Control of perception

[From Rick Marken (940726.0850)]

Bill Leach (940725.21:17) --

To a young child such as I was at the time, experiencing a "simple"
electro-mechanical system that "controlled" was pretty unique.

This post was wonderful, Bill. A very nice description of the "magic" of
control system behavor. So it is with some reluctance that I make what may
seem like a picky comment on something you say in another post, specifically:

Bill Leach (940725.19:19) --

The real issue is that "of course the effect of the disturbance is 'in'
the perceptual signal" but with a good high gain control system the
magnitude is divided by the available gain and is in the noise. Thus,
impractical for use.

This is not really the reason why there is no information about disturbance
in the perceptual signal. The real reason is quite simple and it is embodied
in this little equation:

p = o + d

The perceptual representation of the controlled variable, p, is AT ALL TIMES
the simultaneous result of variations in disturbance(s), d, AND the system's
own output, o. This is the fact that Bill Powers (940725.0600 MDT) was
alluding to when he said:

The state of the controlled variable, furthermore, is not a unique indicator
of the state of the disturbance, because it is simultaneously affected by
the system's own output

The fact that p does not inform the system about the state of disturbances to
the controlled variable may not be a particularly important point in control
engineering; but it is a very important point when one is trying to
understand the behavior of living systems (including oneself). The idea that
p is "informative" suggests that the outputs (visible actions) of living
systems are "guided" by the information in p. This is just S-R psychology,
using fancy terminology. It preserves the fundamental conceit of S-R
psychology; that what we do depends (in some way) on what happens to us.
Saying, as Martin Taylor (940724 20:00) does, that:

the information that allows the output to oppose the disturbance is
obtained through the perceptual input function

may sound "close" to PCT, but it is really one of those "baby steps" Mary
mentioned, that leaves you stranded on the banks of S-R psychology-land.
Martin's statment isn't "sort of true", or "close to true" or, as Martin
claims "necessarily" true. It is just flat out wrong; it is a restatement of
the view of behavior that PCT shows to have been based on an illusion -- the
"behavioral" illusion described in B:CP and elsewhere.

The behavior of organisms is not guided, informed, caused, determined or
controlled by perceptual inputs. Perceptual inputs are guided, informed,
caused, determined -- well, controlled by the actions of the organism.

Big difference:-)

Best

Rick

[Martin Taylor 940726 17:30]

Rick Marken (940726.0850)

Saying, as Martin Taylor (940724 20:00) does, that:

the information that allows the output to oppose the disturbance is
obtained through the perceptual input function

may sound "close" to PCT, but it is really one of those "baby steps" Mary
mentioned, that leaves you stranded on the banks of S-R psychology-land.
Martin's statment isn't "sort of true", or "close to true" or, as Martin
claims "necessarily" true. It is just flat out wrong;

Yes, Rick has on several occasions asserted that he prefers the magical
to the physical explanation of control systems.

Bill Powers is closer the mark when he says that a signal processing analysis
of control systems is adequate, and that information theory doesn't make for
any more precise description of the behaviour of a control system. As far
as it goes, that statement is correct. What it omits is that taking different
viewpoints on the same process allows you to see different aspects more
readily.

Bill Leach is wrong in arguing that because the effect of the disturbance is
lost in the noise, it does not passinformation through the perceptual
function. It does, and it is the noise limit that eventually determines
(one aspect of) limits to control. Bandwidth and loop delay (transport lag)
in the control system, as related to the bandwidth of the disturbance effect
(Note, Bill P.: not the disturbance or the disturbing variable) are other
factors.

Fascinating to observe.

Martin

<[Bill Leach 940727.17:04 EST(EDT)]

[Rick Marken (940726.0850)]

... So it is with some reluctance that I make what may seem like a picky
comment on something you say in another post, specifically:

This is not really the reason why there is no information about
disturbance in the perceptual signal. The real reason is quite simple
and it is embodied in this little equation:

p = o + d

Bill Powers (940725.0600
MDT) was alluding to when he said:

The state of the controlled variable, furthermore, is not a unique

indicator of the state of the disturbance, because it is simultaneously
affected by the system's own output

This is wierd! I am arguing with both you and Martin and I don't even
think that I am controlling for being a devil's advocate!

In the first place, I have not disagreement with Bill's posting that you
quoted in any fashion whatsoever. As is typical of Bill's usual form,
the statement is not only succinct but it is also complete for not only
the theoritical but for the real (or practical) case as well.

All I have been trying to state is that in a real control system, a
portion of what Martin is asserting is true (and as best I can tell Bill
P. agrees). OTOH, I also agree with Bill, that while certain assertions
in Martin's arguement are indeed valid points, they are irrelevent.

I will try to sum up my own thinking on this subject:

To do so, let consider first the relevent characteristics of a "perfect"
control system:
     Loop gain is infinite
     Loop response time is zero
     Output dynamic range is infinite
     System noise zero

For this sort of system, there would be absolutely no effect of anything
other than the reference upon the perception. A change in disturbance
"could be seen" by monitoring the output but that is the only parameter
that could indicate that there even was a disturbance or change in same.

Such a system would not need any stabilizing or compensation to operate.

A real control system then has a few shortcomings when compared to the
perfect system:

     Loop gain is less than infinite
     Loop response time is greater than zero
     Ouput dynamic range is less than infinite
     System noise is greater than zero

The significance of this (from only a theoritical standpoint as far as I
am concerned and as a practical matter if control exists then p=o+d);

     Since loop gain is less than infinite, any change in the disturbance
     requires a steady state offset in the perception to be sustained to
     generate the error signal that can then only partially compensate
     for the disturbance. The "only" there however means that
     compensation is not perfect as opposed to implying that the
     correction will be "poor". ie: The control error might now has
     "risen" to .0001% or some similar minor value. The magnitude of
     this offset to the perception necessary to generate the needed
     change in the error signal is "related" to the magnitude of the
     disturbance as a function of the capability of the output portion to
     overcome the change in the actual condition in the environment.

     Since loop response time is greater than zero, the perception "will
     track" a change in the disturbance to the extent that the
     disturbance is faster then the control system response time. This
     is an overly simplistic statement but basically means that when the
     perception changes, time is required before the output will act to
     restore the perception.

     Output dynamic range is less than infinite is not relevent to the
     discussion that we have been having.

     System noise is greater than zero. This factor complicates the
     picture greatly (from a theory standpoint but again, I don't see
     where it mean much otherwise). In the first place the control
     system can only achieve a control resolution that is greater than
     the level of system noise. In the second place the control system
     will control perception to that resolution. Thus, any steady state
     disturbances (slow compared to the system response time) will BE in
     the noise making determination of the presence of a disturbance
     difficult if you are measuring other loop parameters and probably
     impossible if you are not -- BTW, the you here is an observer that
     through some "magic" has access to the perceptual variable AND
     specifically NOT the control system itself).

In a real control system, perception MUST change in response to a
disturbance that actually effects the condition in the environment that
is being controlled inorder for the control system to even generate a
changed output to compensate. I fully accept the adamant opposition to
any claim that this is relevent to the idea of control. In the first
place to one that does not understand control, this sounds an awful lot
like stimulus-response but for a high gain control system (say loop gain
of a million) then it is about a million to one removed from stimulus-
response type behaviour.

In a control loop that is "well defined" and this definition is known,
the it is probably possible to extract "information" about the disturbance
through analysis of the perceptual variable. In the first place, I
hesitate to even include the adjective "useful" to the term information.
Also, performing this amazing feat (other than theoritically) would
require that the control system remain absolutely consistant AND that
there actually is someway to "get ahold of" the perceptual signal. I
know that doing the second part is "possible" but I gather luck is the
major player. I don't think that biological control systems in complex
biological entities are either well defined nor are they "absolutely"
consistent, nor do they need to be, they can control quite well with
small variations in such things as gain. These considerations would
probably, in my opinion, make any sort of signal analysis of the
deviations of the perceptual signal within the noise useless.

I don't know (but am sure that I will find out) how well I did with this
explaination but this is the justification for both Martin's statement
that PCT is not magic and Bill's assertion that the information is
irrelevent -- that is irrelevent absent any proof via a functioning
model.

The fact that p does not inform the system about the state of
disturbances to the controlled variable may not be a particularly
important point in control engineering; but it is a very important point
when one is trying to understand the behavior of living systems
(including oneself).

Engineered control system might well "take advantage" of known
characteristics of the environment that can be identified through the
perceptual signal effect additional control actions. I am not so sure
however, that this idea is necessarly at variance with PCT. The physical
manner in which something like that is accomplished could well be "out of
line" but the idea that a single perceptual signal can be the input to
more than one control loop, each with its own reference would not
necessarily be a problem. I think that how such an idea is viewed can be
a problem. In the first place, the organism doesn't "know" things in the
same fashion as the control systems engineer (at least thinks) that he
knows things about what is being controlled.

Also, I don't see any problem with the idea that a perceptual signal
could be the single input to one control loop and at the same time be
combined with other perceptual signals to make up yet another perceptual
signal that would then be the input to another control loop.

However, none of this has anything to do with disturbance and much of
what little I understand about "modern feedforward" control sounds to me
as though it is nothing more that negative feedback control of yet
another perception to its' appropriate reference. That the perception is
made up (at least in part) of a perception that is the perception for
another control loop does not, in my mind, in and of itself invalidate my
thoughts on that matter.

The idea that p is "informative" suggests that the outputs (visible
actions) of living systems are "guided" by the information in p. This is
just S-R psychology, using fancy terminology. It preserves the
fundamental conceit of S-R psychology; that what we do depends (in some
way) on what happens to us. Saying, as Martin Taylor (940724 20:00) does,
that:

the information that allows the output to oppose the disturbance is
obtained through the perceptual input function

may sound "close" to PCT, but it is really one of those "baby steps"
Mary mentioned, that leaves you stranded on the banks of S-R
psychology-land. Martin's statment isn't "sort of true", or "close to
true" or, as Martin claims "necessarily" true. It is just flat out
wrong; it is a restatement of the view of behavior that PCT shows to
have been based on an illusion -- the "behavioral" illusion described in
B:CP and elsewhere.

IF I have been understanding Martin correctly, he has not been asserting
that the control system itself can extract information about the
disturbance by analyzing the perception. My understanding is that he is
asserting that IF an outsider could monitor the perceptual variable then
through the application of IT, some information about the disturbance
could be obtained.

As far as I am concerned, it has not been DEMONSTRATED that ANY
information can be obtained (much less "useful" information). My
assertion in this matter is that knowledge of ONLY the perceptual signal
magnitude would not allow an observer to determine anything. The
observer would at least have to have information concerning the stability
of the output system and the reference to even know if non-noise
variations were even caused by effects external to the control system.
Assuming just that much knowledge, you would then (at least in theory)
only know that SOMETHING was varying the controlled perception besides
the control system. You would not know anything useful.

The behavior of organisms is not guided, informed, caused, determined or
controlled by perceptual inputs. Perceptual inputs are guided, informed,
caused, determined -- well, controlled by the actions of the organism.

I probably should apologize for quoting such a large segment, but since I
had comments to this that might sound as though I disagree with the
"message" here when in fact I believe that the "message" is the single
most important concept in PCT, a full quote seemed the best choice.

Big difference:-)

And finally, yep, I agree.

-bill

<[Bill Leach 940727.19:14 EST(EDT)]

[Martin Taylor 940726 17:30]

Bill Leach is wrong in arguing that because the effect of the
disturbance is lost in the noise, it does not pass information ...

Did I make myself any clearer in the other post to Rick today?

-bill

[From Rick Marken (930921.2000)]

Hans Blom (930921)--

From this follows an important point: It is as if an individual's
reference signal -- which we assume internal in the individual -- can
be modified simply by distorting (offsetting) his observations --
which can be done by an outside agent. In this sense, the discrepancy
between the "inside" (where we assume the source of "autonomy") and
the "outside" (where the "manipulators" live) is just a change of
sign. But this must be too philosophical for you :slight_smile:

Nope. It's just straight PCT (no ice). Your description here makes it
clear that the r in your diagram is part of our perceptual function
(the function that transforms some objective state of affairs, like
the position of the target, t, into a perceptual variable). In this
case the perceptual function just adds an offset to t. So the per-
ception that is controlled is c - t+r. r can be thought of as a prism
that displaces t from it's actual position (if the prism is not angled
relative to the line of sight then the displacement is zero -- r = 0
-- and the controlled perception is c - t). Of course, the prism is
only in front of t (because r only adds to t).

An observer who can see t and c is, indeed, likely to conclude that
the the intended distance between c and t is -r. So when the
"actual" intended distance between c and t is 0, the observer
will see this distance maintained at -r (to compensate for the
displacement of t created by r -- the prism). It is also easy
to make it seem to the observer that the subject is varying the
reference for the distance between c -t even when the subject
is maintaining a fixed reference for that distance; just keep
changing the angle (and, hence, displacement) of the prism. The
subject must continuously vary the actual distance between t and
c in order to keep the perceived distance fixed at the intended
(reference) level.

Your important point is very important indeed, Hans. It is a good way
of showing that people don't control what's "really out there";
they control their perceptions.

It is possible, by the way, to use "the test" to show what the person
is actually controlling in this "prism distortion" situation. Once
the experimenter notices the prism, it would be easy to compute the
expected distance between c and t if the subject has a fixed reference
for the distance between c and the prism distorted t. The hypothesis
would be that the subject is keeping c - t+r = 0 (for example).
When variations in the two manipulable variables (t and r) results
in a sum (c-t+r) that remains nearly zero, there would be strong
evidence that this variable is, indeed, controlled relative to a fixed
reference.

So you do see that control systems control perception, Hans. Maybe
you just don't like to say it that way?

Best

Rick