equivalent in
an actual brain. It’s an analytical
convenience, an abstraction that simply
assumes that the effect of a lot of
neurons firing with their own timings is
the same as though one super-neuron
performed all the firings, and then
smeared them across time so that a
smooth variation was used in further
functions. For most purposes, that’s
fine, but if you really want to think
about it, Bill just said that if it’s
within a few percent (5%, 2%, I forget)
that’s good enough. And it usually is.
But it doesn’t mean that it’s perfect.

                                    Any imperfection and noise in

the biological implementation is
just another disturbance.
Disturbances can enter at any
point in the loop.

                           Indeed, but when they appear

between the controlled variable p and the
complex environmental variable (the CEV)
to which it corresponds, all that means is
that the CEV is not controlled as
precisely as the perception is.

                                    If such disturbances could not

be countered by the control
process in the same way that
environmental disturbances are,
and if they were great enough to
make p depart from its
correspondence to the relevant
aspect of the environment, as
represented by Qi , they
would be pathologies making it
less likely for that organism to
succeed in bringing offspring to
reproductive maturity, so there
is obvious evolutionary pressure
for that coupling to be quite
good enough to support good
control.

                           Yes. That, in essence, is what I

said when I said “it’s good enough for
most purposes”. You have read a lot of my
writings. How often have I written in
things addressed to PCT newbies that
though what is controlled is perception,
it’s what happens in the environment that
matters?

                                    That seems to me a pretty strong

basis for that coupling being
treated as a constant Ki .
rather than as a variable
subject to significant
unpredictable perturbations.

                           Much more likely to be some kind 

of approximation to log(Qi) with some kind
of ceiling and some kind of zero-region
tolerance zone.

                          It doesn't matter, what the function is,

if it’s invertible. noise free, and
consistent (which adapting systems are
not).

                          Look, all I'm trying to do is to emphasize

that PCT is about The Control of
Perception, something that seems in
danger of being forgotten even on CSGnet.

                              Martin

/Bruce

                                  On Tue, Nov

24, 2015 at 2:44 PM, Martin Taylor
mmt-csg@mmtaylor.net
wrote:

                                      [Martin Taylor

2015.11.24.14.02]

                                            [From Rick

Marken
(2015.11.22.0950)]

                                       No. p1 is the

controlled quantity, the
perception that tracks the
reference value -d2 closely if
all the g values are
substantially greater than
unity. In a real control loop,
of course, the “g” multipliers
would represent the long-term
stable values of the leaky
integrators, just as in the
usual analysis of the simple
control loop. The actual loop
could not use simple
multipliers. When there are
loop delays, simple
multipliers inevitably lead to
oscillation and no control. My
analysis was of the stable
equilibrium values, and for
that, the leaky integrators
are well represented by simple
multipliers.

                                      And what's an "ECV"?
                                       Yes, I am interested.

It’s good to have the
spreadsheet example. What gain
and leak rates did you use for
the four “g” functions, to get
the “g” multipliers? The
effect of d4 and d3 is
diminished by the multiplier
ratio each step back round the
loop. My analysis assumed,
g>>1, as we do when we
do an equilibrium analysis of
the ordinary control loop and
assume the loop gain
G>>1.

                                      I imagine that in your

spreadsheet you have a scalar
variable and a simple
multiplier, as I showed in the
example. One can’t actually
run the TCV on a single
scalar, because there is no
function to be found. But it
would be interesting to run a
spreadsheet example in which
each of the paths was a vector
of, say, three scalars, and
each perceptual function was
different, and then run the
TCV to see what you find.

                                      "PCT" doesn't imply it. It's

simply a fact of life (and of
engineering) that ONLY if the
connection from Qi to the
perceptual variable is
invertible, perfect, and
noise-free will there be no
difference between the ECV
(whatever that may be) and the
perception. The perception is
controlled, and as a
consequence, the environmental
variable appears to be. As an
approximation, it’s good
enough for most purposes, but
like Newtonian gravity, it’s
not a good foundation for
theoretical discussion or
precise analysis.

                                       That's at least

equally wrong. I think it
might be worth your while to
look a little more closely
into the actual conditions for
using the TCV, and the
potential and limitations on
what you can determine by
using it. You often seem to
suggest (planning in
imagination) that you might
use the TCV in real-life
situations. Sometimes the
conditions are suitable, but
much more often, they aren’t.
I haven’t done it, so I am
also planning in imagination,
but one ought to be able to
run the TCV on your demo of a
three-level control system to
find what is being controlled
at the top level. You have all
the outputs and disturbances
necessary, so it should work.
But what about in a real-life
situation in which the
circumstances never recur. In
the hammering example, this
might be the only time in the
hammerer’s life that he is so
angry with his wife that he
has to hit something, and
doesn’t want to hit his wife.
How can the TCV be used in
that situation?

                                      And how do you use the TCV

when control is poor? If you
get a poor compensation of the
disturbance by the output, how
do you know whether you
haven’t found the controlled
variable or you have found it
and the control system doesn’t
work very well?

                                      I think your statement is

simply equivalent to saying
“PCT research is impossible”
which is something I don’t
believe, though I do believe
that the control of perception
accounts for what we see
people and other organisms do,
and that we should carefully
study by all available means
just how this works.

                                          Martin
                                                    Bruce

Nevin
(2015.11.21.20:44
ET) to Martin
Taylor

                                                      BN: Thanks

for this nice
demonstration
of the
difficulty
with this
distinction
[between ECV
and p).

                                                  RM: I think

Martin aimed to
demonstrate that
controlling an ECV
is not equivalent
to controlling the
perception that
corresponds to
that ECV. But, in
fact, his
demonstration
doesn’t
demonstrate that
at all. What it
demonstrates is
that when you put
a bunch of
disturbances into
the feedback
connection between
output and input
you control
neither the ECV
nor p. To be
precise, the
disturbance
variables, d3 and
d4. enter the loop
after the output,
p2, and before the
input, (p4+d1).
See Martin’s
diagram below:

                                                  RM: When these

disturbances are
present the effect
of output (p2) on
input (p4+d1) is
constantly
changing. However,
if you remove
these disturbances
from the feedback
function control
is restored.

                                                  RM: In this

control loop p4+d1
is the controlled
quantity (q.i,or
ECV) and
g1*(p4+d1) is the
controlled
perception, p.

                                                  The only

difference between
q.i (the ECV) and
p is the scaling
factor, g1. But
variations in p
(p1) are perfectly
correlated with
variations in the
ECV (p4+d1); the
only difference
between p and ECV
is that the
former is measured
in neural firing
rate units and ECV
is measures in
physical units;
g1 is just a
scaling factor
that converts
physical units
into neural firing
rate units .

                                                  RM: I've

implemented
Martin’s model in
a spreadsheet, in
case anyone is
interested. It
allows you to see
how well the
perception, p1,
and corresponding
ECV (p4+d1) are
controlled when
the disturbances
to the feedback
function (d3 and
d4) are in or out
of the loop. When
these disturbances
are in, control of
both p1 and ECV is
poor but the
correlation
between variations
in p1 and the ECV
is 1.0; when these
disturbances are
out control of
both p1 and ECV
are excellent; and
the correlation
between p1 and ECV
is again 1.0.

                                                  RM:P I don't

know how people
got the idea that
PCT implies that
there is a
difference between
control of a
perception and
control of the
corresponding ECV.

                                                  But it's an

idea that is not
only wrong but one
that, if believed,
make PCT research
impossible.

[Martin Taylor 2015.11.21.10.51]

In many messages over the years, there has been confusion about just

what is controlled in a control loop, the perception or something in
the environment of the controller. Here’s one recent example in
which the confusion is self-contained [From Rick Marken
(2015.11.15.1630)]:

---------quote------
[RM] The input to the perceptual function is ultimately the set of

environmental variables at the sensory surface, call it
(v.1,v.2,…v.n), as per Fig. 1, p. 66 of LCS I. Qi is the function
of this set of variables that corresponds to a controlled
perception: Qi = f(v.1,v.2,…v.n), where f() is the perceptual
function. Another way to look at this is that Qi represents the ** aspect
of the environment** that is controlled when the perceptual
signal, p, that corresponds to Qi, p = f(v.1,v.2,…v.n), is
controlled.

----end quote----

What Rick says is true for the simple isolated control loop with a

passive environmental feedback path, but it isn’t true in general
for controlled perceptions. To see the problem, we can use a simple
illustrative diagram in which Rick’s f(.) is a multiplier “g” with
one input, so p = g*v. Four of these are arranged in a ring, and “v”
for each of them is the sum of two quantities, a “d” from outside
the ring and the “p” from the previous device, like this:

![loop_1234.jpg|447x384](upload://j9kgvva6hF6sowdVD46rW23Se76.jpeg)

If an odd number of the "g" multipliers is negative, this is a

negative feedback loop, and if the “g” absolute values are similar
and appreciably greater than unity, each “p” value closely tracks
the negative of the following “d”. For example p1 will be nearly
equal to -d2. In other words, every “p” in this homeostatic loop is
a controlled variable with a reference value of -d.

Looking at the loop from the viewpoint of p1 by collapsing the loop

from the output of g2 through the input to g1, we get this:

![Re What is controlled10.jpg|885x452](upload://ewz3HRpizOU0HqObmfTv47Ju1LF.jpeg)

This looks remarkably like an ordinary control loop, which is not

strange, because it is one. It controls p1 with a reference value of
-d1. The perceptual input function is the multiplier g1 and the
output function is the multiplier g2. The Environmental Feedback
Path also has gain, g3g4, so the loop gain G is g1g2g3g4. If the
absolute magnitude of each “g” is considerably greater than unity, G
is very much greater than any one of them. The disturbance dx is
some combination of d1, d3, and d4. It doesn’t matter what
combination it is for the purposes of the illustration.

The question at issue is what in the environment is controlled?

Whatever it is must be found where the virtual disturbance dx enters
the loop. But where is that? The physical disturbances that
contribute to dx (reference values for other “p” variables) enter at
different places around the loop.

Let's imagine an external observer (EO) interested in using the Test

for the Controlled Variable (TCV) to determine what the controlled
variable p1 corresponds to. What might EO hypothesize and disturb?
Whichever of the observable variables EO chose, the disturbance
would be opposed, not because p1 was controlled, but because it was
the reference value for the negative of the immediately preceding p.
Only by hypothesising the entire complex of the Environmental
Feedback Function (g3, p3, d4, g4, and d1) could EO discover p1 and
its reference value -d2. If EO simply added a random disturbance to
d3 or to d4, the TCV would lead to the conclusion that p2 or p3
respectively was THE controlled variable. p1 would not appear at
all. This simple loop is a control loop for all four p variables,
and if it failed for one (say one of the g values became much less
than unity or changed sign), it would fail for all.

I suppose one might say that the "controlled" environmental variable

is some function of g3, p3, d4, g4, and d1, but this would
contradict Rick’s assertion that it is defined by the perceptual
function, in this case the multiplier g1 (an assertion with which I
fully agree when the Environmental Feedback Function has only one
disturbance entry point and is otherwise passive).

I conclude that consistency in PCT can be maintained only by

restricting the concept of a controlled variable to perceptions, and
that talking about “environmental controlled variables” can only
confuse the issue, no matter how closely the presumed environmental
variable found by the TCV tracks the perceptual variable in a simple
“single-scalar” control loop.

Martin

[Bruce Nevin (2015.11.21.20:44 ET)]

Am I remembering correctly that it was you who introduced the ECV acronym some years ago to distinguish the ECV (environmental CV) from the CV ?

Thanks for this nice demonstration of the difficulty with this distinction. It always seemed to me that the ECV was the perception (CV) controlled by the observer and imputed to the observed subject by performing the Test (TCV). The observer has to imagine the perceptual capacities and limitations of the subject insofar as they differ from her or his own, but it’s still the observer’s perception.

All we have, observationally, is the ECV, and qo, qi, and d, which can only be specified relative to the ECV. Rick’s spreadsheet is limited to observables, for good reason. The values of p, r, and e are so far inferred, though Henry is getting some actual values.

loop_1234.jpg

···

On Sat, Nov 21, 2015 at 12:41 PM, Martin Taylor mmt-csg@mmtaylor.net wrote:

[Martin Taylor 2015.11.21.10.51]

In many messages over the years, there has been confusion about just

what is controlled in a control loop, the perception or something in
the environment of the controller. Here’s one recent example in
which the confusion is self-contained [From Rick Marken
(2015.11.15.1630)]:

---------quote------
[RM] The input to the perceptual function is ultimately the set of

environmental variables at the sensory surface, call it
(v.1,v.2,…v.n), as per Fig. 1, p. 66 of LCS I. Qi is the function
of this set of variables that corresponds to a controlled
perception: Qi = f(v.1,v.2,…v.n), where f() is the perceptual
function. Another way to look at this is that Qi represents the ** aspect
of the environment** that is controlled when the perceptual
signal, p, that corresponds to Qi, p = f(v.1,v.2,…v.n), is
controlled.

----end quote----



What Rick says is true for the simple isolated control loop with a

passive environmental feedback path, but it isn’t true in general
for controlled perceptions. To see the problem, we can use a simple
illustrative diagram in which Rick’s f(.) is a multiplier “g” with
one input, so p = g*v. Four of these are arranged in a ring, and “v”
for each of them is the sum of two quantities, a “d” from outside
the ring and the “p” from the previous device, like this:

If an odd number of the "g" multipliers is negative, this is a

negative feedback loop, and if the “g” absolute values are similar
and appreciably greater than unity, each “p” value closely tracks
the negative of the following “d”. For example p1 will be nearly
equal to -d2. In other words, every “p” in this homeostatic loop is
a controlled variable with a reference value of -d.

Looking at the loop from the viewpoint of p1 by collapsing the loop

from the output of g2 through the input to g1, we get this:

This looks remarkably like an ordinary control loop, which is not

strange, because it is one. It controls p1 with a reference value of
-d1. The perceptual input function is the multiplier g1 and the
output function is the multiplier g2. The Environmental Feedback
Path also has gain, g3g4, so the loop gain G is g1g2g3g4. If the
absolute magnitude of each “g” is considerably greater than unity, G
is very much greater than any one of them. The disturbance dx is
some combination of d1, d3, and d4. It doesn’t matter what
combination it is for the purposes of the illustration.

The question at issue is what in the environment is controlled?

Whatever it is must be found where the virtual disturbance dx enters
the loop. But where is that? The physical disturbances that
contribute to dx (reference values for other “p” variables) enter at
different places around the loop.

Let's imagine an external observer (EO) interested in using the Test

for the Controlled Variable (TCV) to determine what the controlled
variable p1 corresponds to. What might EO hypothesize and disturb?
Whichever of the observable variables EO chose, the disturbance
would be opposed, not because p1 was controlled, but because it was
the reference value for the negative of the immediately preceding p.
Only by hypothesising the entire complex of the Environmental
Feedback Function (g3, p3, d4, g4, and d1) could EO discover p1 and
its reference value -d2. If EO simply added a random disturbance to
d3 or to d4, the TCV would lead to the conclusion that p2 or p3
respectively was THE controlled variable. p1 would not appear at
all. This simple loop is a control loop for all four p variables,
and if it failed for one (say one of the g values became much less
than unity or changed sign), it would fail for all.

I suppose one might say that the "controlled" environmental variable

is some function of g3, p3, d4, g4, and d1, but this would
contradict Rick’s assertion that it is defined by the perceptual
function, in this case the multiplier g1 (an assertion with which I
fully agree when the Environmental Feedback Function has only one
disturbance entry point and is otherwise passive).

I conclude that consistency in PCT can be maintained only by

restricting the concept of a controlled variable to perceptions, and
that talking about “environmental controlled variables” can only
confuse the issue, no matter how closely the presumed environmental
variable found by the TCV tracks the perceptual variable in a simple
“single-scalar” control loop.

Martin

[Marftin Taylor 2015.11.21.22.57]

The acronym was CEV, for Complex Environmental Variable, not

“Controlled” Environmental Variable. The point was to emphasise
Rick’s point about the CEV being determined by the perceptual
function. If by hypothesis one can segregate a simple control loop
with a perceptual function that has many inputs from the senses and
none from imagination, and has a simple “pass-through” environmental
feedback function, then the CEV in that loop does vary precisely as
does the perceptual signal, which is what the concept of the TCV is
based on.
Yes.
We don’t have the CEV observationally. That was my point (in part)
about Bill’s required precondition: “once behaviour has been defined
in terms of an appropriate variable”. The observer has to have a
perceptual function for the same complex as the subject is
controlling, and, as Bill said, this has to be done by assumption.
Actually, if the situation permits the use of the TCV, it doesn’t,
but it’s always an assumption that the situation does permit the
TCV, the biggest assumption being that the perception being
controlled stays the same while you change the disturbance to what
you think is “the appropriate variable”, the CEV. If all the
conditions apply and you have infinite time, you can get arbitrarily
close to the CEV with an arbitrarily high level of probability. So
it isn’t impossible to be almost certain about the CEV in some
special situations, typically those of an experiment. It’s hard to
find real-life conditions where all the necessary conditions for the
TCV are present.
More commonly, the hypothesized CEV is “self-evident”. In the
driving example, it is “self-evident” that the driver intended to
depress the clutch by the amount it was observably depressed.
Self-evident truths, strange to say, are often correct, but they
aren’t always correct and it is usually hard to determine whether
they are or not. Quite possibly the driver had a muscular spasm,
stretched the leg, and the pedal got in the way. It’s unlikely, but
possible.
Once you get over that problem by saying
“once behaviour has been defined in terms of an appropriate
variable”, it’s quite likely that you will be able to observe
something close to a value of the CEV that corresponds to the
reference value of the controlled perception, but only if control is
reasonably good.
Behaviour that is not very effective in controlling the perception,
such as voting in an election, may not bring the perception of the
elected party to its reference value, and the observer won’t see it,
though the observer will still see the behaviour (voting). There is
a record of whether someone voted, in order to ensure that they
don’t vote more than once, so the observer can know that the
behaviour occurred. But the observer does not know, and would not
know even if the ballot were public, what the voter’s reference
value is for a perfect candidate since none of the available choices
are likely to match that reference state. So I don’t think it is at
all reasonable to say that observable behaviour includes anything to
do with the reference value for the controlled perception. Martin
PS. I notice a typo in the message to which you responded. Just
under the second figure, I said: “This looks remarkably like an
ordinary control loop, which is not strange, because it is one. It
controls p1 with a reference value of -d1.” The reference value is,
of course, -d2.

loop_1234.jpg

···

[Bruce Nevin (2015.11.21.20:44 ET)]

      Am I remembering correctly that it was you who introduced

the ECV acronym some years ago to distinguish the ECV
(environmental CV) from the CV ?

      Thanks for this nice demonstration of the difficulty with

this distinction. It always seemed to me that the ECV was the
perception (CV) controlled by the observer and imputed to the
observed subject by performing the Test (TCV). The observer
has to imagine the perceptual capacities and limitations of
the subject insofar as they differ from her or his own, but
it’s still the observer’s perception.

      All we have, observationally, is the ECV, and qo, qi, and

d, which can only be specified relative to the ECV. Rick’s
spreadsheet is limited to observables, for good reason. The
values of p, r, and e are so far
inferred, though Henry is getting some actual values.

/B

      On Sat, Nov 21, 2015 at 12:41 PM,

Martin Taylor mmt-csg@mmtaylor.net
wrote:

          [Martin Taylor

2015.11.21.10.51]

          In many messages over the years, there has been confusion

about just what is controlled in a control loop, the
perception or something in the environment of the
controller. Here’s one recent example in which the
confusion is self-contained [From Rick Marken
(2015.11.15.1630)]:

          ---------quote------
          [RM] The input to the perceptual function is ultimately

the set of environmental variables at the sensory surface,
call it (v.1,v.2,…v.n), as per Fig. 1, p. 66 of LCS I.
Qi is the function of this set of variables that
corresponds to a controlled perception: Qi =
f(v.1,v.2,…v.n), where f() is the perceptual function.
Another way to look at this is that Qi represents the **
aspect
of the environment** that is controlled when the
perceptual signal, p, that corresponds to Qi, p =
f(v.1,v.2,…v.n), is controlled.

          ----end quote----



          What Rick says is true for the simple isolated control

loop with a passive environmental feedback path, but it
isn’t true in general for controlled perceptions. To see
the problem, we can use a simple illustrative diagram in
which Rick’s f(.) is a multiplier “g” with one input, so p
= g*v. Four of these are arranged in a ring, and “v” for
each of them is the sum of two quantities, a “d” from
outside the ring and the “p” from the previous device,
like this:

          If an odd number of the "g" multipliers is negative, this

is a negative feedback loop, and if the “g” absolute
values are similar and appreciably greater than unity,
each “p” value closely tracks the negative of the
following “d”. For example p1 will be nearly equal to -d2.
In other words, every “p” in this homeostatic loop is a
controlled variable with a reference value of -d.

          Looking at the loop from the viewpoint of p1 by collapsing

the loop from the output of g2 through the input to g1, we
get this:

          This looks remarkably like an ordinary control loop, which

is not strange, because it is one. It controls p1 with a
reference value of -d1. The perceptual input function is
the multiplier g1 and the output function is the
multiplier g2. The Environmental Feedback Path also has
gain, g3g4, so the loop gain G is g1g2g3g4. If the
absolute magnitude of each “g” is considerably greater
than unity, G is very much greater than any one of them.
The disturbance dx is some combination of d1, d3, and d4.
It doesn’t matter what combination it is for the purposes
of the illustration.

          The question at issue is what in the environment is

controlled? Whatever it is must be found where the virtual
disturbance dx enters the loop. But where is that? The
physical disturbances that contribute to dx (reference
values for other “p” variables) enter at different places
around the loop.

          Let's imagine an external observer (EO) interested in

using the Test for the Controlled Variable (TCV) to
determine what the controlled variable p1 corresponds to.
What might EO hypothesize and disturb? Whichever of the
observable variables EO chose, the disturbance would be
opposed, not because p1 was controlled, but because it was
the reference value for the negative of the immediately
preceding p. Only by hypothesising the entire complex of
the Environmental Feedback Function (g3, p3, d4, g4, and
d1) could EO discover p1 and its reference value -d2. If
EO simply added a random disturbance to d3 or to d4, the
TCV would lead to the conclusion that p2 or p3
respectively was THE controlled variable. p1 would not
appear at all. This simple loop is a control loop for all
four p variables, and if it failed for one (say one of the
g values became much less than unity or changed sign), it
would fail for all.

          I suppose one might say that the "controlled"

environmental variable is some function of g3, p3, d4, g4,
and d1, but this would contradict Rick’s assertion that it
is defined by the perceptual function, in this case the
multiplier g1 (an assertion with which I fully agree when
the Environmental Feedback Function has only one
disturbance entry point and is otherwise passive).

          I conclude that consistency in PCT can be maintained only

by restricting the concept of a controlled variable to
perceptions, and that talking about “environmental
controlled variables” can only confuse the issue, no
matter how closely the presumed environmental variable
found by the TCV tracks the perceptual variable in a
simple “single-scalar” control loop.

              Martin

[From Rick Marken (2015.11.22.0950)]

loop_1234.jpg

···

Bruce Nevin (2015.11.21.20:44 ET) to Martin Taylor

BN: Thanks for this nice demonstration of the difficulty with this distinction [between ECV and p).

RM: I think Martin aimed to demonstrate that controlling an ECV is not equivalent to controlling the perception that corresponds to that ECV. But, in fact, his demonstration doesn’t demonstrate that at all. What it demonstrates is that when you put a bunch of disturbances into the feedback connection between output and input you control neither the ECV nor p. To be precise, the disturbance variables, d3 and d4. enter the loop after the output, p2, and before the input, (p4+d1). See Martin’s diagram below:

RM: When these disturbances are present the effect of output (p2) on input (p4+d1) is constantly changing. However, if you remove these disturbances from the feedback function control is restored.

RM: In this control loop p4+d1 is the controlled quantity (q.i,or ECV) and g1*(p4+d1) is the controlled perception, p. The only difference between q.i (the ECV) and p is the scaling factor, g1. But variations in p (p1) are perfectly correlated with variations in the ECV (p4+d1); the only difference between p and ECV is that the former is measured in neural firing rate units and ECV is measures in physical units; g1 is just a scaling factor that converts physical units into neural firing rate units .

RM: I’ve implemented Martin’s model in a spreadsheet, in case anyone is interested. It allows you to see how well the perception, p1, and corresponding ECV (p4+d1) are controlled when the disturbances to the feedback function (d3 and d4) are in or out of the loop. When these disturbances are in, control of both p1 and ECV is poor but the correlation between variations in p1 and the ECV is 1.0; when these disturbances are out control of both p1 and ECV are excellent; and the correlation between p1 and ECV is again 1.0.

RM: So there is no evidence in this demonstration of any difference between control of a perception and control of the corresponding ECV. When you are controlling a perception of some aspect of the environment you are also controlling that aspect of the environment.

RM: Why this is even an issue is astounding to me. If it were actually true that PCT implied a difference between control of a perception and control of the corresponding ECV then PCT would be untestable; research on PCT would be impossible. That’s because all we observe is the ECV, which is the basis for our inferences about the perception that is being controlled. But clearly PCT was developed as a testable theory of behavior; there is even a chapter on experimental methods in B:CP. These methods are aimed at determining the ECV around which a particular behavior is organized. Once we know the ECV we also know the controlled perception, which can now be used in models of the behavior under study.

RM:P I don’t know how people got the idea that PCT implies that there is a difference between control of a perception and control of the corresponding ECV. But it’s an idea that is not only wrong but one that, if believed, make PCT research impossible. Thus, as a PCT researcher who would like to see others join in, it’s an idea that I would really like to see discarded asap.

Best

Rick

It always seemed to me that the ECV was the perception (CV) controlled by the observer and imputed to the observed subject by performing the Test (TCV). The observer has to imagine the perceptual capacities and limitations of the subject insofar as they differ from her or his own, but it’s still the observer’s perception.

All we have, observationally, is the ECV, and qo, qi, and d, which can only be specified relative to the ECV. Rick’s spreadsheet is limited to observables, for good reason. The values of p, r, and e are so far inferred, though Henry is getting some actual values.

/B


Richard S. Marken

www.mindreadings.com
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

On Sat, Nov 21, 2015 at 12:41 PM, Martin Taylor mmt-csg@mmtaylor.net wrote:

[Martin Taylor 2015.11.21.10.51]

In many messages over the years, there has been confusion about just

what is controlled in a control loop, the perception or something in
the environment of the controller. Here’s one recent example in
which the confusion is self-contained [From Rick Marken
(2015.11.15.1630)]:

---------quote------
[RM] The input to the perceptual function is ultimately the set of

environmental variables at the sensory surface, call it
(v.1,v.2,…v.n), as per Fig. 1, p. 66 of LCS I. Qi is the function
of this set of variables that corresponds to a controlled
perception: Qi = f(v.1,v.2,…v.n), where f() is the perceptual
function. Another way to look at this is that Qi represents the ** aspect
of the environment** that is controlled when the perceptual
signal, p, that corresponds to Qi, p = f(v.1,v.2,…v.n), is
controlled.

----end quote----



What Rick says is true for the simple isolated control loop with a

passive environmental feedback path, but it isn’t true in general
for controlled perceptions. To see the problem, we can use a simple
illustrative diagram in which Rick’s f(.) is a multiplier “g” with
one input, so p = g*v. Four of these are arranged in a ring, and “v”
for each of them is the sum of two quantities, a “d” from outside
the ring and the “p” from the previous device, like this:

If an odd number of the "g" multipliers is negative, this is a

negative feedback loop, and if the “g” absolute values are similar
and appreciably greater than unity, each “p” value closely tracks
the negative of the following “d”. For example p1 will be nearly
equal to -d2. In other words, every “p” in this homeostatic loop is
a controlled variable with a reference value of -d.

Looking at the loop from the viewpoint of p1 by collapsing the loop

from the output of g2 through the input to g1, we get this:

This looks remarkably like an ordinary control loop, which is not

strange, because it is one. It controls p1 with a reference value of
-d1. The perceptual input function is the multiplier g1 and the
output function is the multiplier g2. The Environmental Feedback
Path also has gain, g3g4, so the loop gain G is g1g2g3g4. If the
absolute magnitude of each “g” is considerably greater than unity, G
is very much greater than any one of them. The disturbance dx is
some combination of d1, d3, and d4. It doesn’t matter what
combination it is for the purposes of the illustration.

The question at issue is what in the environment is controlled?

Whatever it is must be found where the virtual disturbance dx enters
the loop. But where is that? The physical disturbances that
contribute to dx (reference values for other “p” variables) enter at
different places around the loop.

Let's imagine an external observer (EO) interested in using the Test

for the Controlled Variable (TCV) to determine what the controlled
variable p1 corresponds to. What might EO hypothesize and disturb?
Whichever of the observable variables EO chose, the disturbance
would be opposed, not because p1 was controlled, but because it was
the reference value for the negative of the immediately preceding p.
Only by hypothesising the entire complex of the Environmental
Feedback Function (g3, p3, d4, g4, and d1) could EO discover p1 and
its reference value -d2. If EO simply added a random disturbance to
d3 or to d4, the TCV would lead to the conclusion that p2 or p3
respectively was THE controlled variable. p1 would not appear at
all. This simple loop is a control loop for all four p variables,
and if it failed for one (say one of the g values became much less
than unity or changed sign), it would fail for all.

I suppose one might say that the "controlled" environmental variable

is some function of g3, p3, d4, g4, and d1, but this would
contradict Rick’s assertion that it is defined by the perceptual
function, in this case the multiplier g1 (an assertion with which I
fully agree when the Environmental Feedback Function has only one
disturbance entry point and is otherwise passive).

I conclude that consistency in PCT can be maintained only by

restricting the concept of a controlled variable to perceptions, and
that talking about “environmental controlled variables” can only
confuse the issue, no matter how closely the presumed environmental
variable found by the TCV tracks the perceptual variable in a simple
“single-scalar” control loop.

Martin

[Bruce Nevin (2015.11.24. ET)]

Rick Marken (2015.11.22.0950)–

I don’t know how people got the idea that PCT implies that there is a difference between control of a perception and control of the corresponding ECV.

As far as I am aware, Boris Hartman is the only one here who claims that p is controlled but Qi is not.

Martin Taylor (2015.11.24.14.02) –

“PCT” doesn’t imply it. It’s simply a fact of life (and of engineering) that ONLY if the connection from Qi to the perceptual variable is invertible, perfect, and noise-free will there be no difference between the ECV (whatever that may be) and the perception. The perception is controlled, and as a consequence, the environmental variable appears to be. As an approximation, it’s good enough for most purposes, but like Newtonian gravity, it’s not a good foundation for theoretical discussion or precise analysis.

In the equations that I’m familiar with the connection from Qi to p is represented by a constant Ki. Hasn’t that sufficed for implementing simulations, or have I missed something? Any imperfection and noise in the biological implementation is just another disturbance. Disturbances can enter at any point in the loop. If such disturbances could not be countered by the control process in the same way that environmental disturbances are, and if they were great enough to make p depart from its correspondence to the relevant aspect of the environment, as represented by Qi, they would be pathologies making it less likely for that organism to succeed in bringing offspring to reproductive maturity, so there is obvious evolutionary pressure for that coupling to be quite good enough to support good control. That seems to me a pretty strong basis for that coupling being treated as a constant Ki. rather than as a variable subject to significant unpredictable perturbations.

···

On Tue, Nov 24, 2015 at 2:44 PM, Martin Taylor mmt-csg@mmtaylor.net wrote:

[Martin Taylor 2015.11.24.14.02]

[From Rick Marken (2015.11.22.0950)]

No. p1 is the controlled quantity, the perception that tracks the

reference value -d2 closely if all the g values are substantially
greater than unity. In a real control loop, of course, the “g”
multipliers would represent the long-term stable values of the leaky
integrators, just as in the usual analysis of the simple control
loop. The actual loop could not use simple multipliers. When there
are loop delays, simple multipliers inevitably lead to oscillation
and no control. My analysis was of the stable equilibrium values,
and for that, the leaky integrators are well represented by simple
multipliers.

And what's an "ECV"?
Yes, I am interested. It's good to have the spreadsheet example.

What gain and leak rates did you use for the four “g” functions, to
get the “g” multipliers? The effect of d4 and d3 is diminished by
the multiplier ratio each step back round the loop. My analysis
assumed, g>>1, as we do when we do an equilibrium analysis of
the ordinary control loop and assume the loop gain G>>1.

I imagine that in your spreadsheet you have a scalar variable and a

simple multiplier, as I showed in the example. One can’t actually
run the TCV on a single scalar, because there is no function to be
found. But it would be interesting to run a spreadsheet example in
which each of the paths was a vector of, say, three scalars, and
each perceptual function was different, and then run the TCV to see
what you find.

"PCT" doesn't imply it. It's simply a fact of life (and of

engineering) that ONLY if the connection from Qi to the perceptual
variable is invertible, perfect, and noise-free will there be no
difference between the ECV (whatever that may be) and the
perception. The perception is controlled, and as a consequence, the
environmental variable appears to be. As an approximation, it’s good
enough for most purposes, but like Newtonian gravity, it’s not a
good foundation for theoretical discussion or precise analysis.

That's at least equally wrong. I think it might be worth your while

to look a little more closely into the actual conditions for using
the TCV, and the potential and limitations on what you can determine
by using it. You often seem to suggest (planning in imagination)
that you might use the TCV in real-life situations. Sometimes the
conditions are suitable, but much more often, they aren’t. I haven’t
done it, so I am also planning in imagination, but one ought to be
able to run the TCV on your demo of a three-level control system to
find what is being controlled at the top level. You have all the
outputs and disturbances necessary, so it should work. But what
about in a real-life situation in which the circumstances never
recur. In the hammering example, this might be the only time in the
hammerer’s life that he is so angry with his wife that he has to hit
something, and doesn’t want to hit his wife. How can the TCV be used
in that situation?

And how do you use the TCV when control is poor? If you get a poor

compensation of the disturbance by the output, how do you know
whether you haven’t found the controlled variable or you have found
it and the control system doesn’t work very well?

I think your statement is simply equivalent to saying "PCT research

is impossible" which is something I don’t believe, though I do
believe that the control of perception accounts for what we see
people and other organisms do, and that we should carefully study by
all available means just how this works.

Martin
            Bruce Nevin (2015.11.21.20:44 ET) to Martin

Taylor

              BN: Thanks for this nice demonstration of the

difficulty with this distinction [between ECV and p).

          RM: I think Martin aimed to demonstrate that

controlling an ECV is not equivalent to controlling the
perception that corresponds to that ECV. But, in fact, his
demonstration doesn’t demonstrate that at all. What it
demonstrates is that when you put a bunch of disturbances
into the feedback connection between output and input you
control neither the ECV nor p. To be precise, the
disturbance variables, d3 and d4. enter the loop after the
output, p2, and before the input, (p4+d1). See Martin’s
diagram below:

          RM: When these disturbances are present the effect of

output (p2) on input (p4+d1) is constantly changing.
However, if you remove these disturbances from the
feedback function control is restored.

          RM: In this control loop p4+d1 is the controlled

quantity (q.i,or ECV) and g1*(p4+d1) is the controlled
perception, p.

          The only difference between q.i (the ECV) and p is the

scaling factor, g1. But variations in p (p1) are perfectly
correlated with variations in the ECV (p4+d1); the only
difference between p and ECV is that the former is
measured in neural firing rate units and ECV is measures
in physical units; g1 is just a scaling factor that
converts physical units into neural firing rate units .

          RM: I've implemented Martin's model in a spreadsheet,

in case anyone is interested. It allows you to see how
well the perception, p1, and corresponding ECV (p4+d1) are
controlled when the disturbances to the feedback function
(d3 and d4) are in or out of the loop. When these
disturbances are in, control of both p1 and ECV is poor
but the correlation between variations in p1 and the ECV
is 1.0; when these disturbances are out control of both p1
and ECV are excellent; and the correlation between p1 and
ECV is again 1.0.

          RM:P I don't know how people got the idea that PCT

implies that there is a difference between control of a
perception and control of the corresponding ECV.

          But it's an idea that is not only wrong but one that,

if believed, make PCT research impossible.

[Martin Taylor 2015.11.24.14.02]

No. p1 is the controlled quantity, the perception that tracks the

reference value -d2 closely if all the g values are substantially
greater than unity. In a real control loop, of course, the “g”
multipliers would represent the long-term stable values of the leaky
integrators, just as in the usual analysis of the simple control
loop. The actual loop could not use simple multipliers. When there
are loop delays, simple multipliers inevitably lead to oscillation
and no control. My analysis was of the stable equilibrium values,
and for that, the leaky integrators are well represented by simple
multipliers.
And what’s an “ECV”? Yes, I am interested. It’s good to have the spreadsheet example.
What gain and leak rates did you use for the four “g” functions, to
get the “g” multipliers? The effect of d4 and d3 is diminished by
the multiplier ratio each step back round the loop. My analysis
assumed, g>>1, as we do when we do an equilibrium analysis of
the ordinary control loop and assume the loop gain G>>1.
I imagine that in your spreadsheet you have a scalar variable and a
simple multiplier, as I showed in the example. One can’t actually
run the TCV on a single scalar, because there is no function to be
found. But it would be interesting to run a spreadsheet example in
which each of the paths was a vector of, say, three scalars, and
each perceptual function was different, and then run the TCV to see
what you find.
“PCT” doesn’t imply it. It’s simply a fact of life (and of
engineering) that ONLY if the connection from Qi to the perceptual
variable is invertible, perfect, and noise-free will there be no
difference between the ECV (whatever that may be) and the
perception. The perception is controlled, and as a consequence, the
environmental variable appears to be. As an approximation, it’s good
enough for most purposes, but like Newtonian gravity, it’s not a
good foundation for theoretical discussion or precise analysis.
That’s at least equally wrong. I think it might be worth your while
to look a little more closely into the actual conditions for using
the TCV, and the potential and limitations on what you can determine
by using it. You often seem to suggest (planning in imagination)
that you might use the TCV in real-life situations. Sometimes the
conditions are suitable, but much more often, they aren’t. I haven’t
done it, so I am also planning in imagination, but one ought to be
able to run the TCV on your demo of a three-level control system to
find what is being controlled at the top level. You have all the
outputs and disturbances necessary, so it should work. But what
about in a real-life situation in which the circumstances never
recur. In the hammering example, this might be the only time in the
hammerer’s life that he is so angry with his wife that he has to hit
something, and doesn’t want to hit his wife. How can the TCV be used
in that situation?
And how do you use the TCV when control is poor? If you get a poor
compensation of the disturbance by the output, how do you know
whether you haven’t found the controlled variable or you have found
it and the control system doesn’t work very well? I think your statement is simply equivalent to saying “PCT research
is impossible” which is something I don’t believe, though I do
believe that the control of perception accounts for what we see
people and other organisms do, and that we should carefully study by
all available means just how this works.
Martin

···

[From Rick Marken (2015.11.22.0950)]

            Bruce Nevin (2015.11.21.20:44 ET) to Martin

Taylor

              BN: Thanks for this nice demonstration of the

difficulty with this distinction [between ECV and p).

          RM: I think Martin aimed to demonstrate that

controlling an ECV is not equivalent to controlling the
perception that corresponds to that ECV. But, in fact, his
demonstration doesn’t demonstrate that at all. What it
demonstrates is that when you put a bunch of disturbances
into the feedback connection between output and input you
control neither the ECV nor p. To be precise, the
disturbance variables, d3 and d4. enter the loop after the
output, p2, and before the input, (p4+d1). See Martin’s
diagram below:

          RM: When these disturbances are present the effect of

output (p2) on input (p4+d1) is constantly changing.
However, if you remove these disturbances from the
feedback function control is restored.

          RM: In this control loop p4+d1 is the controlled

quantity (q.i,or ECV) and g1*(p4+d1) is the controlled
perception, p.

          The only difference between q.i (the ECV) and p is the

scaling factor, g1. But variations in p (p1) are perfectly
correlated with variations in the ECV (p4+d1); the only
difference between p and ECV is that the former is
measured in neural firing rate units and ECV is measures
in physical units; g1 is just a scaling factor that
converts physical units into neural firing rate units .

          RM: I've implemented Martin's model in a spreadsheet,

in case anyone is interested. It allows you to see how
well the perception, p1, and corresponding ECV (p4+d1) are
controlled when the disturbances to the feedback function
(d3 and d4) are in or out of the loop. When these
disturbances are in, control of both p1 and ECV is poor
but the correlation between variations in p1 and the ECV
is 1.0; when these disturbances are out control of both p1
and ECV are excellent; and the correlation between p1 and
ECV is again 1.0.

          RM:P I don't know how people got the idea that PCT

implies that there is a difference between control of a
perception and control of the corresponding ECV.

          But it's an idea that is not only wrong but one that,

if believed, make PCT research impossible.

[Martin Taylor 2015.11.24.23.37]

So it is, but how realistic do you think that is in the real world

of live organisms?
As I said: "
Indeed, but when they appear between the controlled variable p and
the complex environmental variable (the CEV) to which it
corresponds, all that means is that the CEV is not controlled as
precisely as the perception is.
Yes. That, in essence, is what I said when I said “it’s good enough
for most purposes”. You have read a lot of my writings. How often
have I written in things addressed to PCT newbies that though what
is controlled is perception, it’s what happens in the environment
that matters?
Much more likely to be some kind of approximation to log(Qi) with
some kind of ceiling and some kind of zero-region tolerance zone.
It doesn’t matter, what the function is, if it’s invertible. noise
free, and consistent (which adapting systems are not).
Look, all I’m trying to do is to emphasize that PCT is about The
Control of Perception, something that seems in danger of being
forgotten even on CSGnet.
Martin

···

On 2015/11/24 10:37 PM, Bruce Nevin
wrote:

[Bruce Nevin (2015.11.24. ET)]

          Martin Taylor

(2015.11.24.14.02) –

          "PCT" doesn't imply it.

It’s simply a fact of life (and of engineering) that ONLY
if the connection from Qi to the perceptual variable is
invertible, perfect, and noise-free will there be no
difference between the ECV (whatever that may be) and the
perception. The perception is controlled, and as a
consequence, the environmental variable appears to be. As
an approximation, it’s good enough for most purposes, but
like Newtonian gravity, it’s not a good foundation for
theoretical discussion or precise analysis.

        In the equations that I'm

familiar with the connection from Qi to p is
represented by a constant Ki.

        Hasn't that sufficed for

implementing simulations, or have I missed something?

As an approximation,
it’s good enough for most purposes,". Does anyone claim that the
simulations actually represent what goes on inside the organism?
Even the concept of a neural current has no equivalent in an
actual brain. It’s an analytical convenience, an abstraction that
simply assumes that the effect of a lot of neurons firing with
their own timings is the same as though one super-neuron performed
all the firings, and then smeared them across time so that a
smooth variation was used in further functions. For most purposes,
that’s fine, but if you really want to think about it, Bill just
said that if it’s within a few percent (5%, 2%, I forget) that’s
good enough. And it usually is. But it doesn’t mean that it’s
perfect.

Any imperfection and noise
in the biological implementation is just another
disturbance. Disturbances can enter at any point in the
loop.

If such disturbances could
not be countered by the control process in the same way that
environmental disturbances are, and if they were great
enough to make p depart from its correspondence to
the relevant aspect of the environment, as represented by Qi ,
they would be pathologies making it less likely for that
organism to succeed in bringing offspring to reproductive
maturity, so there is obvious evolutionary pressure for that
coupling to be quite good enough to support good control.

That seems to me a pretty
strong basis for that coupling being treated as a constant Ki .
rather than as a variable subject to significant
unpredictable perturbations.

/Bruce

      On Tue, Nov 24, 2015 at 2:44 PM, Martin

Taylor mmt-csg@mmtaylor.net
wrote:

          [Martin Taylor

2015.11.24.14.02]

[From Rick Marken (2015.11.22.0950)]

                        Bruce Nevin (2015.11.21.20:44

ET) to Martin Taylor

                          BN: Thanks for this nice demonstration

of the difficulty with this distinction
[between ECV and p).

                      RM: I think Martin aimed to demonstrate

that controlling an ECV is not equivalent to
controlling the perception that corresponds to
that ECV. But, in fact, his demonstration
doesn’t demonstrate that at all. What it
demonstrates is that when you put a bunch of
disturbances into the feedback connection
between output and input you control neither
the ECV nor p. To be precise, the disturbance
variables, d3 and d4. enter the loop after the
output, p2, and before the input, (p4+d1). See
Martin’s diagram below:

                      RM: When these disturbances are present the

effect of output (p2) on input (p4+d1) is
constantly changing. However, if you remove
these disturbances from the feedback function
control is restored.

                      RM: In this control loop p4+d1 is the

controlled quantity (q.i,or ECV) and
g1*(p4+d1) is the controlled perception, p.

           No. p1 is the controlled quantity, the perception

that tracks the reference value -d2 closely if all the g
values are substantially greater than unity. In a real
control loop, of course, the “g” multipliers would
represent the long-term stable values of the leaky
integrators, just as in the usual analysis of the simple
control loop. The actual loop could not use simple
multipliers. When there are loop delays, simple
multipliers inevitably lead to oscillation and no control.
My analysis was of the stable equilibrium values, and for
that, the leaky integrators are well represented by simple
multipliers.

          And what's an "ECV"?
                      The only difference between q.i (the ECV)

and p is the scaling factor, g1. But
variations in p (p1) are perfectly correlated
with variations in the ECV (p4+d1); the only
difference between p and ECV is that the
former is measured in neural firing rate units
and ECV is measures in physical units; g1 is
just a scaling factor that converts physical
units into neural firing rate units .

                      RM: I've implemented Martin's model in a

spreadsheet, in case anyone is interested. It
allows you to see how well the perception, p1,
and corresponding ECV (p4+d1) are controlled
when the disturbances to the feedback function
(d3 and d4) are in or out of the loop. When
these disturbances are in, control of both p1
and ECV is poor but the correlation between
variations in p1 and the ECV is 1.0; when
these disturbances are out control of both p1
and ECV are excellent; and the correlation
between p1 and ECV is again 1.0.

           Yes, I am interested. It's good to have the

spreadsheet example. What gain and leak rates did you use
for the four “g” functions, to get the “g” multipliers?
The effect of d4 and d3 is diminished by the multiplier
ratio each step back round the loop. My analysis assumed,
g>>1, as we do when we do an equilibrium analysis of
the ordinary control loop and assume the loop gain
G>>1.

          I imagine that in your spreadsheet you have a scalar

variable and a simple multiplier, as I showed in the
example. One can’t actually run the TCV on a single
scalar, because there is no function to be found. But it
would be interesting to run a spreadsheet example in which
each of the paths was a vector of, say, three scalars, and
each perceptual function was different, and then run the
TCV to see what you find.

                      RM:P I don't know how people got the idea

that PCT implies that there is a difference
between control of a perception and control of
the corresponding ECV.

          "PCT" doesn't imply it. It's simply a fact of life (and of

engineering) that ONLY if the connection from Qi to the
perceptual variable is invertible, perfect, and noise-free
will there be no difference between the ECV (whatever that
may be) and the perception. The perception is controlled,
and as a consequence, the environmental variable appears
to be. As an approximation, it’s good enough for most
purposes, but like Newtonian gravity, it’s not a good
foundation for theoretical discussion or precise analysis.

                      But it's an idea that is not only wrong but

one that, if believed, make PCT research
impossible.

           That's at least equally wrong. I think it might be

worth your while to look a little more closely into the
actual conditions for using the TCV, and the potential and
limitations on what you can determine by using it. You
often seem to suggest (planning in imagination) that you
might use the TCV in real-life situations. Sometimes the
conditions are suitable, but much more often, they aren’t.
I haven’t done it, so I am also planning in imagination,
but one ought to be able to run the TCV on your demo of a
three-level control system to find what is being
controlled at the top level. You have all the outputs and
disturbances necessary, so it should work. But what about
in a real-life situation in which the circumstances never
recur. In the hammering example, this might be the only
time in the hammerer’s life that he is so angry with his
wife that he has to hit something, and doesn’t want to hit
his wife. How can the TCV be used in that situation?

          And how do you use the TCV when control is poor? If you

get a poor compensation of the disturbance by the output,
how do you know whether you haven’t found the controlled
variable or you have found it and the control system
doesn’t work very well?

          I think your statement is simply equivalent to saying "PCT

research is impossible" which is something I don’t
believe, though I do believe that the control of
perception accounts for what we see people and other
organisms do, and that we should carefully study by all
available means just how this works.

              Martin

[From Rick Marken (2015.11.15.1100)]

···

Martin Taylor (2015.11.24.14.02)–

MT: No. p1 is the controlled quantity, the perception that tracks the

reference value -d2 closely if all the g values are substantially
greater than unity.

RM: Usually the term “controlled quantity” is used to refer to the environmental correlate of the perceptual signal. But it doesn’t matter what it’s called; based on your figures (copies below) p1 is, as you say, the perceptual signal that is controlled and p4+d1 is the environmental correlate of that variable, what we call Q.i.

MT: And what’s an “ECV”?

RM: It’s what you call the CEV, I guess, with the letters rearranged. I don’t care for the term much; it implies that there is a unitary variable in the environment that is being controlled. I prefer the more neutral term “controlled quantity”, often symbolized q.i ot Q.i, because I am used to thinking of it as an aspect or function of environmental variables – the variables that physics tells us are “out there”. In the figure above, for example, Qi is a function of two environmental variables, p4 and d1; specifically, Qi = p4 + d1

MT: Yes, I am interested. It's good to have the spreadsheet example.

What gain and leak rates did you use for the four “g” functions, to
get the “g” multipliers?

RM The spreadsheet is an exact translation of your diagram above into a working model. The g values were constant multipliers (I picked 2 but I could make them any size you like; you just said greater than 1, I believe), except for g2, the output function, which was a leaky integrator tuned to give the best possible control.

MT: I imagine that in your spreadsheet you have a scalar variable and a

simple multiplier, as I showed in the example.

RM: Yes, as I say I copied your figure exactly, along with your caveats about the size of the g values.

MT: One can't actually

run the TCV on a single scalar, because there is no function to be
found.

RM: I don’t understand. We test for “scalar” variables all the time. One possible controlled variable in a tracking task is the position of the cursor, for example. Works fine.

MT: But it would be interesting to run a spreadsheet example in

which each of the paths was a vector of, say, three scalars, and
each perceptual function was different, and then run the TCV to see
what you find.

RM: Actually, I do plan to develop the sheet in that direction. But I implemented the sheet based on your figure because I wanted to see whether it demonstrated what I thought you were trying to demonstrate with it: that an aspect of the environment (the controlled quantity , Qi, CEV or ECV) is not necessarily being controlled when the corresponding perception is being controlled. Here’s the little dialog that was the basis for your development of your model:

RM…Qi represents the aspect of the environment that is controlled when the perceptual signal, p, that corresponds to Qi, p = f(v.1,v.2,…v.n), is controlled.

MT: What Rick says is true for the simple isolated control loop with a passive environmental feedback path, but it isn’t true in general for controlled perceptions.

RM: So your model was posted to show that what I said above – that an aspect of the environment is controlled when the perceptual variable that corresponds to that aspect of the environment is controlled – was “not true in general”. My spreadsheet shows that your model demonstrates no such thing.

MT: "PCT" doesn't imply it. It's simply a fact of life (and of

engineering) that ONLY if the connection from Qi to the perceptual
variable is invertible, perfect, and noise-free will there be no
difference between the ECV (whatever that may be) and the
perception.

RM: If that is all you mean when you say that there is a difference between control of a perception and control of the corresponding aspect of the environment (Qi or ECV or CEV) then I have no problem with that. The fact that p = Qi+error has had no practical effect on the ability of the TCV to determine the aspect of the environment that is represented by Qi (and, thus, p), which is the principle aim of the TCV.

MT: The perception is controlled, and as a consequence, the

environmental variable appears to be.

RM: That’s a funny way of putting it. We only suspect that a perception is controlled if we notice that as aspect of the environment – an environmental variable – is controlled. So it’s not that the environmental variable “appears” to be controlled; we can see that it is controlled. And we imagine (using PCT) that this is happening because a perception is being controlled.

MT: That's at least equally wrong. I think it might be worth your while

to look a little more closely into the actual conditions for using
the TCV, and the potential and limitations on what you can determine
by using it. You often seem to suggest (planning in imagination)
that you might use the TCV in real-life situations. Sometimes the
conditions are suitable, but much more often, they aren’t.

RM: You can only do an informal version of the TCV in “real life” conditions. In order to use the TCV as part of the science of studying living control systems you have to use formal scientific methodology: precise manipulation of variables, controlled conditions, all that stuff. But either way, formal or informal, PCT research (the TCV) would be impossible if it were actually true that control of Qi were not the same as control of p.

MT: I haven't

done it, so I am also planning in imagination, but one ought to be
able to run the TCV on your demo of a three-level control system to
find what is being controlled at the top level.

RM: That’s a great suggestion and is very similar to my idea for extending my simulation of your purported demonstration of how control of p is not the same as control of Qi. I want to show that control of p is exactly the same as control of Qi and that this can be demonstrated by showing that getting the right description of Qi will tell you exactly what perception, p, a control system is controlling.

MT: And how do you use the TCV when control is poor?

RM: This is one of the important things to consider when doing research on PCT. You have to use disturbances that are not overwhelming and you have to be aware of the possibility that a hypothetical controlled variable may be varying under disturbance, not because it is not the variable under control but because control is poor. This, and many other things, are a challenge for the study of living control systems. But these challenges don’t contradict the fact that such research would be impossible if it were not true that control of Qi is the same as control of p.

RM: If people don’t want to do PCT research that’s fine with me (though very disappointing) but it’s not necessary to make up stories about control of Qi being different than control of p in order to do that.

Best

Rick

If you get a poor

compensation of the disturbance by the output, how do you know
whether you haven’t found the controlled variable or you have found
it and the control system doesn’t work very well?

I think your statement is simply equivalent to saying "PCT research

is impossible" which is something I don’t believe, though I do
believe that the control of perception accounts for what we see
people and other organisms do, and that we should carefully study by
all available means just how this works.

Martin


Richard S. Marken

www.mindreadings.com
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

          RM: In this control loop p4+d1 is the controlled

quantity (q.i,or ECV) and g1*(p4+d1) is the controlled
perception, p.

          RM:P I don't know how people got the idea that PCT

implies that there is a difference between control of a
perception and control of the corresponding ECV.

          RM: But it's an idea [the idea that control of Qi is not the same as control of p] that is not only wrong but one that,

if believed, make PCT research impossible.

[Martin Taylor 2015.11.25.14.09]

[From Rick Marken (2015.11.15.1100)]

There's no point in doing a census of the uses of the term in CSGnet

postings and publications on PCT. All I can say is that if the term
is usually used that way, the people using it are not talking about
perceptual control theory. Nor are tehy correct in respect of
engineering terms, which PCT is supposed to respect.

"CEV" = Complex Environmental Variable. Is "ECV" Environmental

Complex Variable? One of the Bruces used it a couple of weeks ago to
mean Environmental Controlled variable. I was asking which meaning
you attached to it.

No, it most definitely implies that there is a complex function of

environmental variables being apparently controlled. But if you
believe in PCT or in engineering control, it’s easy to see that the
CEV is not being controlled, and the appearance that it is is a
consequence of the perception being controlled. For the CEV to be
actually controlled, somewhere in the environment there must be a
reference value for the function that defines the CEV and also in
the environment there must be something that produces the difference
between the CEV and its reference value.

"controlled quantity" is hardly a neutral term, when the whole

problem with propagating PCT to the rest of the world is to get them
to understand that perception is what is controlled, not some
abstract function of environmental variables that may or may not be
accessible to another person.

True. But the biggest disturbance to p comes from d3 if d2 is taken

as the reference value. More to the point, if the influence of d2 on
the value of p is 1 unit, then the influence of d3 is 1/g, of d4 is
1/g^2, and of d1 is 1/g^3. For p to be controlled, with a reference
of -d1, g has to be big enough that 1/g isn’t important.

I'm glad you used at least one integrator, because otherwise the

whole thing would oscillate terribly. Anyway, I think I said, and
should have if I didn’t, that is was necessary for g to be
substantially greater than 1.

You have to compare it against other possibilities. What you are

talking about in the tracking task isn’t the TCV, it’s the
parameters in a model in which the TCV is not used because the
controlled variable is presumed to be known.

You are correct when the environmental boundary is placed where I

placed it. The external observer can measure p4+d1, being able to
see each of them. I should have put d1 and perhaps g4 inside the
organism, and d3 outside, because that’s the situation I was trying
to exemplify.

In fact, I have said "no practical effect" or similar more than once

in this and related threads. It does, to some extent, contradict
what I say at the end of this message, but I’m going to let that
contradiction stand for now, because it’s not Black and White.

The minimum possible difference between variation in p and variation

in Qi depends on how good control actually is. If control is
excellent, there can’t be much difference between Qi and the
controlled variable p, and if you hypothesize the right function in
your TCV search, that function will clearly beat out other
possibilities that aren’t too similar to the correct one. But let’s
suppose someone is controlling log(x+y), and you hypothesize that
they are controlling sqrt(x+y) or even just (x+y), their control
will have to be really very good before you could reliably tell
which they were controlling.

Looking at it from a different angle, p at time t always corresponds

to Qi at some earlier time or a weighted average of what Qi was over
some prior period, because it takes time for all the neural inputs
to sum to a value that causes a spike, and then for that spike to
travel to a point where it can influence the perception. If Qi is
changing at all, the autocorrelation function of Qi is not 1.0 at
that time delay, so p(t) != Qi(t).

You could finish your paragraph by adding "So, " followed by my

statement, which would complete the thought. Apart from your first
sentence and the “only” in the second, to do so would make a good
statement of an approach that often works in simple circumstances.
But what about the times when we can’t see what in the environment
is being controlled? That’s when we try to use the TCV to find out
what is being controlled – because we can’t see it otherwise, but
we suspect something is being controlled because the theory suggests
so. But “try to” are the operative words.

How would you use the TCV informally to determine that the guy is

hammering the nail so that he doesn’t hit his wife, rather than
because he wanted a flush surface on the plank? How would you use
the TCV to determine that the stranger you see ringing the doorbell
across the road is visiting his aunt, and not casing the joint to
see if it is empty or testing the doorbell circuitry?

That's logically wrong, but it would be correct if you replaced in

the last line “not the same as” by “unrelated to”. I don’t think
even Boris is claiming that variation in Qi is unrelated to
variation in p. I certainly am not. But I am saying that it is not
the same. It’s “close, but no cigar” except in highly idealized
conditions.

I come back to a problem I have had in different guises in

interchanges with you. You see no distinction between “not quite
perfect” and “utterly worthless”, just as two decades ago you could
see no distinction between “imperfect information” and “no
information”…

If control of Qi is not "the same as" p, for you the whole PCT

enterprise fails. The fact that engineering analysis shows that
control of Qi cannot be the same as, but must be very close to,
control of p, if p is to be controlled well, is irrelevant to you.
“Not the same as” equates to a moral disaster, even though
technically it has to be true.

And the fact that lots of controlled perceptions are not well

controlled also must, I suppose, mean the destruction of the whole
idea of PCT, because under those circumstances, looking from
outside, lots of different functions could equally well be the CEV
of the perception for which control is being attempted. How well
could you tell who I voted for in an election by looking at the
externally observable result of the election? I very definitely do
control a perception of who is elected, with a particular candidate
or party as a reference, but my control of that perception is (by
historical evidence) extremely poor. That may be an extreme example,
but I think excellent control by anyone who isn’t trained to a
professional level in a particular skill is also rather unusual. How
many golfers go around a golf course without having to hit any
putts? If they had perfect control of the golf ball, every shot when
the hole was in range would go into the hole.

I suppose my point is that as scientists we have to work with what

we’ve got, and that is the external observables. But the theory that
accounts for those observables will include unobservable constructs,
and in PCT one of the unobservable constructs is a perception. To do
PCT research we have to assess what should be the result if this or
that is true, including what would happen to the observables if this
perception were controlled in that way and we manipulated the
observables thus and so. Sometimes, one can actually do the
manipulation observe what happens, and compare it with a finely
computed model instantiation of the theory, and that’s great. Bill’s
insight of genius was that what is controlled is nothing on the
outside that we can see, but an unobservable construct on the
inside, the perception, and that by making the giant leap that ALL
intentional behaviour is for the control of these unobservable
constructs called perceptions, a whole lot made sense as a unitary
whole that previously had been a big mess of unrelated statistical
correlations.

At other times, one can't do the required manipulations, and one has

to extrapolate, the way they do in other sciences, and as Bill did
for the general case. The extrapolation may or may not provide
precise descriptions as to what might happen to the observables in
this or that case. You won’t be able to predict who will win the
election by correctly assessing which way I will vote, but you might
if you can correctly assess the way thousands or millions of people
will vote. That doesn’t mean that these thousands or millions of
people aren’t controlling a perception of the candidates. It means
you can’t use the TCV on individuals by looking at what happens to
their controlled variable.

We see the Universe in different ways. You in Black and White, me in

shades of grey, and I don’t imagine that will ever change.

Martin

···
            Martin Taylor

(2015.11.24.14.02)–

            MT: No. p1 is the controlled quantity, the

perception that tracks the reference value -d2 closely
if all the g values are substantially greater than
unity.

          RM: Usually the term "controlled quantity" is used to

refer to the environmental correlate of the perceptual
signal.

                        RM: In this control loop p4+d1 is the

controlled quantity (q.i,or ECV) and
g1*(p4+d1) is the controlled perception, p.

          But it doesn't matter what it's called; based on your

figures (copies below) p1 is, as you say, the perceptual
signal that is controlled and p4+d1 is the environmental
correlate of that variable, what we call Q.i.

            MT: And what's an "ECV"?
          RM: It's what you call the CEV, I guess, with the

letters rearranged.

          I don't care for the term much; it implies that there

is a unitary variable in the environment that is being
controlled.

          I prefer the more neutral term "controlled quantity",

often symbolized q.i ot Q.i, because I am used to thinking
of it as an aspect or function of
environmental variables – the variables that physics
tells us are “out there”.

          In the figure above, for example, Qi is a function of

two environmental variables, p4 and d1; specifically, Qi =
p4 + d1

                            MT: Yes, I am interested. It's good to have the

spreadsheet example. What gain and leak rates did you
use for the four “g” functions, to get the “g”
multipliers?

          RM The spreadsheet is an exact translation of your

diagram above into a working model. The g values were
constant multipliers (I picked 2 but I could make them any
size you like; you just said greater than 1, I believe),
except for g2, the output function, which was a leaky
integrator tuned to give the best possible control.

            MT: I imagine that

in your spreadsheet you have a scalar variable and a
simple multiplier, as I showed in the example.

          RM: Yes, as I say I copied your figure exactly, along

with your caveats about the size of the g values.

            MT: One can't

actually run the TCV on a single scalar, because there
is no function to be found.

          RM: I don't understand. We test for "scalar" variables

all the time. One possible controlled variable in a
tracking task is the position of the cursor, for example.
Works fine.

            MT: But it would be

interesting to run a spreadsheet example in which each
of the paths was a vector of, say, three scalars, and
each perceptual function was different, and then run the
TCV to see what you find.

          RM: Actually, I do plan to develop the sheet in that

direction. But I implemented the sheet based on your
figure because I wanted to see whether it demonstrated
what I thought you were trying to demonstrate with it:
that an aspect of the environment (the controlled quantity
, Qi, CEV or ECV) is not necessarily being controlled when
the corresponding perception is being controlled. Here’s
the little dialog that was the basis for your development
of your model:

RM…Qi represents the aspect of the environment that is controlled when the
perceptual signal, p, that corresponds to Qi, p =
f(v.1,v.2,…v.n), is controlled.

            MT: What Rick says is true for

the simple isolated control loop with a passive
environmental feedback path, but it isn’t true in
general for controlled perceptions.

          RM: So your model was posted to show that what I said

above – that an aspect of the environment is controlled
when the perceptual variable that corresponds to that
aspect of the environment is controlled – was “not true
in general”. My spreadsheet shows that your model
demonstrates no such thing.

            MT: "PCT" doesn't imply it. It's simply a fact of life

(and of engineering) that ONLY if the connection from Qi
to the perceptual variable is invertible, perfect, and
noise-free will there be no difference between the ECV
(whatever that may be) and the perception.

          RM: If that is all you mean when you say that there is

a difference between control of a perception and control
of the corresponding aspect of the environment (Qi or ECV
or CEV) then I have no problem with that. The fact that p
= Qi+error has had no practical effect on the ability of
the TCV to determine the aspect of the environment that is
represented by Qi (and, thus, p), which is the principle
aim of the TCV.

                      RM:P I don't know how people got the idea

that PCT implies that there is a difference
between control of a perception and control of
the corresponding ECV.

            MT: The perception

is controlled, and as a consequence, the environmental
variable appears to be.

          RM: That's a funny way of putting it. We only suspect

that a perception is controlled if we notice that as
aspect of the environment – an environmental variable –
is controlled. So it’s not that the environmental variable
“appears” to be controlled; we can see that it is
controlled. And we imagine (using PCT) that this is
happening because a perception is being controlled.

            MT: That's at least equally wrong. I think it

might be worth your while to look a little more closely
into the actual conditions for using the TCV, and the
potential and limitations on what you can determine by
using it. You often seem to suggest (planning in
imagination) that you might use the TCV in real-life
situations. Sometimes the conditions are suitable, but
much more often, they aren’t.

          RM: You can only do an informal version of the TCV in

“real life” conditions.

                        RM: But it's an idea [the idea that

control of Qi is not the same as control of
p] that is not only wrong but one that, if
believed, make PCT research impossible.

          In order to use the TCV as part of the science of

studying living control systems you have to use formal
scientific methodology: precise manipulation of variables,
controlled conditions, all that stuff. But either way,
formal or informal, PCT research (the TCV) would be
impossible if it were actually true that control of Qi
were not the same as control of p.

          RM: If people don't want to do PCT research that's fine

with me (though very disappointing) but it’s not necessary
to make up stories about control of Qi being different
than control of p in order to do that.

[From Rick Marken (2015.11.25.1620)]

···

Martin Taylor (2015.11.25.14.09)–

MT: “CEV” = Complex Environmental Variable.

RM: Yes, that’s much better than ECV (Environmental Controlled Variable).

MT: No, it most definitely implies that there is a complex function of

environmental variables being apparently controlled.

RM: Yes, I think it’s a good term. But what’s with the emphasis on “apparently”? Does CEV just refer to aspects of the environment that have not yet been determined to be under control? If so, then what do we call the controlled aspect of the environment? Can we go back to calling it the controlled quantity, like Bill did? Or how about CCEV: Controlled Complex Environmental Variable?

MT: But if you

believe in PCT or in engineering control, it’s easy to see that the
CEV is not being controlled, and the appearance that it is is a
consequence of the perception being controlled.

RM: I believe in PCT and engineering control and it is not easy, indeed, it is impossible for me to see that the CEV is not being controlled (unless it’s not a CEV that is controlled). Are you saying that the temperature in my living room is not being controlled, it just appears to be because the thermostat is controlling a perception?

MT: For the CEV to be

actually controlled, somewhere in the environment there must be a
reference value for the function that defines the CEV and also in
the environment there must be something that produces the difference
between the CEV and its reference value.

RM: I am all astonishment. If this is what you think then we have a disconnect that is so great that it is truly unbridgeable.

MT: "controlled quantity" is hardly a neutral term, when the whole

problem with propagating PCT to the rest of the world is to get them
to understand that perception is what is controlled, not some
abstract function of environmental variables that may or may not be
accessible to another person.

RM: In PCT a perception is an abstract function of environmental variables. And it has to be accessible to a researcher, either via his own perceptual system or via instrumentation, or he has no way of knowing that it even exists. That’s how we know that bats control a perception of the echos of ultra high frequency acoustical signals, for example. I think before we try to get the rest of the world to understand that perception is what is controlled by living systems we should get them to understand that what living organisms do is control. Other wise I think they will just be baffled getting an answer to a question that they never had.

MT: I'm glad you used at least one integrator, because otherwise the

whole thing would oscillate terribly. Anyway, I think I said, and
should have if I didn’t, that is was necessary for g to be
substantially greater than 1.

RM: Ok, I can make g as big as you like. How will I know when they are sufficiently greater than 1.0?

MT: You are correct when the environmental boundary is placed where I

placed it. The external observer can measure p4+d1, being able to
see each of them. I should have put d1 and perhaps g4 inside the
organism, and d3 outside, because that’s the situation I was trying
to exemplify.

RM: Then why did you put the environmental boundary where you did. I wasted a lot of time trying to see if your little model actually demonstrates what you claim it did, and it didn’t. So now you say that what you really meant was that d1 and maybe g4 should have been inside the organism. How about taking a deep breath and posting a nice clear concise post explaining what exactly you want to show (which is presumably that a CEV is only “apparently” controlled when the perception that corresponds to that CEV is controlled) , what exactly that means, and how to “wire up” the model that tests this. I’m getting a little wind chill over here from all the hand waving. And that’s no mean trick considering that I live in California and no one out here even knows what wind chill is!

MT: The minimum possible difference between variation in p and variation

in Qi depends on how good control actually is.

RM: Please explain what you mean by the "difference between variation in p and variation in Q.i. How do you measure it?

MT: If control is

excellent, there can’t be much difference between Qi and the
controlled variable p, and if you hypothesize the right function in
your TCV search, that function will clearly beat out other
possibilities that aren’t too similar to the correct one. But let’s
suppose someone is controlling log(x+y), and you hypothesize that
they are controlling sqrt(x+y) or even just (x+y), their control
will have to be really very good before you could reliably tell
which they were controlling.

RM: Yes, it can be very difficult to accurately determine exactly what perceptual variable a person is controlling. But then it was difficult to measure the mass of an electron or the speed of light. Experimental science is hard. But the problem of distinguishing control of log(x+y) from control of sqrt(x+y) has nothing to do with control of Qi being different than control of p. When you calculate log(x+y) and sqrt(x+y) you are calculating possible values of the controlled quantity, Qi – variables that are functions (sqrt and log in this case) of environmental variables (x and y in this case). If we find evidence that Qi = log(x+y) is under control then we conclude that p is proportional to log (x+y). That’s how we do the TCV. We develop hypotheses about the perceptions under control by proposing functions of environmental variables that we think are approximations to the perceptual functions that produce the perceptions that are actually under control. Doing this accurately is, indeed, going to be a challenge. But it can be done. I think a good example of using the TCV to discriminate two very nearly identical hypotheses about the function of environmental variables that corresponds to the perceptual function that produces the perception controlled in an experiment is described in Chapter 4 of “Doing Research on Purpose”. Indeed, I used your data to do this TCV. You really should read that Chapter or read it again if you’ve read it before.

MT: If control of Qi is not "the same as" p, for you the whole PCT

enterprise fails.

RM: No, if someone says that control of Q.i is not the same as control of p then they just don’t understand the difference between fact and theory. Control of Qi is an observable fact; control of p is a theory that explains that fact. Your saying that control of Qi is not the same as control of p doesn’t affect how I go about my business because in my business – doing PCT research --that’s just nonsense.

MT: The fact that engineering analysis shows that

control of Qi cannot be the same as, but must be very close to,
control of p, if p is to be controlled well, is irrelevant to you.

RM: Yes. It has never been a factor in my work. When my models are picking up 99% of the predictable variable in my research I assume that some of that tiny amount of unknown variance is a result of noise in the system or a still very slightly imprecise definition of the controlled perception.

MT: And the fact that lots of controlled perceptions are not well

controlled also must, I suppose, mean the destruction of the whole
idea of PCT,

RM: Not at all; As I said earlier, it’s something that someone doing research in PCT has to be aware of so that they can design methods that can determine whether an apparently poor example of control results from an incorrect definition of the controlled variable, a varying reference for a correctly defined controlled variable or from poor control of a correctly defined controlled variable. There was ways to figure this out which of these is the case.

MT: because under those circumstances, looking from

outside, lots of different functions could equally well be the CEV
of the perception for which control is being attempted.

RM: Right. So what is needed is ingenuity. That’s what makes research fun!

MT: I suppose my point is that as scientists we have to work with what

we’ve got, and that is the external observables.

RM: I couldn’t agree more. Though we also get to work with our imagination and reason – ie, theory.

MT: But the theory that

accounts for those observables will include unobservable constructs, and in PCT one of the unobservable constructs is a perception.

RM: Exactly!

MT: To do

PCT research we have to assess what should be the result if this or
that is true, including what would happen to the observables if this
perception were controlled in that way and we manipulated the
observables thus and so.

RM: Righto!!

MT: Sometimes, one can actually do the

manipulation observe what happens, and compare it with a finely
computed model instantiation of the theory, and that’s great. Bill’s
insight of genius was that what is controlled is nothing on the
outside that we can see, but an unobservable construct on the
inside, the perception,

RM: Whoops! Off the tracks. There is nothing in PCT about “nothing on the outside that we can see” being controlled. Have you ever noticed that there’s a variable out there in the environment side of the control diagram called the controlled quantity! All that stuff on the environment side of the diagram is the “observables” that the theory of control of perception explains.

MT: and that by making the giant leap that ALL

intentional behaviour is for the control of these unobservable
constructs called perceptions, a whole lot made sense as a unitary
whole that previously had been a big mess of unrelated statistical
correlations.

RM: What you are leaving out is what the “whole lot” was that was made sense of. The “whole lot” was the “observables”! And the most important of those observables that was made sense of by PCT is the existence of reference states for controlled variables (as Bill mentions on p. 175 of LCS I).

Best

Rick


Richard S. Marken

www.mindreadings.com
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

          RM: I don't care for the term [ECV] much; it implies that there

is a unitary variable in the environment that is being
controlled.

          RM: So your model was posted to show that what I said

above – that an aspect of the environment is controlled
when the perceptual variable that corresponds to that
aspect of the environment is controlled – was “not true
in general”. My spreadsheet shows that your model
demonstrates no such thing.

[Bruce Nevin (2015.11.25.20:04 ET)]

Martin Taylor 2015.11.24.23.37 –

when [disturbances] appear between the controlled variable p and the complex environmental variable (the CEV) to which it corresponds, all that means is that the CEV is not controlled as precisely as the perception is

Yes, but it is controlled, however imperfectly that may be.

You’re denying my assertion that a disturbance at that point in the loop can be resisted. In the case of a pathology, it is certainly the case that control is impaired, as I said. Example: before the invention of corrective lenses, my astigmatism would require me to rely on others to make out details of a scene and report them, as would my relatively slight myopia.

Look, all I’m trying to do is to emphasize that PCT is about The Control of Perception, something that seems in danger of being forgotten even on CSGnet.

Yes, but consider the context. This was in response to a person who denies that Qi is controlled at all. I agree with Rick: if that were the case, there would be no way for an observer to notice the fact of control. No stabilization of the environment against disturbances would be perceptible to anyone except the organism that was doing the controlling.

···

On Tue, Nov 24, 2015 at 11:59 PM, Martin Taylor mmt-csg@mmtaylor.net wrote:

[Martin Taylor 2015.11.24.23.37]

  On 2015/11/24 10:37 PM, Bruce Nevin

wrote:

[Bruce Nevin (2015.11.24. ET)]

          Martin Taylor

(2015.11.24.14.02) –

          "PCT" doesn't imply it.

It’s simply a fact of life (and of engineering) that ONLY
if the connection from Qi to the perceptual variable is
invertible, perfect, and noise-free will there be no
difference between the ECV (whatever that may be) and the
perception. The perception is controlled, and as a
consequence, the environmental variable appears to be. As
an approximation, it’s good enough for most purposes, but
like Newtonian gravity, it’s not a good foundation for
theoretical discussion or precise analysis.

        In the equations that I'm

familiar with the connection from Qi to p is
represented by a constant Ki.

So it is, but how realistic do you think that is in the real world

of live organisms?

        Hasn't that sufficed for

implementing simulations, or have I missed something?

As I said: " As an approximation,

it’s good enough for most purposes,". Does anyone claim that the
simulations actually represent what goes on inside the organism?
Even the concept of a neural current has no equivalent in an
actual brain. It’s an analytical convenience, an abstraction that
simply assumes that the effect of a lot of neurons firing with
their own timings is the same as though one super-neuron performed
all the firings, and then smeared them across time so that a
smooth variation was used in further functions. For most purposes,
that’s fine, but if you really want to think about it, Bill just
said that if it’s within a few percent (5%, 2%, I forget) that’s
good enough. And it usually is. But it doesn’t mean that it’s
perfect.

Any imperfection and noise
in the biological implementation is just another
disturbance. Disturbances can enter at any point in the
loop.

Indeed, but when they appear between the controlled variable p and

the complex environmental variable (the CEV) to which it
corresponds, all that means is that the CEV is not controlled as
precisely as the perception is.

If such disturbances could
not be countered by the control process in the same way that
environmental disturbances are, and if they were great
enough to make p depart from its correspondence to
the relevant aspect of the environment, as represented by Qi ,
they would be pathologies making it less likely for that
organism to succeed in bringing offspring to reproductive
maturity, so there is obvious evolutionary pressure for that
coupling to be quite good enough to support good control.

Yes. That, in essence, is what I said when I said "it's good enough

for most purposes". You have read a lot of my writings. How often
have I written in things addressed to PCT newbies that though what
is controlled is perception, it’s what happens in the environment
that matters?

That seems to me a pretty
strong basis for that coupling being treated as a constant Ki .
rather than as a variable subject to significant
unpredictable perturbations.

Much more likely to be some kind  of approximation to log(Qi) with

some kind of ceiling and some kind of zero-region tolerance zone.

It doesn't matter, what the function is, if it's invertible. noise

free, and consistent (which adapting systems are not).

Look, all I'm trying to do is to emphasize that PCT is about The

Control of Perception, something that seems in danger of being
forgotten even on CSGnet.

Martin

/Bruce

      On Tue, Nov 24, 2015 at 2:44 PM, Martin

Taylor mmt-csg@mmtaylor.net
wrote:

          [Martin Taylor

2015.11.24.14.02]

[From Rick Marken (2015.11.22.0950)]

           No. p1 is the controlled quantity, the perception

that tracks the reference value -d2 closely if all the g
values are substantially greater than unity. In a real
control loop, of course, the “g” multipliers would
represent the long-term stable values of the leaky
integrators, just as in the usual analysis of the simple
control loop. The actual loop could not use simple
multipliers. When there are loop delays, simple
multipliers inevitably lead to oscillation and no control.
My analysis was of the stable equilibrium values, and for
that, the leaky integrators are well represented by simple
multipliers.

          And what's an "ECV"?
           Yes, I am interested. It's good to have the

spreadsheet example. What gain and leak rates did you use
for the four “g” functions, to get the “g” multipliers?
The effect of d4 and d3 is diminished by the multiplier
ratio each step back round the loop. My analysis assumed,
g>>1, as we do when we do an equilibrium analysis of
the ordinary control loop and assume the loop gain
G>>1.

          I imagine that in your spreadsheet you have a scalar

variable and a simple multiplier, as I showed in the
example. One can’t actually run the TCV on a single
scalar, because there is no function to be found. But it
would be interesting to run a spreadsheet example in which
each of the paths was a vector of, say, three scalars, and
each perceptual function was different, and then run the
TCV to see what you find.

          "PCT" doesn't imply it. It's simply a fact of life (and of

engineering) that ONLY if the connection from Qi to the
perceptual variable is invertible, perfect, and noise-free
will there be no difference between the ECV (whatever that
may be) and the perception. The perception is controlled,
and as a consequence, the environmental variable appears
to be. As an approximation, it’s good enough for most
purposes, but like Newtonian gravity, it’s not a good
foundation for theoretical discussion or precise analysis.

           That's at least equally wrong. I think it might be

worth your while to look a little more closely into the
actual conditions for using the TCV, and the potential and
limitations on what you can determine by using it. You
often seem to suggest (planning in imagination) that you
might use the TCV in real-life situations. Sometimes the
conditions are suitable, but much more often, they aren’t.
I haven’t done it, so I am also planning in imagination,
but one ought to be able to run the TCV on your demo of a
three-level control system to find what is being
controlled at the top level. You have all the outputs and
disturbances necessary, so it should work. But what about
in a real-life situation in which the circumstances never
recur. In the hammering example, this might be the only
time in the hammerer’s life that he is so angry with his
wife that he has to hit something, and doesn’t want to hit
his wife. How can the TCV be used in that situation?

          And how do you use the TCV when control is poor? If you

get a poor compensation of the disturbance by the output,
how do you know whether you haven’t found the controlled
variable or you have found it and the control system
doesn’t work very well?

          I think your statement is simply equivalent to saying "PCT

research is impossible" which is something I don’t
believe, though I do believe that the control of
perception accounts for what we see people and other
organisms do, and that we should carefully study by all
available means just how this works.

              Martin
                        Bruce Nevin (2015.11.21.20:44

ET) to Martin Taylor

                          BN: Thanks for this nice demonstration

of the difficulty with this distinction
[between ECV and p).

                      RM: I think Martin aimed to demonstrate

that controlling an ECV is not equivalent to
controlling the perception that corresponds to
that ECV. But, in fact, his demonstration
doesn’t demonstrate that at all. What it
demonstrates is that when you put a bunch of
disturbances into the feedback connection
between output and input you control neither
the ECV nor p. To be precise, the disturbance
variables, d3 and d4. enter the loop after the
output, p2, and before the input, (p4+d1). See
Martin’s diagram below:

                      RM: When these disturbances are present the

effect of output (p2) on input (p4+d1) is
constantly changing. However, if you remove
these disturbances from the feedback function
control is restored.

                      RM: In this control loop p4+d1 is the

controlled quantity (q.i,or ECV) and
g1*(p4+d1) is the controlled perception, p.

                      The only difference between q.i (the ECV)

and p is the scaling factor, g1. But
variations in p (p1) are perfectly correlated
with variations in the ECV (p4+d1); the only
difference between p and ECV is that the
former is measured in neural firing rate units
and ECV is measures in physical units; g1 is
just a scaling factor that converts physical
units into neural firing rate units .

                      RM: I've implemented Martin's model in a

spreadsheet, in case anyone is interested. It
allows you to see how well the perception, p1,
and corresponding ECV (p4+d1) are controlled
when the disturbances to the feedback function
(d3 and d4) are in or out of the loop. When
these disturbances are in, control of both p1
and ECV is poor but the correlation between
variations in p1 and the ECV is 1.0; when
these disturbances are out control of both p1
and ECV are excellent; and the correlation
between p1 and ECV is again 1.0.

                      RM:P I don't know how people got the idea

that PCT implies that there is a difference
between control of a perception and control of
the corresponding ECV.

                      But it's an idea that is not only wrong but

one that, if believed, make PCT research
impossible.

[Bruce Nevin (2015.11.25.22:20 ET)]

Rick Marken (2015.11.25.1620)–

Martin Taylor (2015.11.25.14.09)–

MT: For the CEV to be actually controlled, somewhere in the environment there must be a reference value for the function that defines the CEV and also in the environment there must be something that produces the difference between the CEV and its reference value.

RM: I am all astonishment. If this is what you think then we have a disconnect that is so great that it is truly unbridgeable.

Well, Martin is correct here, but maybe not in the intended way. Yes, when I observe behavior, conduct the test, and narrow down what appears to be the controlled variable, out there in my environment there is “a reference value for the function that defines the CEV and […] something that produces the difference between the CEV and its reference value.” The reference value and the control hierarchy that produces the difference between the CEV and its reference value are inside the subject whose behavior I observe out there in my environment.

/Bruce

···

On Wed, Nov 25, 2015 at 7:20 PM, Richard Marken rsmarken@gmail.com wrote:

[From Rick Marken (2015.11.25.1620)]

Martin Taylor (2015.11.25.14.09)–

MT: “CEV” = Complex Environmental Variable.

RM: Yes, that’s much better than ECV (Environmental Controlled Variable).

MT: No, it most definitely implies that there is a complex function of

environmental variables being apparently controlled.

RM: Yes, I think it’s a good term. But what’s with the emphasis on “apparently”? Does CEV just refer to aspects of the environment that have not yet been determined to be under control? If so, then what do we call the controlled aspect of the environment? Can we go back to calling it the controlled quantity, like Bill did? Or how about CCEV: Controlled Complex Environmental Variable?

MT: But if you

believe in PCT or in engineering control, it’s easy to see that the
CEV is not being controlled, and the appearance that it is is a
consequence of the perception being controlled.

RM: I believe in PCT and engineering control and it is not easy, indeed, it is impossible for me to see that the CEV is not being controlled (unless it’s not a CEV that is controlled). Are you saying that the temperature in my living room is not being controlled, it just appears to be because the thermostat is controlling a perception?

MT: For the CEV to be

actually controlled, somewhere in the environment there must be a
reference value for the function that defines the CEV and also in
the environment there must be something that produces the difference
between the CEV and its reference value.

RM: I am all astonishment. If this is what you think then we have a disconnect that is so great that it is truly unbridgeable.

MT: "controlled quantity" is hardly a neutral term, when the whole

problem with propagating PCT to the rest of the world is to get them
to understand that perception is what is controlled, not some
abstract function of environmental variables that may or may not be
accessible to another person.

RM: In PCT a perception is an abstract function of environmental variables. And it has to be accessible to a researcher, either via his own perceptual system or via instrumentation, or he has no way of knowing that it even exists. That’s how we know that bats control a perception of the echos of ultra high frequency acoustical signals, for example. I think before we try to get the rest of the world to understand that perception is what is controlled by living systems we should get them to understand that what living organisms do is control. Other wise I think they will just be baffled getting an answer to a question that they never had.

MT: I'm glad you used at least one integrator, because otherwise the

whole thing would oscillate terribly. Anyway, I think I said, and
should have if I didn’t, that is was necessary for g to be
substantially greater than 1.

RM: Ok, I can make g as big as you like. How will I know when they are sufficiently greater than 1.0?

MT: You are correct when the environmental boundary is placed where I

placed it. The external observer can measure p4+d1, being able to
see each of them. I should have put d1 and perhaps g4 inside the
organism, and d3 outside, because that’s the situation I was trying
to exemplify.

RM: Then why did you put the environmental boundary where you did. I wasted a lot of time trying to see if your little model actually demonstrates what you claim it did, and it didn’t. So now you say that what you really meant was that d1 and maybe g4 should have been inside the organism. How about taking a deep breath and posting a nice clear concise post explaining what exactly you want to show (which is presumably that a CEV is only “apparently” controlled when the perception that corresponds to that CEV is controlled) , what exactly that means, and how to “wire up” the model that tests this. I’m getting a little wind chill over here from all the hand waving. And that’s no mean trick considering that I live in California and no one out here even knows what wind chill is!

MT: The minimum possible difference between variation in p and variation

in Qi depends on how good control actually is.

RM: Please explain what you mean by the "difference between variation in p and variation in Q.i. How do you measure it?

MT: If control is

excellent, there can’t be much difference between Qi and the
controlled variable p, and if you hypothesize the right function in
your TCV search, that function will clearly beat out other
possibilities that aren’t too similar to the correct one. But let’s
suppose someone is controlling log(x+y), and you hypothesize that
they are controlling sqrt(x+y) or even just (x+y), their control
will have to be really very good before you could reliably tell
which they were controlling.

RM: Yes, it can be very difficult to accurately determine exactly what perceptual variable a person is controlling. But then it was difficult to measure the mass of an electron or the speed of light. Experimental science is hard. But the problem of distinguishing control of log(x+y) from control of sqrt(x+y) has nothing to do with control of Qi being different than control of p. When you calculate log(x+y) and sqrt(x+y) you are calculating possible values of the controlled quantity, Qi – variables that are functions (sqrt and log in this case) of environmental variables (x and y in this case). If we find evidence that Qi = log(x+y) is under control then we conclude that p is proportional to log (x+y). That’s how we do the TCV. We develop hypotheses about the perceptions under control by proposing functions of environmental variables that we think are approximations to the perceptual functions that produce the perceptions that are actually under control. Doing this accurately is, indeed, going to be a challenge. But it can be done. I think a good example of using the TCV to discriminate two very nearly identical hypotheses about the function of environmental variables that corresponds to the perceptual function that produces the perception controlled in an experiment is described in Chapter 4 of “Doing Research on Purpose”. Indeed, I used your data to do this TCV. You really should read that Chapter or read it again if you’ve read it before.

MT: If control of Qi is not "the same as" p, for you the whole PCT

enterprise fails.

RM: No, if someone says that control of Q.i is not the same as control of p then they just don’t understand the difference between fact and theory. Control of Qi is an observable fact; control of p is a theory that explains that fact. Your saying that control of Qi is not the same as control of p doesn’t affect how I go about my business because in my business – doing PCT research --that’s just nonsense.

MT: The fact that engineering analysis shows that

control of Qi cannot be the same as, but must be very close to,
control of p, if p is to be controlled well, is irrelevant to you.

RM: Yes. It has never been a factor in my work. When my models are picking up 99% of the predictable variable in my research I assume that some of that tiny amount of unknown variance is a result of noise in the system or a still very slightly imprecise definition of the controlled perception.

MT: And the fact that lots of controlled perceptions are not well

controlled also must, I suppose, mean the destruction of the whole
idea of PCT,

RM: Not at all; As I said earlier, it’s something that someone doing research in PCT has to be aware of so that they can design methods that can determine whether an apparently poor example of control results from an incorrect definition of the controlled variable, a varying reference for a correctly defined controlled variable or from poor control of a correctly defined controlled variable. There was ways to figure this out which of these is the case.

MT: because under those circumstances, looking from

outside, lots of different functions could equally well be the CEV
of the perception for which control is being attempted.

RM: Right. So what is needed is ingenuity. That’s what makes research fun!

MT: I suppose my point is that as scientists we have to work with what

we’ve got, and that is the external observables.

RM: I couldn’t agree more. Though we also get to work with our imagination and reason – ie, theory.

MT: But the theory that

accounts for those observables will include unobservable constructs, and in PCT one of the unobservable constructs is a perception.

RM: Exactly!

MT: To do

PCT research we have to assess what should be the result if this or
that is true, including what would happen to the observables if this
perception were controlled in that way and we manipulated the
observables thus and so.

RM: Righto!!

MT: Sometimes, one can actually do the

manipulation observe what happens, and compare it with a finely
computed model instantiation of the theory, and that’s great. Bill’s
insight of genius was that what is controlled is nothing on the
outside that we can see, but an unobservable construct on the
inside, the perception,

RM: Whoops! Off the tracks. There is nothing in PCT about “nothing on the outside that we can see” being controlled. Have you ever noticed that there’s a variable out there in the environment side of the control diagram called the controlled quantity! All that stuff on the environment side of the diagram is the “observables” that the theory of control of perception explains.

MT: and that by making the giant leap that ALL

intentional behaviour is for the control of these unobservable
constructs called perceptions, a whole lot made sense as a unitary
whole that previously had been a big mess of unrelated statistical
correlations.

RM: What you are leaving out is what the “whole lot” was that was made sense of. The “whole lot” was the “observables”! And the most important of those observables that was made sense of by PCT is the existence of reference states for controlled variables (as Bill mentions on p. 175 of LCS I).

Best

Rick


Richard S. Marken

www.mindreadings.com
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

          RM: I don't care for the term [ECV] much; it implies that there

is a unitary variable in the environment that is being
controlled.

          RM: So your model was posted to show that what I said

above – that an aspect of the environment is controlled
when the perceptual variable that corresponds to that
aspect of the environment is controlled – was “not true
in general”. My spreadsheet shows that your model
demonstrates no such thing.

[From Rick Marken (2015.11.26.0920)]

···

Bruce Nevin (2015.11.25.22:20 ET)

MT: For the CEV to be actually controlled, somewhere in the environment there must be a reference value for the function that defines the CEV and also in the environment there must be something that produces the difference between the CEV and its reference value.

RM: I am all astonishment. If this is what you think then we have a disconnect that is so great that it is truly unbridgeable.

BN: Well, Martin is correct here, but maybe not in the intended way. Yes, when I observe behavior, conduct the test, and narrow down what appears to be the controlled variable, out there in my environment there is “a reference value for the function that defines the CEV and […] something that produces the difference between the CEV and its reference value.” The reference value and the control hierarchy that produces the difference between the CEV and its reference value are inside the subject whose behavior I observe out there in my environment.

RM: I see, so when Martin said that “somewhere in the environment there must be a reference value for the function that defines the CEV” he was talking about the environment of the CEV, which includes the system (subject) doing the controlling. So what Martin meant is that the reference value for “the function that defines the CEV” (which is the perceptual function in PCT) is the reference signal inside the controlling system. And the “something that produces the difference between CEV and its reference value” is also inside the controlling system; presumably it’s the comparator that calculates the difference between the perceptual signal (the representation of the CEV inside the controlling system) and reference signal.

RM: That’s a fairly odd way to say it; it would have made more sense to me if Martin had said: “For the CEV to be actually controlled, somewhere in the controlling system there must be a reference specification for the value for the perceptual representation of the CEV and also in that system there must be something that produces the difference between this perception and its reference value”. It was the word “environment” that made things kind of confusing.

RM: Happy Thanksgiving everyone.

Best

Rick

Richard S. Marken

www.mindreadings.com
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

[Martin Taylor 2015.11.26.14.54]

[Bruce Nevin (2015.11.25.20:04 ET)]

Martin Taylor 2015.11.24.23.37 –

        when [disturbances] appear between the controlled

variable p and the complex environmental variable (the CEV)
to which it corresponds, all that means is that the CEV is
not controlled as precisely as the perception is

Yes, but it is controlled, however imperfectly that may be.

I deny that.

The appearance of control of teh Complex Environmental Variable is,

if I understood you correctly when you used the term in another
context, a spandrel. The appearance that is is controlled is a
consequence of something else truly being controlled. It used to be
quite obvious that phlogiston flowed in and out of objects, and even
now we observe heat flowing in and out of objects, but there’s
“really” no flow of anything. All there is is a bunch of molecules
moving around and beating the hell out of each other. It’s the same
kind of thing. When we talk casually, I have no objection to saying
that the CEV is controlled. I do it myself quite often. But when we
want to explain the theory to anyone in or out of CSGnet, one of the
very first things we have to explain is that the CEV is not
controlled, however much it looks as though it is. The perception of
it is controlled, and that is the reason it looks as though it is
controlled (as also is everyone else’s perception of anything
correlated with the CEV, though that fact is never mentioned in this
discussion; why should the argument not be that the controller is
controlling what some undefined other person is perceiving? The
logic is the same.).

      You're denying my assertion that a disturbance at that

point in the loop can be resisted.

No I most definitely am not!!! If it were not resisted, how could

the corresponding perception be controlled?

      In the case of a pathology, it is certainly the case that

control is impaired, as I said. Example: before the invention
of corrective lenses, my astigmatism would require me to rely
on others to make out details of a scene and report them, as
would my relatively slight myopia.

        Look, all I'm trying to do is to emphasize that PCT is

about The Control of Perception, something that seems in
danger of being forgotten even on CSGnet.

      Yes, but consider the context. This was in response to a

person who denies that Qi is controlled at all.

I guess we bring different prior assumptions to our reading of what

always must be ambiguous, and that’s foubly true of someone whose
first language is no variety of English. I often disagree with what
Boris says, but on this I have read Boris as simply pointing out
that the output affects Qi in order that perception is controlled.
I say the same, but I am not usually told I don’t understand PCT –
at least not in the 20 or so years since Bill challenged Rick when
he made that claim, saying something along the lines of “Who do you
think you are saying doesn’t understand PCT?”

      I agree with Rick: if that were the case, there would be

no way for an observer to notice the fact of control. No
stabilization of the environment against disturbances would be
perceptible to anyone except the organism that was doing the
controlling.

Why not? I never perceived you as being of the all-or-none Black or

White persuasion, but here you are saying that if (as must be the
case) an observer has a different set of inputs to the senses than
the person doing the controlling, no matter how similar their inputs
and perceptual functions may be, what the observer sees must be
totally unrelated to what the controller sees. Sure, if the
controller is controlling the placement of a glass on a table, and
the observer is looking at the degree to which a door is open, the
observer will say there’s no control. But that’s not what we are
talking about, is it? The observer sees the glass on the table, and
if he wants to know whether the controller cared where it was
placed, the observer can become an experimenter and move it. The
fact that they see it from different angles may matter, but probably
doesn’t.

Martin

···

On Tue, Nov 24, 2015 at 11:59 PM,
Martin Taylor mmt-csg@mmtaylor.net
wrote:

          [Martin Taylor

2015.11.24.23.37]

On 2015/11/24 10:37 PM, Bruce Nevin wrote:

[Bruce Nevin (2015.11.24. ET)]

                      Martin Taylor

(2015.11.24.14.02) –

                      "PCT" doesn't

imply it. It’s simply a fact of life (and of
engineering) that ONLY if the connection from
Qi to the perceptual variable is invertible,
perfect, and noise-free will there be no
difference between the ECV (whatever that may
be) and the perception. The perception is
controlled, and as a consequence, the
environmental variable appears to be. As an
approximation, it’s good enough for most
purposes, but like Newtonian gravity, it’s not
a good foundation for theoretical discussion
or precise analysis.

                    In the equations

that I’m familiar with the connection from * Qi*to p is represented by a constant Ki .

          So it is, but how realistic do you think that is in the

real world of live organisms?

                    Hasn't that

sufficed for implementing simulations, or have I
missed something?

           As I said: " As an

approximation, it’s good enough for most purposes,".
Does anyone claim that the simulations actually
represent what goes on inside the organism? Even the
concept of a neural current has no equivalent in an
actual brain. It’s an analytical convenience, an
abstraction that simply assumes that the effect of a lot
of neurons firing with their own timings is the same as
though one super-neuron performed all the firings, and
then smeared them across time so that a smooth variation
was used in further functions. For most purposes, that’s
fine, but if you really want to think about it, Bill
just said that if it’s within a few percent (5%, 2%, I
forget) that’s good enough. And it usually is. But it
doesn’t mean that it’s perfect.

Any
imperfection and noise in the biological
implementation is just another disturbance.
Disturbances can enter at any point in the loop.

           Indeed, but when they appear between the

controlled variable p and the complex environmental
variable (the CEV) to which it corresponds, all that means
is that the CEV is not controlled as precisely as the
perception is.

If such
disturbances could not be countered by the
control process in the same way that
environmental disturbances are, and if they were
great enough to make p depart from its
correspondence to the relevant aspect of the
environment, as represented by Qi , they
would be pathologies making it less likely for
that organism to succeed in bringing offspring
to reproductive maturity, so there is obvious
evolutionary pressure for that coupling to be
quite good enough to support good control.

           Yes. That, in essence, is what I said when I said

“it’s good enough for most purposes”. You have read a lot
of my writings. How often have I written in things
addressed to PCT newbies that though what is controlled is
perception, it’s what happens in the environment that
matters?

That seems to
me a pretty strong basis for that coupling being
treated as a constant Ki . rather than as
a variable subject to significant unpredictable
perturbations.

           Much more likely to be some kind  of approximation

to log(Qi) with some kind of ceiling and some kind of
zero-region tolerance zone.

          It doesn't matter, what the function is, if it's

invertible. noise free, and consistent (which adapting
systems are not).

          Look, all I'm trying to do is to emphasize that PCT is

about The Control of Perception, something that seems in
danger of being forgotten even on CSGnet.

              Martin

/Bruce

                  On Tue, Nov 24, 2015 at

2:44 PM, Martin Taylor mmt-csg@mmtaylor.net
wrote:

                      [Martin

Taylor 2015.11.24.14.02]

                            [From Rick Marken

(2015.11.22.0950)]

                       No. p1 is the controlled quantity, the

perception that tracks the reference value -d2
closely if all the g values are substantially
greater than unity. In a real control loop, of
course, the “g” multipliers would represent
the long-term stable values of the leaky
integrators, just as in the usual analysis of
the simple control loop. The actual loop could
not use simple multipliers. When there are
loop delays, simple multipliers inevitably
lead to oscillation and no control. My
analysis was of the stable equilibrium values,
and for that, the leaky integrators are well
represented by simple multipliers.

                      And what's an "ECV"?
                       Yes, I am interested. It's good to

have the spreadsheet example. What gain and
leak rates did you use for the four “g”
functions, to get the “g” multipliers? The
effect of d4 and d3 is diminished by the
multiplier ratio each step back round the
loop. My analysis assumed, g>>1, as we
do when we do an equilibrium analysis of the
ordinary control loop and assume the loop gain
G>>1.

                      I imagine that in your spreadsheet you have a

scalar variable and a simple multiplier, as I
showed in the example. One can’t actually run
the TCV on a single scalar, because there is
no function to be found. But it would be
interesting to run a spreadsheet example in
which each of the paths was a vector of, say,
three scalars, and each perceptual function
was different, and then run the TCV to see
what you find.

                      "PCT" doesn't imply it. It's simply a fact of

life (and of engineering) that ONLY if the
connection from Qi to the perceptual variable
is invertible, perfect, and noise-free will
there be no difference between the ECV
(whatever that may be) and the perception. The
perception is controlled, and as a
consequence, the environmental variable
appears to be. As an approximation, it’s good
enough for most purposes, but like Newtonian
gravity, it’s not a good foundation for
theoretical discussion or precise analysis.

                       That's at least equally wrong. I think

it might be worth your while to look a little
more closely into the actual conditions for
using the TCV, and the potential and
limitations on what you can determine by using
it. You often seem to suggest (planning in
imagination) that you might use the TCV in
real-life situations. Sometimes the conditions
are suitable, but much more often, they
aren’t. I haven’t done it, so I am also
planning in imagination, but one ought to be
able to run the TCV on your demo of a
three-level control system to find what is
being controlled at the top level. You have
all the outputs and disturbances necessary, so
it should work. But what about in a real-life
situation in which the circumstances never
recur. In the hammering example, this might be
the only time in the hammerer’s life that he
is so angry with his wife that he has to hit
something, and doesn’t want to hit his wife.
How can the TCV be used in that situation?

                      And how do you use the TCV when control is

poor? If you get a poor compensation of the
disturbance by the output, how do you know
whether you haven’t found the controlled
variable or you have found it and the control
system doesn’t work very well?

                      I think your statement is simply equivalent to

saying “PCT research is impossible” which is
something I don’t believe, though I do believe
that the control of perception accounts for
what we see people and other organisms do, and
that we should carefully study by all
available means just how this works.

                          Martin
                                    Bruce Nevin

(2015.11.21.20:44 ET) to Martin
Taylor

                                      BN: Thanks for this nice

demonstration of the
difficulty with this
distinction [between ECV and
p).

                                  RM: I think Martin aimed to

demonstrate that controlling an
ECV is not equivalent to
controlling the perception that
corresponds to that ECV. But, in
fact, his demonstration doesn’t
demonstrate that at all. What it
demonstrates is that when you put
a bunch of disturbances into the
feedback connection between output
and input you control neither the
ECV nor p. To be precise, the
disturbance variables, d3 and d4.
enter the loop after the output,
p2, and before the input, (p4+d1).
See Martin’s diagram below:

                                  RM: When these disturbances are

present the effect of output (p2)
on input (p4+d1) is constantly
changing. However, if you remove
these disturbances from the
feedback function control is
restored.

                                  RM: In this control loop p4+d1

is the controlled quantity (q.i,or
ECV) and g1*(p4+d1) is the
controlled perception, p.

                                  The only difference between

q.i (the ECV) and p is the scaling
factor, g1. But variations in p
(p1) are perfectly correlated with
variations in the ECV (p4+d1); the
only difference between p and ECV
is that the former is measured in
neural firing rate units and ECV
is measures in physical units; g1
is just a scaling factor that
converts physical units into
neural firing rate units .

                                  RM: I've implemented Martin's

model in a spreadsheet, in case
anyone is interested. It allows
you to see how well the
perception, p1, and corresponding
ECV (p4+d1) are controlled when
the disturbances to the feedback
function (d3 and d4) are in or out
of the loop. When these
disturbances are in, control of
both p1 and ECV is poor but the
correlation between variations in
p1 and the ECV is 1.0; when these
disturbances are out control of
both p1 and ECV are excellent; and
the correlation between p1 and ECV
is again 1.0.

                                  RM:P I don't know how people

got the idea that PCT implies that
there is a difference between
control of a perception and
control of the corresponding ECV.

                                  But it's an idea that is not

only wrong but one that, if
believed, make PCT research
impossible.

[Bruce Nevin (2015.11.28.22:53 ET)]

Martin Taylor (2015.11.26.14.54) –

BN: This was in response to a person who denies that Qi is controlled at all.

MMT: I have read Boris as simply pointing out that the output affects Qi in order that perception is controlled. I say the same.

BN: if that were the case, there would be no way for an observer to notice the fact of control. No stabilization of the environment against disturbances would be perceptible to anyone except the organism that was doing the controlling.

MMT: Why not? I never perceived you as being of the all-or-none Black or White persuasion, but here you are saying that if (as must be the case) an observer has a different set of inputs to the senses than the person doing the controlling, no matter how similar their inputs and perceptual functions may be, what the observer sees must be totally unrelated to what the controller sees. Sure, if the controller is controlling the placement of a glass on a table, and the observer is looking at the degree to which a door is open, the observer will say there’s no control. But that’s not what we are talking about, is it? The observer sees the glass on the table, and if he wants to know whether the controller cared where it was placed, the observer can become an experimenter and move it. The fact that they see it from different angles may matter, but probably doesn’t.

I am not at all saying that since inputs to the observer’s senses are different from the inputs to the senses of the subject, “what the observer sees must be totally unrelated to what the controller sees”. Although as I attempt in vain to relate that to what I said, it does seem that you may be exemplifying what you said.

I am saying that in your glass scenario or in the TCV the perception that each of the participants controls is related to the perception that the other controls by way of that aspect of their common environment which they are controlling. To talk about that relationship of the observer’s perception to the subject’s perception, you prefer to say that the perception that each of them controls is related to the perception that the other controls by way of that aspect of their common environment which they are influencing. I assume you have a purpose for that choice of words, but you have not stated it. I have a purpose in saying that Qi is controlled. I will explain that here.

In the TCV, the tester controls variables until a (gentle) conflict with the subject is confirmed. That conflict affirms that they are both controlling the same aspect of the environment. Or in your words, they are both influencing the same aspect of the environment. That controlled or influenced aspect of the environment is quantified as Qi. The controlled perception p is a transform of Qi from physical units measured in the environment to (per the PCT model) a rate of firing in a nerve or nerve bundle. The transformation by the input function is quantified as a constant Ki. You have objected that imperfections in the sensory apparatus make Ki a noisy variable. My rejoinder was that if that has any significant effect at all, and is not zeroed out as just another disturbance in the loop, the effect is that Qi is less well controlled than p is, but Qi is nonetheless still controlled.

As far as I can see, to say that the tester and the subject are merely influencing Qi (or that aspect of the environment which is quantified as Qi) as means of controlling their respective perceptions is sophistry, a terminological distinction without a difference, serving no purpose and confusing the issue. Or if you do have a purpose in making that distinction, please do say what it is. But even my astigmatism does not interfere with my ability to put that glass back where I want it, so perfect me no run of the mill sensory imperfections, please. Or, more politely, let us say that I remain unconvinced.

Perceptual control has environmental consequences that are perceived (and can be controlled) by others. Your position is that when a perception is controlled the environmental consequences are not controlled. In my view, environmental consequences that are not controlled are called side effects.

In your view, the environment is merely influenced by control activities in order that the perception may be controlled. The perceived influence is controlled, but the influence that is perceived is not controlled. The intended environmental consequences of that influence do not constitute control of the affected aspect of the environment. I say that there is evidence that the affected aspect of the environment is controlled, and that the environmental consequences of control, as perceived by others, measured by instruments, etc., are controlled. The effect is intentional. Indeed, the nature of that effect is precisely, control. One kind of evidence is that it is perceived by another as control. “What are you doing to that glass?” Another is that the tester’s perception (from the imagined point of view of the subject) is sufficient basis from which successfully to deduce the subject’s internally maintained reference value for p. Another is that conflict often has environmental consequences (“Now see what you’ve done! You’ve spilled the water!”) which may disturb collectively controlled variables. Collective control is yet another kind of evidence: stabilization of what? An environmental feedback path.

Perhaps you are brought to your position in part by the testimony of the physical sciences that the objects, relations, and events that we perceive devolve to shifting arrangements of subatomic particles and energy.

Did old Sam Johnson bruise his foot in vain?

The assumption that perceptions are veridical, and that control of a perception indicates control of that which is perceived, is the converse of a sacrament. A sacrament, as you may recall, is said to be an outward and visible sign of an inward and spiritual reality. A controlled perception is an inward and perceptible sign of an outward reality which, aside from perceptions, is unknowable. The latter is as much an article of faith as the former. Yet it certainly seems not so, because of our existential reliance on perceptions. Indeed, faith of the sacramental sort is characterized by belief without evidence; and your claim seems to amount to saying that the only evidence we have, our perceptions, is no evidence at all. As Alice would say, curiouser and curiouser.

I have two questions:

  1. How do you avoid solipsism?
  2. What explanatory principles do you invoke to account for how the Test for the controlled variable discloses the subject’s CV on the basis of your perceptions? (Let the TCV serve as first proxy for the other kinds of evidence enumerated above.)

I postulate only one explanatory principle: that an aspect of the environment is controlled when a perception is controlled.

We derive our conviction as to the veridicality of perception from the mutual consistency of many perceptions, including our incessant informal testing of what variables those around us are controlling. Collaboration, collective control, conflict and its resolution, all hinge upon a public actuality that is commonly affected by the separate and private control of perceptions by the participants, thereby confirming again and again that control of perceptions is by means of control of the perceived environment. Is that confidence ill founded?

The physical universe, whatever it is, is resistant to our control activities. When you shift the alignment of a dime in the coin game or a glass on the table it stays put when you take your hand away. Presumably, that resistance emerges from what seems to be an infinite plasticity of subatomic phenomena somehow–collective control by infinitesimal points of energy/consciousness?–but however it comes about, a consequence is that control through the environment is very different from control in imagination. When we control our perceptions, we do so by overcoming the inertial character of material things, by making changes in the environment which are perceived as effects of our control of perceptions. And a great many of those effects endure in our absence until our return. The furniture is where we left it. Ah, that’s where I left my glasses, now I remember.

I know he’s a crotchety old fellow, but let Mr. Ockham have a word. On offer is a single explanatory principle to account for all this: an aspect of reality is controlled when the perception of it is controlled. Please show us how any other account avoids multiplying explanatory principles.

···

On Thu, Nov 26, 2015 at 3:21 PM, Martin Taylor mmt-csg@mmtaylor.net wrote:

[Martin Taylor 2015.11.26.14.54]

[Bruce Nevin (2015.11.25.20:04 ET)]

Martin Taylor 2015.11.24.23.37 –

        when [disturbances] appear between the controlled

variable p and the complex environmental variable (the CEV)
to which it corresponds, all that means is that the CEV is
not controlled as precisely as the perception is

Yes, but it is controlled, however imperfectly that may be.

I deny that.



The appearance of control of teh Complex Environmental Variable is,

if I understood you correctly when you used the term in another
context, a spandrel. The appearance that is is controlled is a
consequence of something else truly being controlled. It used to be
quite obvious that phlogiston flowed in and out of objects, and even
now we observe heat flowing in and out of objects, but there’s
“really” no flow of anything. All there is is a bunch of molecules
moving around and beating the hell out of each other. It’s the same
kind of thing. When we talk casually, I have no objection to saying
that the CEV is controlled. I do it myself quite often. But when we
want to explain the theory to anyone in or out of CSGnet, one of the
very first things we have to explain is that the CEV is not
controlled, however much it looks as though it is. The perception of
it is controlled, and that is the reason it looks as though it is
controlled (as also is everyone else’s perception of anything
correlated with the CEV, though that fact is never mentioned in this
discussion; why should the argument not be that the controller is
controlling what some undefined other person is perceiving? The
logic is the same.).

      You're denying my assertion that a disturbance at that

point in the loop can be resisted.

No I most definitely am not!!! If it were not resisted, how could

the corresponding perception be controlled?

      In the case of a pathology, it is certainly the case that

control is impaired, as I said. Example: before the invention
of corrective lenses, my astigmatism would require me to rely
on others to make out details of a scene and report them, as
would my relatively slight myopia.

        Look, all I'm trying to do is to emphasize that PCT is

about The Control of Perception, something that seems in
danger of being forgotten even on CSGnet.

      Yes, but consider the context. This was in response to a

person who denies that Qi is controlled at all.

I guess we bring different prior assumptions to our reading of what

always must be ambiguous, and that’s foubly true of someone whose
first language is no variety of English. I often disagree with what
Boris says, but on this I have read Boris as simply pointing out
that the output affects Qi in order that perception is controlled.
I say the same, but I am not usually told I don’t understand PCT –
at least not in the 20 or so years since Bill challenged Rick when
he made that claim, saying something along the lines of “Who do you
think you are saying doesn’t understand PCT?”

      I agree with Rick: if that were the case, there would be

no way for an observer to notice the fact of control. No
stabilization of the environment against disturbances would be
perceptible to anyone except the organism that was doing the
controlling.

Why not? I never perceived you as being of the all-or-none Black or

White persuasion, but here you are saying that if (as must be the
case) an observer has a different set of inputs to the senses than
the person doing the controlling, no matter how similar their inputs
and perceptual functions may be, what the observer sees must be
totally unrelated to what the controller sees. Sure, if the
controller is controlling the placement of a glass on a table, and
the observer is looking at the degree to which a door is open, the
observer will say there’s no control. But that’s not what we are
talking about, is it? The observer sees the glass on the table, and
if he wants to know whether the controller cared where it was
placed, the observer can become an experimenter and move it. The
fact that they see it from different angles may matter, but probably
doesn’t.

Martin

/Bruce

      On Tue, Nov 24, 2015 at 11:59 PM,

Martin Taylor mmt-csg@mmtaylor.net
wrote:

          [Martin Taylor

2015.11.24.23.37]

On 2015/11/24 10:37 PM, Bruce Nevin wrote:

[Bruce Nevin (2015.11.24. ET)]

                      Martin Taylor

(2015.11.24.14.02) –

                      "PCT" doesn't

imply it. It’s simply a fact of life (and of
engineering) that ONLY if the connection from
Qi to the perceptual variable is invertible,
perfect, and noise-free will there be no
difference between the ECV (whatever that may
be) and the perception. The perception is
controlled, and as a consequence, the
environmental variable appears to be. As an
approximation, it’s good enough for most
purposes, but like Newtonian gravity, it’s not
a good foundation for theoretical discussion
or precise analysis.

                    In the equations

that I’m familiar with the connection from * Qi*to p is represented by a constant Ki .

          So it is, but how realistic do you think that is in the

real world of live organisms?

                    Hasn't that

sufficed for implementing simulations, or have I
missed something?

           As I said: " As an

approximation, it’s good enough for most purposes,".
Does anyone claim that the simulations actually
represent what goes on inside the organism? Even the
concept of a neural current has no equivalent in an
actual brain. It’s an analytical convenience, an
abstraction that simply assumes that the effect of a lot
of neurons firing with their own timings is the same as
though one super-neuron performed all the firings, and
then smeared them across time so that a smooth variation
was used in further functions. For most purposes, that’s
fine, but if you really want to think about it, Bill
just said that if it’s within a few percent (5%, 2%, I
forget) that’s good enough. And it usually is. But it
doesn’t mean that it’s perfect.

Any
imperfection and noise in the biological
implementation is just another disturbance.
Disturbances can enter at any point in the loop.

           Indeed, but when they appear between the

controlled variable p and the complex environmental
variable (the CEV) to which it corresponds, all that means
is that the CEV is not controlled as precisely as the
perception is.

If such
disturbances could not be countered by the
control process in the same way that
environmental disturbances are, and if they were
great enough to make p depart from its
correspondence to the relevant aspect of the
environment, as represented by Qi , they
would be pathologies making it less likely for
that organism to succeed in bringing offspring
to reproductive maturity, so there is obvious
evolutionary pressure for that coupling to be
quite good enough to support good control.

           Yes. That, in essence, is what I said when I said

“it’s good enough for most purposes”. You have read a lot
of my writings. How often have I written in things
addressed to PCT newbies that though what is controlled is
perception, it’s what happens in the environment that
matters?

That seems to
me a pretty strong basis for that coupling being
treated as a constant Ki . rather than as
a variable subject to significant unpredictable
perturbations.

           Much more likely to be some kind  of approximation

to log(Qi) with some kind of ceiling and some kind of
zero-region tolerance zone.

          It doesn't matter, what the function is, if it's

invertible. noise free, and consistent (which adapting
systems are not).

          Look, all I'm trying to do is to emphasize that PCT is

about The Control of Perception, something that seems in
danger of being forgotten even on CSGnet.

              Martin

/Bruce

                  On Tue, Nov 24, 2015 at

2:44 PM, Martin Taylor mmt-csg@mmtaylor.net
wrote:

                      [Martin

Taylor 2015.11.24.14.02]

                            [From Rick Marken

(2015.11.22.0950)]

                       No. p1 is the controlled quantity, the

perception that tracks the reference value -d2
closely if all the g values are substantially
greater than unity. In a real control loop, of
course, the “g” multipliers would represent
the long-term stable values of the leaky
integrators, just as in the usual analysis of
the simple control loop. The actual loop could
not use simple multipliers. When there are
loop delays, simple multipliers inevitably
lead to oscillation and no control. My
analysis was of the stable equilibrium values,
and for that, the leaky integrators are well
represented by simple multipliers.

                      And what's an "ECV"?
                       Yes, I am interested. It's good to

have the spreadsheet example. What gain and
leak rates did you use for the four “g”
functions, to get the “g” multipliers? The
effect of d4 and d3 is diminished by the
multiplier ratio each step back round the
loop. My analysis assumed, g>>1, as we
do when we do an equilibrium analysis of the
ordinary control loop and assume the loop gain
G>>1.

                      I imagine that in your spreadsheet you have a

scalar variable and a simple multiplier, as I
showed in the example. One can’t actually run
the TCV on a single scalar, because there is
no function to be found. But it would be
interesting to run a spreadsheet example in
which each of the paths was a vector of, say,
three scalars, and each perceptual function
was different, and then run the TCV to see
what you find.

                      "PCT" doesn't imply it. It's simply a fact of

life (and of engineering) that ONLY if the
connection from Qi to the perceptual variable
is invertible, perfect, and noise-free will
there be no difference between the ECV
(whatever that may be) and the perception. The
perception is controlled, and as a
consequence, the environmental variable
appears to be. As an approximation, it’s good
enough for most purposes, but like Newtonian
gravity, it’s not a good foundation for
theoretical discussion or precise analysis.

                       That's at least equally wrong. I think

it might be worth your while to look a little
more closely into the actual conditions for
using the TCV, and the potential and
limitations on what you can determine by using
it. You often seem to suggest (planning in
imagination) that you might use the TCV in
real-life situations. Sometimes the conditions
are suitable, but much more often, they
aren’t. I haven’t done it, so I am also
planning in imagination, but one ought to be
able to run the TCV on your demo of a
three-level control system to find what is
being controlled at the top level. You have
all the outputs and disturbances necessary, so
it should work. But what about in a real-life
situation in which the circumstances never
recur. In the hammering example, this might be
the only time in the hammerer’s life that he
is so angry with his wife that he has to hit
something, and doesn’t want to hit his wife.
How can the TCV be used in that situation?

                      And how do you use the TCV when control is

poor? If you get a poor compensation of the
disturbance by the output, how do you know
whether you haven’t found the controlled
variable or you have found it and the control
system doesn’t work very well?

                      I think your statement is simply equivalent to

saying “PCT research is impossible” which is
something I don’t believe, though I do believe
that the control of perception accounts for
what we see people and other organisms do, and
that we should carefully study by all
available means just how this works.

                          Martin
                                    Bruce Nevin

(2015.11.21.20:44 ET) to Martin
Taylor

                                      BN: Thanks for this nice

demonstration of the
difficulty with this
distinction [between ECV and
p).

                                  RM: I think Martin aimed to

demonstrate that controlling an
ECV is not equivalent to
controlling the perception that
corresponds to that ECV. But, in
fact, his demonstration doesn’t
demonstrate that at all. What it
demonstrates is that when you put
a bunch of disturbances into the
feedback connection between output
and input you control neither the
ECV nor p. To be precise, the
disturbance variables, d3 and d4.
enter the loop after the output,
p2, and before the input, (p4+d1).
See Martin’s diagram below:

                                  RM: When these disturbances are

present the effect of output (p2)
on input (p4+d1) is constantly
changing. However, if you remove
these disturbances from the
feedback function control is
restored.

                                  RM: In this control loop p4+d1

is the controlled quantity (q.i,or
ECV) and g1*(p4+d1) is the
controlled perception, p.

                                  The only difference between

q.i (the ECV) and p is the scaling
factor, g1. But variations in p
(p1) are perfectly correlated with
variations in the ECV (p4+d1); the
only difference between p and ECV
is that the former is measured in
neural firing rate units and ECV
is measures in physical units; g1
is just a scaling factor that
converts physical units into
neural firing rate units .

                                  RM: I've implemented Martin's

model in a spreadsheet, in case
anyone is interested. It allows
you to see how well the
perception, p1, and corresponding
ECV (p4+d1) are controlled when
the disturbances to the feedback
function (d3 and d4) are in or out
of the loop. When these
disturbances are in, control of
both p1 and ECV is poor but the
correlation between variations in
p1 and the ECV is 1.0; when these
disturbances are out control of
both p1 and ECV are excellent; and
the correlation between p1 and ECV
is again 1.0.

                                  RM:P I don't know how people

got the idea that PCT implies that
there is a difference between
control of a perception and
control of the corresponding ECV.

                                  But it's an idea that is not

only wrong but one that, if
believed, make PCT research
impossible.

[From Rick Marken (2015.11.28.1800)]

···

Martin Taylor (2015.11.26.14.54)–

MT: The appearance of control of the Complex Environmental Variable is,

if I understood you correctly when you used the term in another
context, a spandrel. The appearance that is is controlled is a
consequence of something else truly being controlled.

RM: According to Google, a spandrel, in biology, is “a phenotypic characteristic that is a byproduct of the evolution of some other characteristic”. Given this definition, the fact that q.i (the CEV) is controlled can, indeed, be considered a spandrel – a byproduct of control of the perceptual correlate of q.i. But that is quite different than control of q.i being an “appearance”, implying that it is not really happening.

MT: When we talk casually, I have no objection to saying

that the CEV is controlled. I do it myself quite often. But when we
want to explain the theory to anyone in or out of CSGnet, one of the
very first things we have to explain is that the CEV is not
controlled, however much it looks as though it is.

RM: This implies that the person to whom you are doing the explaining has seen that a variable, the CEV (q.i), is under control. So that person has observed the phenomenon of control that control theory explains; they have observed the fact that q.i is kept in a reference state, protected from disturbance.

RM: I would say that the “first thing” to do now is the “first thing” Bill did in B:CP after explaining what the phenomenon of control is (Ch. 4): show that this phenomenon is accounted for by a model that says behavior (control) is the control of perception, the model of course being PCT. And be sure to explain that the perception that is being controlled does not necessarily correspond to the person’s perception of q.i – the variable that the person sees being controlled. In order to understand the observed behavior-- the fact that q.i is being controlled – you have to discover what perception is actually being controlled.

RM: For example, the person to whom we are doing the explaining may have observed the phenomenon of control in the rubber band demo by seeing that S’s behavior involves keeping the knot about 2 cm to the left of the dot despite E’s disturbances to knot position (I’m using the terminology used on pp 243-245 of B:CP 2nd Ed). So our person has observed that a CEV (q.i), the distance from knot to dot, is being controlled – maintained in a reference state of 2 cm to the left of the dot.

RM: Rather than telling the person that this is not really happening, I would tell the person that the knot is being kept in a reference state – controlled – because S is controlling a perception of the distance from knot to dot relative to a reference signal in their brain that specifies the reference state of this perception. And S is perceiving this distance from S’s perspective, to the left of the dot, so while it may look to you like the knot is being kept 2 cm to the left of the dot, from S’s position, due to parallax, it looks like the knot is being kept over the dot; S’s perception of the controlled state of affairs is not the same as yours. This fact would be readily revealed using the Test for the Controlled Variable.

RM: So you could explain to the person that the goal of research based on PCT is to discover the controlled variable (perception) that corresponds to an observer’s perception of the controlled quantity, q.i; what you call the CEV.

MT: I often disagree with what

Boris says, but on this I have read Boris as simply pointing out
that the output affects Qi in order that perception is controlled.
I say the same, but I am not usually told I don’t understand PCT

RM: What you don’t seem to understand is that the output that affects q.i is part of a control loop, as is q.i and p. So the output affects q.i, but in precisely the right way so that q.i (and p) are maintained in a reference state, protected from disturbance. So in a control loop both q.i and p are controlled. Today, Bruce Abbott (2015.11.28.0940 EST) posted an excellent demonstration of this fact.And, as Bruce Nevin has pointed out, if this weren’t true we would have no way of even knowing that control is going on. We can’t see the perceptions other people are controlling. In PCT we infer these perceptions based on our own perception of the fact that certain variables, such as the distance between the knot and dot in the rubber band demo, are being controlled.

MT: --

at least not in the 20 or so years since Bill challenged Rick when
he made that claim, saying something along the lines of “Who do you
think you are saying doesn’t understand PCT?”

RM: That’s not even worth a reply.

RM: But thanks for the conversation. It’s helping me flesh out my ideas for a paper on PCT research called (what else) “Testing for Controlled Variables”.

RM: Oh, and one quick terminology note. I think we should stick to referring to what you call the CEV as the controlled quantity and to the perceptual correlate of that variable as the controlled variable. That is, q.i is the controlled quantity and p is the controlled variable. This is the terminology Bill uses in B:CP, though I don’t know how consistent Bill was in it’s use throughout the book. But it was certainly the terminology Bill used in the chapter on Experimental Methods and it’s the terminology I will use in my paper on the TCV.

Best regards

Rick


Richard S. Marken

www.mindreadings.com
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

Bruce,

I will ask you three questions. The first is about your use of

language, the second is contingent on your answer to the first, and
the third on your answer to the second.

1. Under which of the following conditions, if any, do you consider

a variable to be controlled:

(a) when disturbed from an apparent resting position it returns to

or near that position.
(b) when disturbed from an apparent resting position, it returns to
or near that position as the result of an observable effect from a
source other than the observed disturbing influence.
(c) when disturbed from a resting position apparently determined by
some other variable, it returns to or near that resting position as
the result of an observable effect from a source other than the
observed disturbing influence.

2. If your answer includes (a) or (b), do you think it possible to

have two independent controlled variables in a standard PCT control
loop?

3. If your answer to (2) is no, which variable in the standard PCT

control loop is controlled?

···

On 2015/11/28 10:54 PM, Bruce Nevin
wrote:

[Bruce Nevin (2015.11.28.22:53 ET)]

      Martin Taylor (2015.11.26.14.54)

            BN: This was in response

to a person who denies that Qi is controlled at all.

            MMT: I have read Boris

as simply pointing out that the output affects Qi in
order that
perception is controlled. I say the same.

              BN: if that were the

case, there would be no way for an observer to notice
the fact of control. No stabilization of the
environment against disturbances would be perceptible
to anyone except the organism that was doing the
controlling.

              MMT: Why not? I never perceived you as being of the

all-or-none Black or White persuasion, but here you
are saying that if (as must be the case) an observer
has a different set of inputs to the senses than the
person doing the controlling, no matter how similar
their inputs and perceptual functions may be, what the
observer sees must be totally unrelated to what the
controller sees. Sure, if the controller is
controlling the placement of a glass on a table, and
the observer is looking at the degree to which a door
is open, the observer will say there’s no control. But
that’s not what we are talking about, is it? The
observer sees the glass on the table, and if he wants
to know whether the controller cared where it was
placed, the observer can become an experimenter and
move it. The fact that they see it from different
angles may matter, but probably doesn’t.

        I am not at all saying that since inputs to the

observer’s senses are different from the inputs to the
senses of the subject, “what the observer sees must be
totally unrelated to what the controller sees”. Although as
I attempt in vain to relate that to what I said, it does
seem that you may be exemplifying what you said.

        I am saying that in your glass scenario or in the TCV the

perception that each of the participants controls is related
to the perception that the other controls by way of
that aspect of their common environment which they are controlling .
To talk about that relationship of the observer’s perception
to the subject’s perception, you prefer to say that the
perception that each of them controls is related to the
perception that the other controls by way of that aspect of
their common environment which they are influencing .
I assume you have a purpose for that choice of words, but
you have not stated it. I have a purpose in saying that * Qi*is controlled. I will explain that here.

        In the TCV, the tester controls variables until a

(gentle) conflict with the subject is confirmed. That
conflict affirms that they are both controlling the same
aspect of the environment. Or in your words, they are both
influencing the same aspect of the environment. That
controlled or influenced aspect of the environment is
quantified as Qi. The controlled perception p is a
transform of Qi from physical units measured in the
environment to (per the PCT model) a rate of firing in a
nerve or nerve bundle. The transformation by the input
function is quantified as a constant Ki . You have
objected that imperfections in the sensory apparatus make Ki a
noisy variable. My rejoinder was that if that has any
significant effect at all, and is not zeroed out as just
another disturbance in the loop, the effect is that Qi is
less well controlled than p is, but Qi is
nonetheless still controlled.

        As far as I can see, to say that the tester and the

subject are merely influencing Qi (or that aspect of
the environment which is quantified as Qi ) as means
of controlling their respective perceptions is sophistry, a
terminological distinction without a difference, serving no
purpose and confusing the issue. Or if you do have a purpose
in making that distinction, please do say what it is. But
even my astigmatism does not interfere with my ability to
put that glass back where I want it, so perfect me no run of
the mill sensory imperfections, please. Or, more politely,
let us say that I remain unconvinced.

        Perceptual control has environmental consequences that

are perceived (and can be controlled) by others. Your
position is that when a perception is controlled the
environmental consequences are not controlled. In my view,
environmental consequences that are not controlled are
called side effects.

        In your view, the environment is merely influenced by

control activities in order that the perception may be
controlled. The perceived influence is controlled, but the
influence that is perceived is not controlled. The intended
environmental consequences of that influence do not
constitute control of the affected aspect of the
environment. I say that there is evidence that the affected
aspect of the environment is controlled, and that the
environmental consequences of control, as perceived by
others, measured by instruments, etc., are controlled. The
effect is intentional. Indeed, the nature of that effect is
precisely, control. One kind of evidence is that it is
perceived by another as control. “What are you doing to that
glass?” Another is that the tester’s perception (from the
imagined point of view of the subject) is sufficient basis
from which successfully to deduce the subject’s internally
maintained reference value for p . Another is that
conflict often has environmental consequences (“Now see what
you’ve done! You’ve spilled the water!”) which may disturb
collectively controlled variables. Collective control is yet
another kind of evidence: stabilization of what? An * environmental*feedback path.

          Perhaps you are brought to your position in part by the

testimony of the physical sciences that the objects,
relations, and events that we perceive devolve to shifting
arrangements of subatomic particles and energy.

Did old Sam Johnson bruise his foot in vain?

          The assumption that perceptions are veridical, and that

control of a perception indicates control of that which is
perceived, is the converse of a sacrament. A sacrament, as
you may recall, is said to be an outward and visible sign
of an inward and spiritual reality. A controlled
perception is an inward and perceptible sign of an outward
reality which, aside from perceptions, is unknowable. The
latter is as much an article of faith as the former. Yet
it certainly seems not so, because of our existential
reliance on perceptions. Indeed, faith of the sacramental
sort is characterized by belief without evidence; and your
claim seems to amount to saying that the only evidence we
have, our perceptions, is no evidence at all. As Alice
would say, curiouser and curiouser.

I have two questions:

  1. How do you avoid solipsism?
  2.               What explanatory principles do you invoke to account
    

for how the Test for the controlled variable discloses
the subject’s CV on the basis of your perceptions?
(Let the TCV serve as first proxy for the other kinds
of evidence enumerated above.)

          I postulate only one explanatory principle: that an

aspect of the environment is controlled when a perception
is controlled.

          We derive our conviction as to the veridicality of

perception from the mutual consistency of many
perceptions, including our incessant informal testing of
what variables those around us are controlling.
Collaboration, collective control, conflict and its
resolution, all hinge upon a public actuality that is
commonly affected by the separate and private control of
perceptions by the participants, thereby confirming again
and again that control of perceptions is by means of
control of the perceived environment. Is that confidence
ill founded?

          The physical universe, whatever it is, is resistant to

our control activities. When you shift the alignment of a
dime in the coin game or a glass on the table it stays put
when you take your hand away. Presumably, that resistance
emerges from what seems to be an infinite plasticity of
subatomic phenomena somehow–collective control by
infinitesimal points of energy/consciousness?–but however
it comes about, a consequence is that control through the
environment is very different from control in imagination.
When we control our perceptions, we do so by overcoming
the inertial character of material things, by making
changes in the environment which are perceived as effects
of our control of perceptions. And a great many of those
effects endure in our absence until our return. The
furniture is where we left it. Ah, that’s where I left my
glasses, now I remember.

          I know he's a crotchety old fellow, but let Mr. Ockham

have a word. On offer is a single explanatory principle to
account for all this: an aspect of reality is controlled
when the perception of it is controlled. Please show us
how any other account avoids multiplying explanatory
principles.

/Bruce

        On Thu, Nov 26, 2015 at 3:21 PM,

Martin Taylor mmt-csg@mmtaylor.net
wrote:

            [Martin Taylor

2015.11.26.14.54]

[Bruce Nevin (2015.11.25.20:04 ET)]

Martin Taylor 2015.11.24.23.37 –

                      when [disturbances] appear between the

controlled variable p and the complex
environmental variable (the CEV) to which it
corresponds, all that means is that the CEV is
not controlled as precisely as the perception
is

                    Yes, but it is controlled, however

imperfectly that may be.

             I deny that.



            The appearance of control of teh Complex Environmental

Variable is, if I understood you correctly when you used
the term in another context, a spandrel. The appearance
that is is controlled is a consequence of something else
truly being controlled. It used to be quite obvious that
phlogiston flowed in and out of objects, and even now we
observe heat flowing in and out of objects, but there’s
“really” no flow of anything. All there is is a bunch of
molecules moving around and beating the hell out of each
other. It’s the same kind of thing. When we talk
casually, I have no objection to saying that the CEV is
controlled. I do it myself quite often. But when we want
to explain the theory to anyone in or out of CSGnet, one
of the very first things we have to explain is that the
CEV is not controlled, however much it looks as though
it is. The perception of it is controlled, and that is
the reason it looks as though it is controlled (as also
is everyone else’s perception of anything correlated
with the CEV, though that fact is never mentioned in
this discussion; why should the argument not be that the
controller is controlling what some undefined other
person is perceiving? The logic is the same.).

                    You're denying my assertion that a

disturbance at that point in the loop can be
resisted.

             No I most definitely am not!!! If it were not

resisted, how could the corresponding perception be
controlled?

                    In the case of a pathology, it is certainly

the case that control is impaired, as I said.
Example: before the invention of corrective
lenses, my astigmatism would require me to rely
on others to make out details of a scene and
report them, as would my relatively slight
myopia.

                      Look, all I'm trying to do is to emphasize

that PCT is about The Control of Perception,
something that seems in danger of being
forgotten even on CSGnet.

                    Yes, but consider the context. This was in

response to a person who denies that Qi is
controlled at all.

             I guess we bring different prior assumptions to

our reading of what always must be ambiguous, and that’s
foubly true of someone whose first language is no
variety of English. I often disagree with what Boris
says, but on this I have read Boris as simply pointing
out that the output affects Qi in order that
perception is controlled. I say the same, but I am not
usually told I don’t understand PCT – at least not in
the 20 or so years since Bill challenged Rick when he
made that claim, saying something along the lines of
“Who do you think you are saying doesn’t understand
PCT?”

                    I agree with Rick: if that were the case,

there would be no way for an observer to notice
the fact of control. No stabilization of the
environment against disturbances would be
perceptible to anyone except the organism that
was doing the controlling.

Why not? I never perceived you as being of the
all-or-none Black or White persuasion, but here you are
saying that if (as must be the case) an observer has a
different set of inputs to the senses than the person
doing the controlling, no matter how similar their
inputs and perceptual functions may be, what the
observer sees must be totally unrelated to what the
controller sees. Sure, if the controller is controlling
the placement of a glass on a table, and the observer is
looking at the degree to which a door is open, the
observer will say there’s no control. But that’s not
what we are talking about, is it? The observer sees the
glass on the table, and if he wants to know whether the
controller cared where it was placed, the observer can
become an experimenter and move it. The fact that they
see it from different angles may matter, but probably
doesn’t.

                Martin

/Bruce

                      On Tue, Nov 24, 2015 at

11:59 PM, Martin Taylor <>
wrote:

[Martin Taylor 2015.11.24.23.37]

                            On 2015/11/24 10:37 PM, Bruce Nevin

wrote:

                              [Bruce Nevin (2015.11.24.

ET)]

                                      Martin

Taylor (2015.11.24.14.02) –

                                      "PCT"

doesn’t imply it. It’s simply
a fact of life (and of
engineering) that ONLY if the
connection from Qi to the
perceptual variable is
invertible, perfect, and
noise-free will there be no
difference between the ECV
(whatever that may be) and the
perception. The perception is
controlled, and as a
consequence, the environmental
variable appears to be. As an
approximation, it’s good
enough for most purposes, but
like Newtonian gravity, it’s
not a good foundation for
theoretical discussion or
precise analysis.

                                    In

the equations that I’m familiar
with the connection from * Qi*to p is
represented by a constant Ki .

                          So it is, but how realistic do you think

that is in the real world of live
organisms?

                                    Hasn't

that sufficed for implementing
simulations, or have I missed
something?

                           As I said: " As an

approximation, it’s good enough for most
purposes,". Does anyone claim that the
simulations actually represent what goes
on inside the organism? Even the concept
of a neural current has no equivalent in
an actual brain. It’s an analytical
convenience, an abstraction that simply
assumes that the effect of a lot of
neurons firing with their own timings is
the same as though one super-neuron
performed all the firings, and then
smeared them across time so that a
smooth variation was used in further
functions. For most purposes, that’s
fine, but if you really want to think
about it, Bill just said that if it’s
within a few percent (5%, 2%, I forget)
that’s good enough. And it usually is.
But it doesn’t mean that it’s perfect.

                                    Any imperfection and noise in

the biological implementation is
just another disturbance.
Disturbances can enter at any
point in the loop.

                           Indeed, but when they appear

between the controlled variable p and the
complex environmental variable (the CEV)
to which it corresponds, all that means is
that the CEV is not controlled as
precisely as the perception is.

                                    If such disturbances could not

be countered by the control
process in the same way that
environmental disturbances are,
and if they were great enough to
make p depart from its
correspondence to the relevant
aspect of the environment, as
represented by Qi , they
would be pathologies making it
less likely for that organism to
succeed in bringing offspring to
reproductive maturity, so there
is obvious evolutionary pressure
for that coupling to be quite
good enough to support good
control.

                           Yes. That, in essence, is what I

said when I said “it’s good enough for
most purposes”. You have read a lot of my
writings. How often have I written in
things addressed to PCT newbies that
though what is controlled is perception,
it’s what happens in the environment that
matters?

                                    That seems to me a pretty strong

basis for that coupling being
treated as a constant Ki .
rather than as a variable
subject to significant
unpredictable perturbations.

                           Much more likely to be some kind 

of approximation to log(Qi) with some kind
of ceiling and some kind of zero-region
tolerance zone.

                          It doesn't matter, what the function is,

if it’s invertible. noise free, and
consistent (which adapting systems are
not).

                          Look, all I'm trying to do is to emphasize

that PCT is about The Control of
Perception, something that seems in
danger of being forgotten even on CSGnet.

                              Martin

/Bruce

                                  On Tue, Nov

24, 2015 at 2:44 PM, Martin Taylor
<>
wrote:

                                            [From Rick

Marken
(2015.11.22.0950)]

                                                    Bruce

Nevin
(2015.11.21.20:44
ET) to Martin
Taylor

                                                      BN: Thanks

for this nice
demonstration
of the
difficulty
with this
distinction
[between ECV
and p).

                                                  RM: I think

Martin aimed to
demonstrate that
controlling an ECV
is not equivalent
to controlling the
perception that
corresponds to
that ECV. But, in
fact, his
demonstration
doesn’t
demonstrate that
at all. What it
demonstrates is
that when you put
a bunch of
disturbances into
the feedback
connection between
output and input
you control
neither the ECV
nor p. To be
precise, the
disturbance
variables, d3 and
d4. enter the loop
after the output,
p2, and before the
input, (p4+d1).
See Martin’s
diagram below:

                                                  RM: When these

disturbances are
present the effect
of output (p2) on
input (p4+d1) is
constantly
changing. However,
if you remove
these disturbances
from the feedback
function control
is restored.

                                                  RM: In this

control loop p4+d1
is the controlled
quantity (q.i,or
ECV) and
g1*(p4+d1) is the
controlled
perception, p.

                                                  The only

difference between
q.i (the ECV) and
p is the scaling
factor, g1. But
variations in p
(p1) are perfectly
correlated with
variations in the
ECV (p4+d1); the
only difference
between p and ECV
is that the
former is measured
in neural firing
rate units and ECV
is measures in
physical units;
g1 is just a
scaling factor
that converts
physical units
into neural firing
rate units .

                                                  RM: I've

implemented
Martin’s model in
a spreadsheet, in
case anyone is
interested. It
allows you to see
how well the
perception, p1,
and corresponding
ECV (p4+d1) are
controlled when
the disturbances
to the feedback
function (d3 and
d4) are in or out
of the loop. When
these disturbances
are in, control of
both p1 and ECV is
poor but the
correlation
between variations
in p1 and the ECV
is 1.0; when these
disturbances are
out control of
both p1 and ECV
are excellent; and
the correlation
between p1 and ECV
is again 1.0.

                                      [Martin Taylor

2015.11.24.14.02]

                                       No. p1 is the

controlled quantity, the
perception that tracks the
reference value -d2 closely if
all the g values are
substantially greater than
unity. In a real control loop,
of course, the “g” multipliers
would represent the long-term
stable values of the leaky
integrators, just as in the
usual analysis of the simple
control loop. The actual loop
could not use simple
multipliers. When there are
loop delays, simple
multipliers inevitably lead to
oscillation and no control. My
analysis was of the stable
equilibrium values, and for
that, the leaky integrators
are well represented by simple
multipliers.

                                      And what's an "ECV"?




                                       Yes, I am interested.

It’s good to have the
spreadsheet example. What gain
and leak rates did you use for
the four “g” functions, to get
the “g” multipliers? The
effect of d4 and d3 is
diminished by the multiplier
ratio each step back round the
loop. My analysis assumed,
g>>1, as we do when we
do an equilibrium analysis of
the ordinary control loop and
assume the loop gain
G>>1.

                                      I imagine that in your

spreadsheet you have a scalar
variable and a simple
multiplier, as I showed in the
example. One can’t actually
run the TCV on a single
scalar, because there is no
function to be found. But it
would be interesting to run a
spreadsheet example in which
each of the paths was a vector
of, say, three scalars, and
each perceptual function was
different, and then run the
TCV to see what you find.

                                                  RM:P I don't

know how people
got the idea that
PCT implies that
there is a
difference between
control of a
perception and
control of the
corresponding ECV.

                                                  But it's an

idea that is not
only wrong but one
that, if believed,
make PCT research
impossible.

                                      "PCT" doesn't imply it. It's

simply a fact of life (and of
engineering) that ONLY if the
connection from Qi to the
perceptual variable is
invertible, perfect, and
noise-free will there be no
difference between the ECV
(whatever that may be) and the
perception. The perception is
controlled, and as a
consequence, the environmental
variable appears to be. As an
approximation, it’s good
enough for most purposes, but
like Newtonian gravity, it’s
not a good foundation for
theoretical discussion or
precise analysis.

                                       That's at least

equally wrong. I think it
might be worth your while to
look a little more closely
into the actual conditions for
using the TCV, and the
potential and limitations on
what you can determine by
using it. You often seem to
suggest (planning in
imagination) that you might
use the TCV in real-life
situations. Sometimes the
conditions are suitable, but
much more often, they aren’t.
I haven’t done it, so I am
also planning in imagination,
but one ought to be able to
run the TCV on your demo of a
three-level control system to
find what is being controlled
at the top level. You have all
the outputs and disturbances
necessary, so it should work.
But what about in a real-life
situation in which the
circumstances never recur. In
the hammering example, this
might be the only time in the
hammerer’s life that he is so
angry with his wife that he
has to hit something, and
doesn’t want to hit his wife.
How can the TCV be used in
that situation?

                                      And how do you use the TCV

when control is poor? If you
get a poor compensation of the
disturbance by the output, how
do you know whether you
haven’t found the controlled
variable or you have found it
and the control system doesn’t
work very well?

                                      I think your statement is

simply equivalent to saying
“PCT research is impossible”
which is something I don’t
believe, though I do believe
that the control of perception
accounts for what we see
people and other organisms do,
and that we should carefully
study by all available means
just how this works.

                                          Martin

mmt-csg@mmtaylor.netmmt-csg@mmtaylor.net

[Martin Taylor 2015.11.29.12.52]

[Bruce Nevin (2015.11.28.22:53 ET)]

      Martin Taylor (2015.11.26.14.54)

            BN: This was in response

to a person who denies that Qi is controlled at all.

            MMT: I have read Boris

as simply pointing out that the output affects Qi in
order that
perception is controlled. I say the same.

              BN: if that were the

case, there would be no way for an observer to notice
the fact of control. No stabilization of the
environment against disturbances would be perceptible
to anyone except the organism that was doing the
controlling.

              MMT: Why not? I never perceived you as being of the

all-or-none Black or White persuasion, but here you
are saying that if (as must be the case) an observer
has a different set of inputs to the senses than the
person doing the controlling, no matter how similar
their inputs and perceptual functions may be, what the
observer sees must be totally unrelated to what the
controller sees. Sure, if the controller is
controlling the placement of a glass on a table, and
the observer is looking at the degree to which a door
is open, the observer will say there’s no control. But
that’s not what we are talking about, is it? The
observer sees the glass on the table, and if he wants
to know whether the controller cared where it was
placed, the observer can become an experimenter and
move it. The fact that they see it from different
angles may matter, but probably doesn’t.

…I have a purpose in saying that Qi is controlled.
I will explain that here.

        In the TCV, the tester controls variables until a

(gentle) conflict with the subject is confirmed. That
conflict affirms that they are both controlling the same
aspect of the environment. Or in your words, they are both
influencing the same aspect of the environment. That
controlled or influenced aspect of the environment is
quantified as Qi.

So far, so good.

The controlled perception p is a transform of Qi
from physical units measured in the environment to (per the
PCT model) a rate of firing in a nerve or nerve bundle. The
transformation by the input function is quantified as a
constant Ki.

This "constant multiplier" notion is at he heart of my difficulty in

thinking of there being two separate controlled variables in a
control loop. If it were ever so in a living organism, there would
be no argument, since control of perception would be mathematically
indistinguishable from control of Qi. But for most, if not all
sensor systems, as well as higher-level perceptual variables, there
is a phenomenon called “adaptation”, or to put it another way, they
sense change more than they sense magnitude. Have you ever walked
from a dark room into bright sunlight, or the reverse? What
perceived intensity do you get from different places in the field of
view? I think in the first case, everywhere in the field of view is
bright white, or very nearly, and in the latter case, every point in
the visual field is solid black. But after a while in both cases,
different parts of the visual field are perceived to have different
lightnesses.

There's far from a constant Ki connecting the rate of photon

impingement on the sensors and any kind of perceptual variable. At a
higher level, I have already pointed out the steady increase in
rotary velocity required to maintain a constant perceived rotation
rate.

        You have objected that imperfections in the sensory

apparatus make Ki a noisy variable.

More than that, it's not a simple noisy multiplier, but a function

of time.

        My rejoinder was that if that has any significant effect

at all, and is not zeroed out as just another disturbance in
the loop, the effect is that Qi is less well
controlled than p is, but Qi is nonetheless
still controlled.

I suspect the problem is the usual philosopher's way of inducing a

long and abstract argument. What do you mean by control, as I asked
in my previous message this morning (which has not yet come back to
me)?

If you mean by a variable being "controlled" that when its value is

influenced by a disturbance some other influence appears from some
unspecified place to oppose the disturbing influence, then your
perception of the variable is indeed being controlled, but not by
you. That’s the everyday casual use of the term “control”. As I
said, I use the word that way quite often, but not when I’m trying
to be clear about PCT. When I am trying to be accurate about PCT I
say that Qi is stabilized because p is controlled, not that p and Qi
are both controlled at the same time, because if they were both
controlled, each would have its own reference value that could be
independently varied. It’s the same problem as that of trying
simultaneously and independently to control the position of the car
in its lane and the angle of the steering wheel.

        As far as I can see, to say that the tester and the

subject are merely influencing Qi (or that aspect of
the environment which is quantified as Qi ) as means
of controlling their respective perceptions is sophistry, a
terminological distinction without a difference, serving no
purpose and confusing the issue. Or if you do have a purpose
in making that distinction, please do say what it is.

I find it hard to say more than that if you want to have people

understand PCT, it is better not to confuse them by saying that in
the “classical” control loop there are two independently controlled
variables, and I’ve said that often enough that if it still is
unclear, I don’t know what more to do.

        But even my astigmatism does not interfere with my

ability to put that glass back where I want it, so perfect
me no run of the mill sensory imperfections, please. Or,
more politely, let us say that I remain unconvinced.

        Perceptual control has environmental consequences that

are perceived (and can be controlled) by others.

Perceptual control has environmental consequences that are perceived

by others, who may control those perceptions.

        Your position is that when a perception is controlled

the environmental consequences are not controlled.

No. Theenvironmental consequences are set, stabilized, consistently

produced, or what have you, but technically, NOT controlled.

        In my view, environmental consequences that are not

controlled are called side effects.

In my view, environmental consequences that are not intended are

called side-effects.

        In your view, the environment is merely influenced by

control activities in order that the perception may be
controlled. The perceived influence is controlled, but the
influence that is perceived is not controlled.

If the perceived influence were controlled, that would be control of

output, and that isn’t even in the vernacular. The perception of the
result of influence is controlled, and it would not be if the
environmental correlate were unstabilized.

Let's do a gedankenexperiment. If we set a pole into moderately soft

ground and it is gravitationally vertical, it will stand
indefinitely. If it is tilted off vertical, it will slowly increase
its tilt and fall down. We want it not to fall down so we control
our perception of it until we perceive it to be precisely vertical.
But when we let go of it, it begins to fall. Why? Puzzlement.

Ah, the ground was sloping, so that what we perceived as vertical

was not actually vertical as far as gravity is concerned. We have
controlled our perception to achieve the vertical, but we have not
controlled the environmental consequence to be vertical. [This
example is actually a record of personal experience. A friend was
setting the pole in concrete, and he controlled for perceiving it ti
be vertical. I was trying to control a relationship between the pole
angle and the wall of a house and could see very well that the pole
was not at the same angle as the house wall, but he overwhelmed me
and set the pole off gravitational vertical, and year by year in the
spring thaw, it increased its tilt despite the concrete.]


The assumption that perceptions are veridical, and that
control of a perception indicates control of that which is
perceived, is the converse of a sacrament. A sacrament, as
you may recall, is said to be an outward and visible sign
of an inward and spiritual reality. A controlled
perception is an inward and perceptible sign of an outward
reality which, aside from perceptions, is unknowable. The
latter is as much an article of faith as the former. Yet
it certainly seems not so, because of our existential
reliance on perceptions. Indeed, faith of the sacramental
sort is characterized by belief without evidence; and your
claim seems to amount to saying that the only evidence we
have, our perceptions, is no evidence at all. As Alice
would say, curiouser and curiouser.

Your interpretations are indeed curiouser and curiouser. My "article

of faith" and described as such, is that PCT is correct. No more
than that. All else is consequent on that. No organism can be sure
of what is outside, and what we call perception is never assumed to
be a veridical representation of what is outside. Nevertheless, if
we control appropriate perceptions effectively, we live, and the
reason (part of that one article of faith) is that if we control
them, we do stabilize or alter aspects of the real environment in a
consistent way. That way makes the environmental variable appear to
be controlled.

I have two questions:

  1. How do you avoid solipsism?
  2.               What explanatory principles do you invoke to account
    

for how the Test for the controlled variable discloses
the subject’s CV on the basis of your perceptions?
(Let the TCV serve as first proxy for the other kinds
of evidence enumerated above.)

          I postulate only one explanatory principle: that an

aspect of the environment is controlled when a perception
is controlled.

An I postulate that this is physically impossible unless control of

the aspect of the environment is mathematically equivalent to
control of the perception.

Just as wiht "Behaviour" we have an argument based not on what we

think is happening when perceptions are controlled, but on how to
use words in ways that clarify PCT to a newcomer interested in
learning what it is all about – namely “The Control of Perception”.

          We derive our conviction as to the veridicality of

perception from the mutual consistency of many
perceptions, including our incessant informal testing of
what variables those around us are controlling.
Collaboration, collective control, conflict and its
resolution, all hinge upon a public actuality that is
commonly affected by the separate and private control of
perceptions by the participants, thereby confirming again
and again that control of perceptions is by means of
control of the perceived environment. Is that confidence
ill founded?

No, I would have said the same.

          I know he's a crotchety old fellow, but let Mr. Ockham

have a word. On offer is a single explanatory principle to
account for all this: an aspect of reality is controlled
when the perception of it is controlled. Please show us
how any other account avoids multiplying explanatory
principles.

I would argue that Judge Ockham would decided in my favour in that

only one variable is controlled, not two, in any control loop.

Martin

···

On Thu, Nov 26, 2015 at 3:21 PM,
Martin Taylor mmt-csg@mmtaylor.net
wrote:

            [Martin Taylor

2015.11.26.14.54]

[Bruce Nevin (2015.11.25.20:04 ET)]

Martin Taylor 2015.11.24.23.37 –

                      when [disturbances] appear between the

controlled variable p and the complex
environmental variable (the CEV) to which it
corresponds, all that means is that the CEV is
not controlled as precisely as the perception
is

                    Yes, but it is controlled, however

imperfectly that may be.

             I deny that.



            The appearance of control of teh Complex Environmental

Variable is, if I understood you correctly when you used
the term in another context, a spandrel. The appearance
that is is controlled is a consequence of something else
truly being controlled. It used to be quite obvious that
phlogiston flowed in and out of objects, and even now we
observe heat flowing in and out of objects, but there’s
“really” no flow of anything. All there is is a bunch of
molecules moving around and beating the hell out of each
other. It’s the same kind of thing. When we talk
casually, I have no objection to saying that the CEV is
controlled. I do it myself quite often. But when we want
to explain the theory to anyone in or out of CSGnet, one
of the very first things we have to explain is that the
CEV is not controlled, however much it looks as though
it is. The perception of it is controlled, and that is
the reason it looks as though it is controlled (as also
is everyone else’s perception of anything correlated
with the CEV, though that fact is never mentioned in
this discussion; why should the argument not be that the
controller is controlling what some undefined other
person is perceiving? The logic is the same.).

                    You're denying my assertion that a

disturbance at that point in the loop can be
resisted.

             No I most definitely am not!!! If it were not

resisted, how could the corresponding perception be
controlled?

                    In the case of a pathology, it is certainly

the case that control is impaired, as I said.
Example: before the invention of corrective
lenses, my astigmatism would require me to rely
on others to make out details of a scene and
report them, as would my relatively slight
myopia.

                      Look, all I'm trying to do is to emphasize

that PCT is about The Control of Perception,
something that seems in danger of being
forgotten even on CSGnet.

                    Yes, but consider the context. This was in

response to a person who denies that Qi is
controlled at all.

             I guess we bring different prior assumptions to

our reading of what always must be ambiguous, and that’s
foubly true of someone whose first language is no
variety of English. I often disagree with what Boris
says, but on this I have read Boris as simply pointing
out that the output affects Qi in order that
perception is controlled. I say the same, but I am not
usually told I don’t understand PCT – at least not in
the 20 or so years since Bill challenged Rick when he
made that claim, saying something along the lines of
“Who do you think you are saying doesn’t understand
PCT?”

                    I agree with Rick: if that were the case,

there would be no way for an observer to notice
the fact of control. No stabilization of the
environment against disturbances would be
perceptible to anyone except the organism that
was doing the controlling.

Why not? I never perceived you as being of the
all-or-none Black or White persuasion, but here you are
saying that if (as must be the case) an observer has a
different set of inputs to the senses than the person
doing the controlling, no matter how similar their
inputs and perceptual functions may be, what the
observer sees must be totally unrelated to what the
controller sees. Sure, if the controller is controlling
the placement of a glass on a table, and the observer is
looking at the degree to which a door is open, the
observer will say there’s no control. But that’s not
what we are talking about, is it? The observer sees the
glass on the table, and if he wants to know whether the
controller cared where it was placed, the observer can
become an experimenter and move it. The fact that they
see it from different angles may matter, but probably
doesn’t.

                Martin

/Bruce

                      On Tue, Nov 24, 2015 at

11:59 PM, Martin Taylor mmt-csg@mmtaylor.net
wrote:

[Martin Taylor 2015.11.24.23.37]

                            On 2015/11/24 10:37 PM, Bruce Nevin

wrote:

                              [Bruce Nevin (2015.11.24.

ET)]

                                      Martin

Taylor (2015.11.24.14.02) –

                                      "PCT"

doesn’t imply it. It’s simply
a fact of life (and of
engineering) that ONLY if the
connection from Qi to the
perceptual variable is
invertible, perfect, and
noise-free will there be no
difference between the ECV
(whatever that may be) and the
perception. The perception is
controlled, and as a
consequence, the environmental
variable appears to be. As an
approximation, it’s good
enough for most purposes, but
like Newtonian gravity, it’s
not a good foundation for
theoretical discussion or
precise analysis.

                                    In

the equations that I’m familiar
with the connection from * Qi*to p is
represented by a constant Ki .

                          So it is, but how realistic do you think

that is in the real world of live
organisms?

                                    Hasn't

that sufficed for implementing
simulations, or have I missed
something?

                           As I said: " As an

approximation, it’s good enough for most
purposes,". Does anyone claim that the
simulations actually represent what goes
on inside the organism? Even the concept
of a neural current has no equivalent in
an actual brain. It’s an analytical
convenience, an abstraction that simply
assumes that the effect of a lot of
neurons firing with their own timings is
the same as though one super-neuron
performed all the firings, and then
smeared them across time so that a
smooth variation was used in further
functions. For most purposes, that’s
fine, but if you really want to think
about it, Bill just said that if it’s
within a few percent (5%, 2%, I forget)
that’s good enough. And it usually is.
But it doesn’t mean that it’s perfect.

                                    Any imperfection and noise in

the biological implementation is
just another disturbance.
Disturbances can enter at any
point in the loop.

                           Indeed, but when they appear

between the controlled variable p and the
complex environmental variable (the CEV)
to which it corresponds, all that means is
that the CEV is not controlled as
precisely as the perception is.

                                    If such disturbances could not

be countered by the control
process in the same way that
environmental disturbances are,
and if they were great enough to
make p depart from its
correspondence to the relevant
aspect of the environment, as
represented by Qi , they
would be pathologies making it
less likely for that organism to
succeed in bringing offspring to
reproductive maturity, so there
is obvious evolutionary pressure
for that coupling to be quite
good enough to support good
control.

                           Yes. That, in essence, is what I

said when I said “it’s good enough for
most purposes”. You have read a lot of my
writings. How often have I written in
things addressed to PCT newbies that
though what is controlled is perception,
it’s what happens in the environment that
matters?

                                    That seems to me a pretty strong

basis for that coupling being
treated as a constant Ki .
rather than as a variable
subject to significant
unpredictable perturbations.

                           Much more likely to be some kind 

of approximation to log(Qi) with some kind
of ceiling and some kind of zero-region
tolerance zone.

                          It doesn't matter, what the function is,

if it’s invertible. noise free, and
consistent (which adapting systems are
not).

                          Look, all I'm trying to do is to emphasize

that PCT is about The Control of
Perception, something that seems in
danger of being forgotten even on CSGnet.

                              Martin

/Bruce

                                  On Tue, Nov

24, 2015 at 2:44 PM, Martin Taylor
mmt-csg@mmtaylor.net
wrote:

                                      [Martin Taylor

2015.11.24.14.02]

                                            [From Rick

Marken
(2015.11.22.0950)]

                                       No. p1 is the

controlled quantity, the
perception that tracks the
reference value -d2 closely if
all the g values are
substantially greater than
unity. In a real control loop,
of course, the “g” multipliers
would represent the long-term
stable values of the leaky
integrators, just as in the
usual analysis of the simple
control loop. The actual loop
could not use simple
multipliers. When there are
loop delays, simple
multipliers inevitably lead to
oscillation and no control. My
analysis was of the stable
equilibrium values, and for
that, the leaky integrators
are well represented by simple
multipliers.

                                      And what's an "ECV"?
                                       Yes, I am interested.

It’s good to have the
spreadsheet example. What gain
and leak rates did you use for
the four “g” functions, to get
the “g” multipliers? The
effect of d4 and d3 is
diminished by the multiplier
ratio each step back round the
loop. My analysis assumed,
g>>1, as we do when we
do an equilibrium analysis of
the ordinary control loop and
assume the loop gain
G>>1.

                                      I imagine that in your

spreadsheet you have a scalar
variable and a simple
multiplier, as I showed in the
example. One can’t actually
run the TCV on a single
scalar, because there is no
function to be found. But it
would be interesting to run a
spreadsheet example in which
each of the paths was a vector
of, say, three scalars, and
each perceptual function was
different, and then run the
TCV to see what you find.

                                      "PCT" doesn't imply it. It's

simply a fact of life (and of
engineering) that ONLY if the
connection from Qi to the
perceptual variable is
invertible, perfect, and
noise-free will there be no
difference between the ECV
(whatever that may be) and the
perception. The perception is
controlled, and as a
consequence, the environmental
variable appears to be. As an
approximation, it’s good
enough for most purposes, but
like Newtonian gravity, it’s
not a good foundation for
theoretical discussion or
precise analysis.

                                       That's at least

equally wrong. I think it
might be worth your while to
look a little more closely
into the actual conditions for
using the TCV, and the
potential and limitations on
what you can determine by
using it. You often seem to
suggest (planning in
imagination) that you might
use the TCV in real-life
situations. Sometimes the
conditions are suitable, but
much more often, they aren’t.
I haven’t done it, so I am
also planning in imagination,
but one ought to be able to
run the TCV on your demo of a
three-level control system to
find what is being controlled
at the top level. You have all
the outputs and disturbances
necessary, so it should work.
But what about in a real-life
situation in which the
circumstances never recur. In
the hammering example, this
might be the only time in the
hammerer’s life that he is so
angry with his wife that he
has to hit something, and
doesn’t want to hit his wife.
How can the TCV be used in
that situation?

                                      And how do you use the TCV

when control is poor? If you
get a poor compensation of the
disturbance by the output, how
do you know whether you
haven’t found the controlled
variable or you have found it
and the control system doesn’t
work very well?

                                      I think your statement is

simply equivalent to saying
“PCT research is impossible”
which is something I don’t
believe, though I do believe
that the control of perception
accounts for what we see
people and other organisms do,
and that we should carefully
study by all available means
just how this works.

                                          Martin
                                                    Bruce

Nevin
(2015.11.21.20:44
ET) to Martin
Taylor

                                                      BN: Thanks

for this nice
demonstration
of the
difficulty
with this
distinction
[between ECV
and p).

                                                  RM: I think

Martin aimed to
demonstrate that
controlling an ECV
is not equivalent
to controlling the
perception that
corresponds to
that ECV. But, in
fact, his
demonstration
doesn’t
demonstrate that
at all. What it
demonstrates is
that when you put
a bunch of
disturbances into
the feedback
connection between
output and input
you control
neither the ECV
nor p. To be
precise, the
disturbance
variables, d3 and
d4. enter the loop
after the output,
p2, and before the
input, (p4+d1).
See Martin’s
diagram below:

                                                  RM: When these

disturbances are
present the effect
of output (p2) on
input (p4+d1) is
constantly
changing. However,
if you remove
these disturbances
from the feedback
function control
is restored.

                                                  RM: In this

control loop p4+d1
is the controlled
quantity (q.i,or
ECV) and
g1*(p4+d1) is the
controlled
perception, p.

                                                  The only

difference between
q.i (the ECV) and
p is the scaling
factor, g1. But
variations in p
(p1) are perfectly
correlated with
variations in the
ECV (p4+d1); the
only difference
between p and ECV
is that the
former is measured
in neural firing
rate units and ECV
is measures in
physical units;
g1 is just a
scaling factor
that converts
physical units
into neural firing
rate units .

                                                  RM: I've

implemented
Martin’s model in
a spreadsheet, in
case anyone is
interested. It
allows you to see
how well the
perception, p1,
and corresponding
ECV (p4+d1) are
controlled when
the disturbances
to the feedback
function (d3 and
d4) are in or out
of the loop. When
these disturbances
are in, control of
both p1 and ECV is
poor but the
correlation
between variations
in p1 and the ECV
is 1.0; when these
disturbances are
out control of
both p1 and ECV
are excellent; and
the correlation
between p1 and ECV
is again 1.0.

                                                  RM:P I don't

know how people
got the idea that
PCT implies that
there is a
difference between
control of a
perception and
control of the
corresponding ECV.

                                                  But it's an

idea that is not
only wrong but one
that, if believed,
make PCT research
impossible.

[From Rick Marken (2015.11.30.1250)]

···

On Sun, Nov 29, 2015 at 6:04 AM, Martin Taylor mmt-csg@mmtaylor.net wrote:

MT: I will ask you three questions.

RM: Hope it’s OK for me to take the quiz too;-)

1. Under which of the following conditions, if any, do you consider

a variable to be controlled:

(a) when disturbed from an apparent resting position it returns to

or near that position.

(b) when disturbed from an apparent resting position, it returns to

or near that position as the result of an observable effect from a
source other than the observed disturbing influence.

(c) when disturbed from a resting position apparently determined by

some other variable, it returns to or near that resting position as
the result of an observable effect from a source other than the
observed disturbing influence.

RM: (d) none of the above

2. If your answer includes (a) or (b), do you think it possible to

have two independent controlled variables in a standard PCT control
loop?

RM: N/A

3. If your answer to (2) is no, which variable in the standard PCT

control loop is controlled?

RM: My answer to (2) was neither yes nor no. However, the answer to (3)) is that there are two variables controlled in a standard PCT loop: q.i and p.

MT: My answers: 1-only c, 2- No, 3-the perception. (I answered 2 as a

tautology, given that 1c precludes a yes answer to 2).

RM: Oh my, it looks like I failed the quiz. I guess I’ll have to study harder for the next one;-)

Best

Rick

Martin



  On 2015/11/28 10:54 PM, Bruce Nevin

wrote:

[Bruce Nevin (2015.11.28.22:53 ET)]

      Martin Taylor (2015.11.26.14.54)

            BN: This was in response

to a person who denies that Qi is controlled at all.

            MMT: I have read Boris

as simply pointing out that the output affects Qi in
order that
perception is controlled. I say the same.

              BN: if that were the

case, there would be no way for an observer to notice
the fact of control. No stabilization of the
environment against disturbances would be perceptible
to anyone except the organism that was doing the
controlling.

              MMT: Why not? I never perceived you as being of the

all-or-none Black or White persuasion, but here you
are saying that if (as must be the case) an observer
has a different set of inputs to the senses than the
person doing the controlling, no matter how similar
their inputs and perceptual functions may be, what the
observer sees must be totally unrelated to what the
controller sees. Sure, if the controller is
controlling the placement of a glass on a table, and
the observer is looking at the degree to which a door
is open, the observer will say there’s no control. But
that’s not what we are talking about, is it? The
observer sees the glass on the table, and if he wants
to know whether the controller cared where it was
placed, the observer can become an experimenter and
move it. The fact that they see it from different
angles may matter, but probably doesn’t.

        I am not at all saying that since inputs to the

observer’s senses are different from the inputs to the
senses of the subject, “what the observer sees must be
totally unrelated to what the controller sees”. Although as
I attempt in vain to relate that to what I said, it does
seem that you may be exemplifying what you said.

        I am saying that in your glass scenario or in the TCV the

perception that each of the participants controls is related
to the perception that the other controls by way of
that aspect of their common environment which they are controlling .
To talk about that relationship of the observer’s perception
to the subject’s perception, you prefer to say that the
perception that each of them controls is related to the
perception that the other controls by way of that aspect of
their common environment which they are influencing .
I assume you have a purpose for that choice of words, but
you have not stated it. I have a purpose in saying that * Qi*is controlled. I will explain that here.

        In the TCV, the tester controls variables until a

(gentle) conflict with the subject is confirmed. That
conflict affirms that they are both controlling the same
aspect of the environment. Or in your words, they are both
influencing the same aspect of the environment. That
controlled or influenced aspect of the environment is
quantified as Qi. The controlled perception p is a
transform of Qi from physical units measured in the
environment to (per the PCT model) a rate of firing in a
nerve or nerve bundle. The transformation by the input
function is quantified as a constant Ki . You have
objected that imperfections in the sensory apparatus make Ki a
noisy variable. My rejoinder was that if that has any
significant effect at all, and is not zeroed out as just
another disturbance in the loop, the effect is that Qi is
less well controlled than p is, but Qi is
nonetheless still controlled.

        As far as I can see, to say that the tester and the

subject are merely influencing Qi (or that aspect of
the environment which is quantified as Qi ) as means
of controlling their respective perceptions is sophistry, a
terminological distinction without a difference, serving no
purpose and confusing the issue. Or if you do have a purpose
in making that distinction, please do say what it is. But
even my astigmatism does not interfere with my ability to
put that glass back where I want it, so perfect me no run of
the mill sensory imperfections, please. Or, more politely,
let us say that I remain unconvinced.

        Perceptual control has environmental consequences that

are perceived (and can be controlled) by others. Your
position is that when a perception is controlled the
environmental consequences are not controlled. In my view,
environmental consequences that are not controlled are
called side effects.

        In your view, the environment is merely influenced by

control activities in order that the perception may be
controlled. The perceived influence is controlled, but the
influence that is perceived is not controlled. The intended
environmental consequences of that influence do not
constitute control of the affected aspect of the
environment. I say that there is evidence that the affected
aspect of the environment is controlled, and that the
environmental consequences of control, as perceived by
others, measured by instruments, etc., are controlled. The
effect is intentional. Indeed, the nature of that effect is
precisely, control. One kind of evidence is that it is
perceived by another as control. “What are you doing to that
glass?” Another is that the tester’s perception (from the
imagined point of view of the subject) is sufficient basis
from which successfully to deduce the subject’s internally
maintained reference value for p . Another is that
conflict often has environmental consequences (“Now see what
you’ve done! You’ve spilled the water!”) which may disturb
collectively controlled variables. Collective control is yet
another kind of evidence: stabilization of what? An * environmental*feedback path.

          Perhaps you are brought to your position in part by the

testimony of the physical sciences that the objects,
relations, and events that we perceive devolve to shifting
arrangements of subatomic particles and energy.

Did old Sam Johnson bruise his foot in vain?

          The assumption that perceptions are veridical, and that

control of a perception indicates control of that which is
perceived, is the converse of a sacrament. A sacrament, as
you may recall, is said to be an outward and visible sign
of an inward and spiritual reality. A controlled
perception is an inward and perceptible sign of an outward
reality which, aside from perceptions, is unknowable. The
latter is as much an article of faith as the former. Yet
it certainly seems not so, because of our existential
reliance on perceptions. Indeed, faith of the sacramental
sort is characterized by belief without evidence; and your
claim seems to amount to saying that the only evidence we
have, our perceptions, is no evidence at all. As Alice
would say, curiouser and curiouser.

I have two questions:

  1. How do you avoid solipsism?
  2.               What explanatory principles do you invoke to account
    

for how the Test for the controlled variable discloses
the subject’s CV on the basis of your perceptions?
(Let the TCV serve as first proxy for the other kinds
of evidence enumerated above.)

          I postulate only one explanatory principle: that an

aspect of the environment is controlled when a perception
is controlled.

          We derive our conviction as to the veridicality of

perception from the mutual consistency of many
perceptions, including our incessant informal testing of
what variables those around us are controlling.
Collaboration, collective control, conflict and its
resolution, all hinge upon a public actuality that is
commonly affected by the separate and private control of
perceptions by the participants, thereby confirming again
and again that control of perceptions is by means of
control of the perceived environment. Is that confidence
ill founded?

          The physical universe, whatever it is, is resistant to

our control activities. When you shift the alignment of a
dime in the coin game or a glass on the table it stays put
when you take your hand away. Presumably, that resistance
emerges from what seems to be an infinite plasticity of
subatomic phenomena somehow–collective control by
infinitesimal points of energy/consciousness?–but however
it comes about, a consequence is that control through the
environment is very different from control in imagination.
When we control our perceptions, we do so by overcoming
the inertial character of material things, by making
changes in the environment which are perceived as effects
of our control of perceptions. And a great many of those
effects endure in our absence until our return. The
furniture is where we left it. Ah, that’s where I left my
glasses, now I remember.

          I know he's a crotchety old fellow, but let Mr. Ockham

have a word. On offer is a single explanatory principle to
account for all this: an aspect of reality is controlled
when the perception of it is controlled. Please show us
how any other account avoids multiplying explanatory
principles.

/Bruce


Richard S. Marken

www.mindreadings.com
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

        On Thu, Nov 26, 2015 at 3:21 PM,

Martin Taylor mmt-csg@mmtaylor.net
wrote:

            [Martin Taylor

2015.11.26.14.54]

[Bruce Nevin (2015.11.25.20:04 ET)]

Martin Taylor 2015.11.24.23.37 –

                      when [disturbances] appear between the

controlled variable p and the complex
environmental variable (the CEV) to which it
corresponds, all that means is that the CEV is
not controlled as precisely as the perception
is

                    Yes, but it is controlled, however

imperfectly that may be.

             I deny that.



            The appearance of control of teh Complex Environmental

Variable is, if I understood you correctly when you used
the term in another context, a spandrel. The appearance
that is is controlled is a consequence of something else
truly being controlled. It used to be quite obvious that
phlogiston flowed in and out of objects, and even now we
observe heat flowing in and out of objects, but there’s
“really” no flow of anything. All there is is a bunch of
molecules moving around and beating the hell out of each
other. It’s the same kind of thing. When we talk
casually, I have no objection to saying that the CEV is
controlled. I do it myself quite often. But when we want
to explain the theory to anyone in or out of CSGnet, one
of the very first things we have to explain is that the
CEV is not controlled, however much it looks as though
it is. The perception of it is controlled, and that is
the reason it looks as though it is controlled (as also
is everyone else’s perception of anything correlated
with the CEV, though that fact is never mentioned in
this discussion; why should the argument not be that the
controller is controlling what some undefined other
person is perceiving? The logic is the same.).

                    You're denying my assertion that a

disturbance at that point in the loop can be
resisted.

             No I most definitely am not!!! If it were not

resisted, how could the corresponding perception be
controlled?

                    In the case of a pathology, it is certainly

the case that control is impaired, as I said.
Example: before the invention of corrective
lenses, my astigmatism would require me to rely
on others to make out details of a scene and
report them, as would my relatively slight
myopia.

                      Look, all I'm trying to do is to emphasize

that PCT is about The Control of Perception,
something that seems in danger of being
forgotten even on CSGnet.

                    Yes, but consider the context. This was in

response to a person who denies that Qi is
controlled at all.

             I guess we bring different prior assumptions to

our reading of what always must be ambiguous, and that’s
foubly true of someone whose first language is no
variety of English. I often disagree with what Boris
says, but on this I have read Boris as simply pointing
out that the output affects Qi in order that
perception is controlled. I say the same, but I am not
usually told I don’t understand PCT – at least not in
the 20 or so years since Bill challenged Rick when he
made that claim, saying something along the lines of
“Who do you think you are saying doesn’t understand
PCT?”

                    I agree with Rick: if that were the case,

there would be no way for an observer to notice
the fact of control. No stabilization of the
environment against disturbances would be
perceptible to anyone except the organism that
was doing the controlling.

Why not? I never perceived you as being of the
all-or-none Black or White persuasion, but here you are
saying that if (as must be the case) an observer has a
different set of inputs to the senses than the person
doing the controlling, no matter how similar their
inputs and perceptual functions may be, what the
observer sees must be totally unrelated to what the
controller sees. Sure, if the controller is controlling
the placement of a glass on a table, and the observer is
looking at the degree to which a door is open, the
observer will say there’s no control. But that’s not
what we are talking about, is it? The observer sees the
glass on the table, and if he wants to know whether the
controller cared where it was placed, the observer can
become an experimenter and move it. The fact that they
see it from different angles may matter, but probably
doesn’t.

                Martin

/Bruce

                      On Tue, Nov 24, 2015 at

11:59 PM, Martin Taylor mmt-csg@mmtaylor.net
wrote:

[Martin Taylor 2015.11.24.23.37]

                            On 2015/11/24 10:37 PM, Bruce Nevin

wrote:

                              [Bruce Nevin (2015.11.24.

ET)]

                                      Martin

Taylor (2015.11.24.14.02) –

                                      "PCT"

doesn’t imply it. It’s simply
a fact of life (and of
engineering) that ONLY if the
connection from Qi to the
perceptual variable is
invertible, perfect, and
noise-free will there be no
difference between the ECV
(whatever that may be) and the
perception. The perception is
controlled, and as a
consequence, the environmental
variable appears to be. As an
approximation, it’s good
enough for most purposes, but
like Newtonian gravity, it’s
not a good foundation for
theoretical discussion or
precise analysis.

                                    In

the equations that I’m familiar
with the connection from * Qi*to p is
represented by a constant Ki .

                          So it is, but how realistic do you think

that is in the real world of live
organisms?

                                    Hasn't

that sufficed for implementing
simulations, or have I missed
something?

                           As I said: " As an

approximation, it’s good enough for most
purposes,". Does anyone claim that the
simulations actually represent what goes
on inside the organism? Even the concept
of a neural current has no equivalent in
an actual brain. It’s an analytical
convenience, an abstraction that simply
assumes that the effect of a lot of
neurons firing with their own timings is
the same as though one super-neuron
performed all the firings, and then
smeared them across time so that a
smooth variation was used in further
functions. For most purposes, that’s
fine, but if you really want to think
about it, Bill just said that if it’s
within a few percent (5%, 2%, I forget)
that’s good enough. And it usually is.
But it doesn’t mean that it’s perfect.

                                    Any imperfection and noise in

the biological implementation is
just another disturbance.
Disturbances can enter at any
point in the loop.

                           Indeed, but when they appear

between the controlled variable p and the
complex environmental variable (the CEV)
to which it corresponds, all that means is
that the CEV is not controlled as
precisely as the perception is.

                                    If such disturbances could not

be countered by the control
process in the same way that
environmental disturbances are,
and if they were great enough to
make p depart from its
correspondence to the relevant
aspect of the environment, as
represented by Qi , they
would be pathologies making it
less likely for that organism to
succeed in bringing offspring to
reproductive maturity, so there
is obvious evolutionary pressure
for that coupling to be quite
good enough to support good
control.

                           Yes. That, in essence, is what I

said when I said “it’s good enough for
most purposes”. You have read a lot of my
writings. How often have I written in
things addressed to PCT newbies that
though what is controlled is perception,
it’s what happens in the environment that
matters?

                                    That seems to me a pretty strong

basis for that coupling being
treated as a constant Ki .
rather than as a variable
subject to significant
unpredictable perturbations.

                           Much more likely to be some kind 

of approximation to log(Qi) with some kind
of ceiling and some kind of zero-region
tolerance zone.

                          It doesn't matter, what the function is,

if it’s invertible. noise free, and
consistent (which adapting systems are
not).

                          Look, all I'm trying to do is to emphasize

that PCT is about The Control of
Perception, something that seems in
danger of being forgotten even on CSGnet.

                              Martin

/Bruce

                                  On Tue, Nov

24, 2015 at 2:44 PM, Martin Taylor
mmt-csg@mmtaylor.net
wrote:

                                      [Martin Taylor

2015.11.24.14.02]

                                            [From Rick

Marken
(2015.11.22.0950)]

                                       No. p1 is the

controlled quantity, the
perception that tracks the
reference value -d2 closely if
all the g values are
substantially greater than
unity. In a real control loop,
of course, the “g” multipliers
would represent the long-term
stable values of the leaky
integrators, just as in the
usual analysis of the simple
control loop. The actual loop
could not use simple
multipliers. When there are
loop delays, simple
multipliers inevitably lead to
oscillation and no control. My
analysis was of the stable
equilibrium values, and for
that, the leaky integrators
are well represented by simple
multipliers.

                                      And what's an "ECV"?
                                       Yes, I am interested.

It’s good to have the
spreadsheet example. What gain
and leak rates did you use for
the four “g” functions, to get
the “g” multipliers? The
effect of d4 and d3 is
diminished by the multiplier
ratio each step back round the
loop. My analysis assumed,
g>>1, as we do when we
do an equilibrium analysis of
the ordinary control loop and
assume the loop gain
G>>1.

                                      I imagine that in your

spreadsheet you have a scalar
variable and a simple
multiplier, as I showed in the
example. One can’t actually
run the TCV on a single
scalar, because there is no
function to be found. But it
would be interesting to run a
spreadsheet example in which
each of the paths was a vector
of, say, three scalars, and
each perceptual function was
different, and then run the
TCV to see what you find.

                                      "PCT" doesn't imply it. It's

simply a fact of life (and of
engineering) that ONLY if the
connection from Qi to the
perceptual variable is
invertible, perfect, and
noise-free will there be no
difference between the ECV
(whatever that may be) and the
perception. The perception is
controlled, and as a
consequence, the environmental
variable appears to be. As an
approximation, it’s good
enough for most purposes, but
like Newtonian gravity, it’s
not a good foundation for
theoretical discussion or
precise analysis.

                                       That's at least

equally wrong. I think it
might be worth your while to
look a little more closely
into the actual conditions for
using the TCV, and the
potential and limitations on
what you can determine by
using it. You often seem to
suggest (planning in
imagination) that you might
use the TCV in real-life
situations. Sometimes the
conditions are suitable, but
much more often, they aren’t.
I haven’t done it, so I am
also planning in imagination,
but one ought to be able to
run the TCV on your demo of a
three-level control system to
find what is being controlled
at the top level. You have all
the outputs and disturbances
necessary, so it should work.
But what about in a real-life
situation in which the
circumstances never recur. In
the hammering example, this
might be the only time in the
hammerer’s life that he is so
angry with his wife that he
has to hit something, and
doesn’t want to hit his wife.
How can the TCV be used in
that situation?

                                      And how do you use the TCV

when control is poor? If you
get a poor compensation of the
disturbance by the output, how
do you know whether you
haven’t found the controlled
variable or you have found it
and the control system doesn’t
work very well?

                                      I think your statement is

simply equivalent to saying
“PCT research is impossible”
which is something I don’t
believe, though I do believe
that the control of perception
accounts for what we see
people and other organisms do,
and that we should carefully
study by all available means
just how this works.

                                          Martin
                                                    Bruce

Nevin
(2015.11.21.20:44
ET) to Martin
Taylor

                                                      BN: Thanks

for this nice
demonstration
of the
difficulty
with this
distinction
[between ECV
and p).

                                                  RM: I think

Martin aimed to
demonstrate that
controlling an ECV
is not equivalent
to controlling the
perception that
corresponds to
that ECV. But, in
fact, his
demonstration
doesn’t
demonstrate that
at all. What it
demonstrates is
that when you put
a bunch of
disturbances into
the feedback
connection between
output and input
you control
neither the ECV
nor p. To be
precise, the
disturbance
variables, d3 and
d4. enter the loop
after the output,
p2, and before the
input, (p4+d1).
See Martin’s
diagram below:

                                                  RM: When these

disturbances are
present the effect
of output (p2) on
input (p4+d1) is
constantly
changing. However,
if you remove
these disturbances
from the feedback
function control
is restored.

                                                  RM: In this

control loop p4+d1
is the controlled
quantity (q.i,or
ECV) and
g1*(p4+d1) is the
controlled
perception, p.

                                                  The only

difference between
q.i (the ECV) and
p is the scaling
factor, g1. But
variations in p
(p1) are perfectly
correlated with
variations in the
ECV (p4+d1); the
only difference
between p and ECV
is that the
former is measured
in neural firing
rate units and ECV
is measures in
physical units;
g1 is just a
scaling factor
that converts
physical units
into neural firing
rate units .

                                                  RM: I've

implemented
Martin’s model in
a spreadsheet, in
case anyone is
interested. It
allows you to see
how well the
perception, p1,
and corresponding
ECV (p4+d1) are
controlled when
the disturbances
to the feedback
function (d3 and
d4) are in or out
of the loop. When
these disturbances
are in, control of
both p1 and ECV is
poor but the
correlation
between variations
in p1 and the ECV
is 1.0; when these
disturbances are
out control of
both p1 and ECV
are excellent; and
the correlation
between p1 and ECV
is again 1.0.

                                                  RM:P I don't

know how people
got the idea that
PCT implies that
there is a
difference between
control of a
perception and
control of the
corresponding ECV.

                                                  But it's an

idea that is not
only wrong but one
that, if believed,
make PCT research
impossible.