Multiplexing, alerting, degrees of freedom

CSGnet Archive CSG1994
1

[From Bill Powers (940719.1515 MDT)]

Martin Taylor (940718.1850) --

Neither of them is where MY concept of Alerting comes from. It comes
from the requirement that Bill P. noted as "multiplexing." You simply
don't have as many output degrees of freedom by a long shot as you have
perceptual degrees of freedom.

"Multiplexing" outputs means applying them (along with the associated
perceptual functions) briefly to one task, then to another, then to
another, then back to the first, and so forth. This is possible only
when the controlled variables, because of inertia or other slowing
effects, can change only very slowly compared with the speed with which
we can switch control from one task to another. You can get away with
correcting a car's position on a straight road only every one to two
seconds, leaving the steering wheel fixed between corrections.
Fortunately, most of the things we control by multiplexing change far
more slowly than a speeding car's position does. We need to correct
hunger signals only a few times per day. The relationships among objects
in our houses require active control only when we have some use for
them, or when someone else disturbs them; you might go for days without
needing to correct the position of the sofa within the room.

Multiplexing must be one facet of the more general class, sequence
control. The sequence in which things are done often makes a difference
in the outcome (as in rotating an alphabet block by 90 degrees in pitch
and in yaw). Sometimes sequence is required just because of the nature
of the world: you might try to fill a glass with water and drink from it
at the same time, but both goals will be accomplished better if done in
the right sequence, one at a time. Multiplexing is just one kind of
sequencing of control processes at lower levels.

After listening to your arguments for some time now, I am convinced that
the supposedly huge excess of input degrees of freedom over output
degrees of freedom is largely a myth. I have told you in private posts
why I feel your calculations are spurious; we can continue that debate
in public if you like. The main limitation on input degrees of freedom
is not the number of individual sensory nerve endings we have; it is the
fact that large parts of any perceptual field are constrained to change
in a coherent way because of the coherence of objects; even the objects
are constrained to behave in specific relationships to each other
because of the nature of the external world. We have enough input
degrees of freedom to handle an environment that comes in rather large-
sized chunks. When it begins to come in larger numbers of smaller
chunks, we fall behind perceptually just as much as output-wise. If all
the pixels in a moving TV picture were being simultaneously disturbed,
you would have a hard time controlling them even if you had one output
per pixel.

I don't believe that there is any great difference between input and
output degrees of freedom. When you switch from steering your car to
tuning the car radio, you not only switch one hand from helping to steer
to turning the knob, you switch the two-degree angle of acceptance for
clear vision from the road to the radio dial. In fact in that case you
have more output degrees of freedom available than input degrees of
freedom. You can't be perceiving the car's position relative to the road
with enough precision for steering at the same time you are perceiving
the position of the needle on the dial with enough precision for tuning,
even though you still have one hand free for steering. We multiplex BOTH
inputs and outputs, because the degrees of freedom of BOTH are limited.
In fact, I think we are hard-pressed to keep our perceptual worlds
updated, even regarding the consequences of our own actions.

If there is no remarkable difference between input and output degrees of
freedom, then the entire edifice you're building on a hasty assumption
of such a difference collapses.

···

-----------------------------------------------------

Uncontrolled perceptions are simply uncontrolled perceptions. That is
to say that when they change value, NOTHING HAPPENS. Why, then, should
we even have perceptual functions to generate those perceptions? They
must sometimes be controlled.

That doesn't follow at all. When a dog is chasing a cat, the
relationship between dog and cat depends on the dog's position and the
cat's position. The dog can control only the dog's position; the cat's
position, which is an equally important part of the relationship
perception, is uncontrollable although it is perceived. The dog can
control this relationship, yet it can never control the part of it due
to the cat's position (until it has caught the cat). Nor does it need
to.

Most perceptions are functions of controlled and uncontrolled
perceptions. As long as changes in the uncontrolled perceptions can be
countered by changes in the controlled ones, the higher-level perception
can remain under control. There is no suggestion here that the
uncontrolled perceptions must come under control. Some of them might,
but there is no requirement that they do.

If nothing happens when an uncontrolled perception changes, then there
is no reason for that perception even to exist. If the bell sounds but
no ice-cream wagon ever appears, perception of the bell will never
become part of the perception of the location of an ice-cream wagon.
Likewise, if one has no unfulfilled reference conditions that can be
satisfied by an ice-cream wagon (i.e., you don't care about getting any
ice cream), there will be no reason to control one's relationship to the
bell (and wagon).

I question whether your concept of "alerting" came from a consideration
of degrees of freedom. I suggest that it existed long before you ever
had such thoughts. "Alerting" or "orienting" signals were proposed long
before you became a psychologist; they came out of the days of primitive
psychological theorizing when every observable phenomenon had its own
specific dormitive principle to make it go. The concept points us toward
a phenomenon, to be sure. But once it has done that, it is of no further
use. We must find a place for the phenomenon in the model (and I don't
think that is hard to do, even without baroque flights of reasoning),
but once we have done that the phenomenon is simply another example of
hierarchical control behavior. The phenomenon would become really
interesting if it challenged the theory. But it does not.
--------------------------------------------------------------------
Best,

Bill P.

2

<[Bill Leach 940720.00:40 EST(EDT)]

[From Bill Powers (940719.1515 MDT)]

I really like your discussion on degrees of freedom and multiplexing.
While I have not tried to dwell on the "degrees of freedom" issue, it had
raised a rather uncomfortable "flag" for me (is that an "Alerting"
signal?). :slight_smile:

I suspect that we do experience overloading of both input and output
degrees of freedom and such an experience is felt as anxiety. "Common
Wisdom" says that the so called "Alcholics Prayer" is probably the
insight to the solution to such a problem: "God (or whatever) grant me
the serenity to accept the things that I can not change, the courage to
change the things that I can... and the wisdom to know the difference!".

Martin:

Uncontrolled perceptions are simply uncontrolled perceptions. That is
to say that when they change value, NOTHING HAPPENS. Why, then, should
we even have perceptual functions to generate those perceptions? They
must sometimes be controlled.

Bill P.

That doesn't follow at all. When a dog is chasing a cat, the
relationship between dog and cat depends on the dog's position and the
cat's position. The dog can control only the dog's position; the cat's
position, which is an equally important part of the relationship
perception, is uncontrollable although it is perceived. The dog can
control this relationship, yet it can never control the part of it due
to the cat's position (until it has caught the cat). Nor does it need
to.

There is a serious terminology problem here (I think). The behaviour of
the cat is NOT a perception with regard to the control of the relative
position between the cat and the dog that the dog is controlling. It is
rather a disturbance.

Now that the dog may indeed perceive things about the cat's behaviour may
well be possible (sounds like perception of disturbance huh?!). I don't
think however, that this is really the case. The dog, just as humans
can, could conceivably perceive it's own behaviour (within the same sort
of limitation that human can do so). Thus, the dog might notice that if
he lunges at the cat in a certain way that the cat "responds" in a
consistant manner and then use this knowledge to develop a strategy to
catch the cat.

Most perceptions are functions of controlled and uncontrolled
perceptions. As long as changes in the uncontrolled perceptions can be
countered by changes in the controlled ones, the higher-level perception
can remain under control. There is no suggestion here that the
uncontrolled perceptions must come under control. Some of them might,
but there is no requirement that they do.

Again, I am having real trouble with this one. As I understand it (and
of course that can very well be wrong), a perception that is controlled
can not have uncontrolled perceptions as a part of THAT control loop. If
it did, any change in the uncontrolled perception would result in a
maximum error for the loop.

It seems to me that we can talk about perceptions that are related to a
controlled perception that are themselves not controlled and indeed there
might even be more such perceptions than the controlled variety but to
the extent that these perceptions affect the controlled perception they
are a disturbance and can not actually be perceived.

Even in my example (using your example of the dog chasing the cat), it is
probably only past control failure viewed after the fact that the dog
could employ to "learn of the nature" of the disturbance -- the
effectiveness of the cat at surviving.

If nothing happens when an uncontrolled perception changes, then there
is no reason for that perception even to exist. If the bell sounds but
no ice-cream wagon ever appears, perception of the bell will never
become part of the perception of the location of an ice-cream wagon.
Likewise, if one has no unfulfilled reference conditions that can be
satisfied by an ice-cream wagon (i.e., you don't care about getting any
ice cream), there will be no reason to control one's relationship to the
bell (and wagon).

I would say that this is true but I also think that one has to realize
that many perceptions are very subtle. For example, if I were the hear
the chimes of an Ice Cream Wagon right now, I might be inclined to think
of it as an "Alerting signal" since in the first place it is after 1:00 AM
and in the second place I have never heard such a sound since I moved to
my present location. I don't think that I have the experience with PCT
to identify the relevent perceptions but I don't have a problem with the
idea that I have "some sense" of the "abnormal sensation" from my
environment and that I have a perception for determining the nature of
abnormal occurances. Thus, though I have no reason whatsoever to be
thinking of an Ice Cream Wagon Chime the perception of one would generate
an error signal for one or more perceptual control loops (loops that have
been idle in my case for years).

-bill

3

From Tom Bourbon [940720.1653]

<[Bill Leach 940720.00:40 EST(EDT)]

[From Bill Powers (940719.1515 MDT)]

. . .
With a jump by Bill Leach to an uncited post by Martin Taylor:

Martin:

Uncontrolled perceptions are simply uncontrolled perceptions. That is
to say that when they change value, NOTHING HAPPENS. Why, then, should
we even have perceptual functions to generate those perceptions? They
must sometimes be controlled.

Bill P.

That doesn't follow at all. When a dog is chasing a cat, the
relationship between dog and cat depends on the dog's position and the
cat's position. The dog can control only the dog's position; the cat's
position, which is an equally important part of the relationship
perception, is uncontrollable although it is perceived. The dog can
control this relationship, yet it can never control the part of it due
to the cat's position (until it has caught the cat). Nor does it need
to.

Bill L:

There is a serious terminology problem here (I think). The behaviour of
the cat is NOT a perception with regard to the control of the relative
position between the cat and the dog that the dog is controlling. It is
rather a disturbance.

??? The dog perceives the cat. The dog intends a specific distance between
itself and the cat. The present perceived distance (the perception of which
distance just might include the dog perceiving the cat :wink: is not the
intended one. The dog acts to reduce the error, whereupon the cat acts to
decrease the now-increasing error for its own intended perception of great
distance between itself and the dog -- the dog that the cat perceives
closing in from over there. The cat's movements certainly do disturb the
dog's perception of the distance the cat is away from the dog's intended
perception, but that is only because the dog _does_ perceive the cat, the
"actual" position of which (reckoned independently of the position of the
dog) is _not_ controlled by the dog.

Reminds me of a tracking experiment, but then, _everything_ reminds me of
a tracking experiment. The position of the target is independent of
anything the person does with the control manipulandum, but the person
must perceive the target in order to make the cursor track it.

Now that the dog may indeed perceive things about the cat's behaviour may
well be possible (sounds like perception of disturbance huh?!).

It sounds more like perception of cat. :slight_smile:

. . .

Bill P.

Most perceptions are functions of controlled and uncontrolled
perceptions. As long as changes in the uncontrolled perceptions can be
countered by changes in the controlled ones, the higher-level perception
can remain under control. There is no suggestion here that the
uncontrolled perceptions must come under control. Some of them might,
but there is no requirement that they do.

Bill L.

Again, I am having real trouble with this one. As I understand it (and
of course that can very well be wrong), a perception that is controlled
can not have uncontrolled perceptions as a part of THAT control loop. If
it did, any change in the uncontrolled perception would result in a
maximum error for the loop.

Try thinking about tracking. Perceptions of cursor and target are just
that. If the person has specified a particular _relationship_ that will
exist between cursor and target, there must also be a perceptual function
that produces a signal for the present value of the relationship. The
person perceives the target, the cursor, and the relationship between cursor
and target, all of them at once. If the task is pursuit tracking, then
_something else_, unperceived by the person, determines the position of the
target, independently of anything the person does. In a tracking
experiment, "something else" is a computer procedure (algorithm, routine)
that calculates momentary values of a target-driving function that is turned
into the plotted position of the target on the computer screen. The person
does see the uncontrolled position of the target, but does not see the
routine that produces target positions -- the person does not see that which
"disturbs" the position of the target. The driving function is the
disturbance.

How might that analysis of a tracking task help clear up the question of
whether the cat is a disturbance that is seen by the person? I have already
suggested that the dog can simultaneously perceive the cat and the cat's
position relative to the dog. The cat's position changes relative to that
of the dog, or, to take the perspective of the dog, the dog sees the
four-legged hairy thing with a tail, but that is not another dog, moving
about in its visual field and disturbing the intended perceived distance.
Does the dog see the disturbance? No. The dog sees the position, relative
to itself, of a mass of hair and claws and teeth; the position of that mass
is determined not by itself, but by the actions of the skeletal muscles on
the skeleton of the cat, and the actions of the skeletal muscles are
determined by error signals coming from comparators in the cat's nervous
system, and those error signals are affected by a comparison (inside the
cat) of the cat's actual and intended perceptions of the distance of the
cat from the dog.

The control system that is the cat's nervous system is the source of the
disturbance that affects the dog's perception of the position of the cat
relative to the dog. The dog does not see the "disturbance;" the dog sees
the mass of hair that gets dragged along (as a side effect) when the cat's
nervous system controls its own perceptions of relative distance.

Bill Powers keeps reminding people that in PCT modeling we must maintain a
distinction between a _disturbance_, defined as an independent source of
influence on variables perceived by a control system, and the _effects_ the
disturbance has on the perceived variables. If you think about what happens
when a person chases a moving target on a computer screen, or a dog chases
a moving cat in the alley, maybe our reasons for maintaining the distinction
will seem a little clearer.

Later,

Tom

4

The dog and cat systems are simple in the example chosen. If both dog
and cat are consider within the master's house and the intended
control is on the antipathy which may be consider causal with respect
to the pursuit of distance change between the animals the issue
becomes a bit less clear. The former "disturbance" from either the
simple cat system or the simple dog system perspective is now an
independent variable used in deriving the antipathy system variable of
the master of dog and cat system. The antipathy variable would be
calculated upon the history of the rate of change in the distance
between the dog and cat in conjunction with other indicators, e.g., a
slow approach may appear as a "stalking" or a "teasing" with different
influences on the evaluation of dog and cat antipathy.

Bill Powers keeps reminding people that in PCT modeling we must
maintain a distinction between a _disturbance_, defined as an
independent source of influence on variables perceived by a control
system, and the _effects_ the disturbance has on the perceived
variables. If you think about what happens when a person chases a
moving target on a computer screen, or a dog chases a moving cat in
the alley, maybe our reasons for maintaining the distinction will seem
a little clearer.

So, is the distance considered a disturbance in a system including
both dog and cat. I will spray one the other or both with the hose
depending on the antipathy indicator. Or am I the one all wet?

Michael Koopman (mike) Associate Computer Systems Specialist
Concurrent Technologies Corporation internet: koopman@ctc.com
1450 Scalp Avenue phone: +1-814-269-2637
Johnstown, PA 15904-3321 USA telefax: +1-814-269-2402

5

<[Bill Leach 940720.22:08 EST(EDT)]

From Tom Bourbon [940720.1653]

Bill L.

Again, I am having real trouble with this one. As I understand it (and
of course that can very well be wrong), a perception that is controlled
can not have uncontrolled perceptions as a part of THAT control loop.
If it did, any change in the uncontrolled perception would result in a
maximum error for the loop.

Try thinking about tracking. Perceptions of cursor and target are just
that.

Ok, I suppose that I see the relationship and understand the meaning of
the term "uncontrolled perceptions". What still bothers me a little is
that in terms of the control loop, it seems that the perception of the
absolute cursor position is incidental to the control action. The
controlled perception is the relative position and control should be fine
as long as the relative position is perceived and changes are within the
response capability of the controller, yes?

I believe that you have done tests where you introduce disturbances to
both the target and the cursor, yes? Was not control roughly the same
for equal magnitude disturbances?

Bill Powers keeps reminding people that in PCT modeling we must maintain
a distinction between a _disturbance_, defined as an independent source
of influence on variables perceived by a control system, and the
_effects_ the disturbance has on the perceived variables.

I must be missing something in this one. If the disturbance to a
variable is related to that variable through a known function then
observation of the effects of the disturbance on a perception relates
back to the disturbance itself.

In the example of the dog and the cat, is not the reference zero and the
controlled perception the relative distance between the two (for the dog
of course)?

If the dog's reference was to maintain a specific distance (greater than
zero) then the dog would not perceive the effects of the disturbance
caused by the cat except to the extent that the control action failed.
That is, if the dog did indeed maintain this reference distance then it
would not know what the cat was doing to disturbe the perception.

Unrelated perceptions, such as the passage through terrain could provide
perceptions about what the cat is doing since they are not controlled
perceptions for the dog. Thus, this dog might later tell its' friends
all about the 'great chase' that the cat provided but the tale would be
generated completely from perceptions that were not being controlled with
the single exception that our intelligent dog would be able to state with
supreme confidence that the distance between the two of them remained
constant for the entire event.

I believe that people, myself included, probably conclude that we can
perceive disturbances (or their effects) from precisely this sort of
additional uncontrolled perception.

Hummm.... how about a Post Script:

"Effects of a disturbance would include actions other than those that
effect the variable in the perceptual control loop, yes?

Thus, the effects of a disturbance can possibly be perceived when such
multiple effects exist and at least one is influencing an uncontrolled
perception AND the behaviour do not incidentally "control" the
uncontrolled perception. That is the multiple effects are at least no
more than somewhat loosely coupled to the controlled perception.

-bill

6

[Martin Taylor 940721 11:00]

Bill Powers (940719.1515 MDT)

It's amazing how much we can agree on while talking past one another!
I agree with ALMOST everything you say in the post that starts:

Martin Taylor (940718.1850) --

Neither of them is where MY concept of Alerting comes from. It comes
from the requirement that Bill P. noted as "multiplexing." You simply
don't have as many output degrees of freedom by a long shot as you have
perceptual degrees of freedom.

"Multiplexing" outputs means applying them (along with the associated
perceptual functions) briefly to one task, then to another, then to
another, then back to the first, and so forth. This is possible only
when the controlled variables, because of inertia or other slowing
effects, can change only very slowly compared with the speed with which
we can switch control from one task to another.

Exactly. We have no disagreement on this, and never have had. Though in
many postings you seem to assume I take a different position. Let's keep
this statement in mind, as one of the basic principles, because that's
what it is.

Multiplexing is just one kind of
sequencing of control processes at lower levels.

Is there another kind? I would take the "sequencing of control processes
at lower levels" to be almost a definition of multiplexing (at least
time-division multiplexing--there are other kinds).

After listening to your arguments for some time now, I am convinced that
the supposedly huge excess of input degrees of freedom over output
degrees of freedom is largely a myth. I have told you in private posts
why I feel your calculations are spurious; we can continue that debate
in public if you like.

All right. Here is probably the only point of disagreement in your whole
posting, and if you would like me to send publicly the response I was
preparing for private communication, I will. It includes some small
quotes from your private message, to which I hope you will not object.

The main limitation on input degrees of freedom
is not the number of individual sensory nerve endings we have; it is the
fact that large parts of any perceptual field are constrained to change
in a coherent way because of the coherence of objects; even the objects
are constrained to behave in specific relationships to each other
because of the nature of the external world.

Yes, that's what allows the whole multiplexing system to work, and why
alerting systems can be effective. I used exactly this same argument
in a commentary on J.G.Taylor's theory in 1973, and I have not changed
my opinion on it since then. Each time you bring up this point, my
response has been either to say in effect "OK, now let's deal with the
issue at hand" or to ignore it as a given.

The fact that we survive as a species is the best demonstration of the
truth of your claim.

We multiplex BOTH
inputs and outputs, because the degrees of freedom of BOTH are limited.
In fact, I think we are hard-pressed to keep our perceptual worlds
updated, even regarding the consequences of our own actions.

Yes, indeed. I see no issue here. The example you use of shifting your
gaze position in order to update your perception of the state of a different
chunk of the real world is a good one. It covers the two modes I called
"search" and "explore." As with any control system, the output that
leads to the action of moving the gaze direction can be initiated because
of external disturbance or internal change of reference value. Either
can affect the error signal in the control system that perceives the gaze
direction. An alerting sound from a particular direction could be such
a disturbance.

Uncontrolled perceptions are simply uncontrolled perceptions. That is
to say that when they change value, NOTHING HAPPENS. Why, then, should
we even have perceptual functions to generate those perceptions? They
must sometimes be controlled.

That doesn't follow at all. When a dog is chasing a cat, the
relationship between dog and cat depends on the dog's position and the
cat's position. The dog can control only the dog's position; the cat's
position, which is an equally important part of the relationship
perception, is uncontrollable although it is perceived. The dog can
control this relationship, yet it can never control the part of it due
to the cat's position (until it has caught the cat). Nor does it need
to.

I suppose you have to make this comment, to the wider audience. But I am
sure you know that often I use the cumbersome but more precise form
"controlled perception or contributing to a controlled perception" when
dealing with the issue. This response is probably unnecessary, except
for self-justification, since I agree wholeheartedly with the rest of this
part of your posting, especially with:

If nothing happens when an uncontrolled perception changes, then there
is no reason for that perception even to exist.

This is a most important comment. It is at the core of J.G.Taylor's theory,
and I have believed it for more than 30 years. I think it is foundational.
It doesn't follow directly from PCT, but is a kind of corollary if you make
the assumption that evolution results in efficient designs most of the time.

I question whether your concept of "alerting" came from a consideration
of degrees of freedom. I suggest that it existed long before you ever
had such thoughts.

Perhaps. That's a question of historical record. As a consequence of your
comment I have just reread my 1972 paper in which I introduced the concept
in the form I now use it, and I find that I did use a degrees of freedom
argument then, though a different and less powerful argument than is
provided by PCT. The current form of the argument is due strictly to my
having learned PCT and understood some of its consequences.

The phenomenon would become really
interesting if it challenged the theory. But it does not.

No, I don't think it does. I never did think it did. As I have said before,
I think of it as one of those many places where PCT makes beautifully clear
something that was separately described, unexplained, and somewhat vague
beforehand. What has puzzled me is why this fact has causes so much
disturbance to the PCT old hands.

Martin

7

From Tom Bourbon [940721.1457]

<[Bill Leach 940720.22:08 EST(EDT)]

From Tom Bourbon [940720.1653]

Bill L.

Again, I am having real trouble with this one. As I understand it (and
of course that can very well be wrong), a perception that is controlled
can not have uncontrolled perceptions as a part of THAT control loop.
If it did, any change in the uncontrolled perception would result in a
maximum error for the loop.

Tom

Try thinking about tracking. Perceptions of cursor and target are just
that.

Bill L.

Ok, I suppose that I see the relationship and understand the meaning of
the term "uncontrolled perceptions". What still bothers me a little is
that in terms of the control loop, it seems that the perception of the
absolute cursor position is incidental to the control action.

That's right. _If_ the person intends to control the position of the cursor
relative to that of the moving target, _then_ the person ignores the absolute
position of the disturbed cursor. Given the person's intended result, there
is no way for her or him to also control the absolute position of the
cursor, which really means to hold its position constant, relative to the
screen coordinates. As you say next:

The
controlled perception is the relative position and control should be fine
as long as the relative position is perceived and changes are within the
response capability of the controller, yes?

Yes. On the other hand, _if_ the person decides to control the absolute
position of the cursor on the screen, then its position relative to the
moving target _must_ go uncontrolled. The situation is like that for the dog
with the bone, in Aesop's fables: if you want to make a grab for the bone
you see reflected in the water, you must first open your mouth, whereupon
you drop the real bone you were already holding.

Speaking of dogs, the same constraint would also apply to the dog and cat
that we began talking about a little earlier, on the subject of distubances.

I believe that you have done tests where you introduce disturbances to
both the target and the cursor, yes? Was not control roughly the same
for equal magnitude disturbances?

Yes, within broad limits on the amplitudes and bandwidths of the
target-driving function and the disturbance to the cursor.

Bill Powers keeps reminding people that in PCT modeling we must maintain
a distinction between a _disturbance_, defined as an independent source
of influence on variables perceived by a control system, and the
_effects_ the disturbance has on the perceived variables.

I must be missing something in this one. If the disturbance to a
variable is related to that variable through a known function then
observation of the effects of the disturbance on a perception relates
back to the disturbance itself.

No problem, so long as you realize that you are talking from the position of
a modeler who knows all of the functions and can see the tables of
disturbance values (perhaps more than one at the same time) that affect the
controlled variable(s), should the need arise. In that role, the modeler
has access to knowledge like that used by the control engineer who designs
and builds a control system. But a living control system doesn't have
access to that kind of knowledge; it neither knows, nor needs to know, the
number or nature of the distrbances that produce a combined effect on its
controlled variables. That said, nothing rules out the possibility that a
living control system, perhaps a person, might discover the sources of some
of those disturing effects, then act to modify them or eliminate them; but
that kind of "search and destroy," or "discover and modify" procedure is not
a _necessary_ element of perceptual control.

In the example of the dog and the cat, is not the reference zero and the
controlled perception the relative distance between the two (for the dog
of course)?

The reference for perceived distance might be zero -- probably so, if the
dog keeps going until it catches the cat. (The opposite for the cat of
course.)

If the dog's reference was to maintain a specific distance (greater than
zero) then the dog would not perceive the effects of the disturbance
caused by the cat except to the extent that the control action failed.
That is, if the dog did indeed maintain this reference distance then it
would not know what the cat was doing to disturbe the perception.

I don't follow you here. If the dog has _no_ reference for "distance to
cat," then nothing the cat does (no relative position of the cat) can
disturb the dog's perception of distance. On the other hand, if the dog has
a reference for a specific non-zero distance, then any deviation from that
distance is error.

By the way, could you make any sense of what I said earlier (Tom Bourbon
[940720.1653]) about the nervous systems of the cat and dog being the
control systems that control their perceptions by acting through their
respective environments, which happen to include their respective bodies?
For the dog, the intended perceived distance to the cat is disturbed by the
cat's nervous system moving the outer surface of the cat's body; the reverse
is true for the cat. Neither of them sees the source of the disturbance.

Unrelated perceptions, such as the passage through terrain could provide
perceptions about what the cat is doing since they are not controlled
perceptions for the dog. Thus, this dog might later tell its' friends
all about the 'great chase' that the cat provided but the tale would be
generated completely from perceptions that were not being controlled with
the single exception that our intelligent dog would be able to state with
supreme confidence that the distance between the two of them remained
constant for the entire event.

OK. (Smart dog, by the way :wink:

I believe that people, myself included, probably conclude that we can
perceive disturbances (or their effects) from precisely this sort of
additional uncontrolled perception.

I think I agree with you that we can conclude such things, but when we do
(conclude we have identified the _source_ of a disturbance), we are often
wrong. Besides, we usually don't need to do such things -- to keep my
balance when I walk along the seawall in Galveston during a strong shore
breeze, do I really need to know about the positions of the jet streams,
the tilt of the earth relative to the sun, the relative movements of warm
and cold air masses, and the like?

Hummm.... how about a Post Script:

"Effects of a disturbance would include actions other than those that
effect the variable in the perceptual control loop, yes?

In other words, in addition to pushing my car around, the wind might also be
blowing down billboards alongside the road? Sure. No problem. Just as
my own actions affect a multitude of variables other than the ones(s) I
have chosen to perceive in a particular state. But there can be multiple
simultaneous sources of disturbance; disentangling them would often be
near impossible. Besides, doing that usually isn't a requirement for
perceptual control.

Thus, the effects of a disturbance can possibly be perceived when such
multiple effects exist and at least one is influencing an uncontrolled
perception AND the behaviour do not incidentally "control" the
uncontrolled perception. That is the multiple effects are at least no
more than somewhat loosely coupled to the controlled perception.

Help. You are losing me here. How about an example or two?

Later,

Tom

8

<[Bill Leach 940721.22:04 EST(EDT)]

From Tom Bourbon [940721.1457]

Me:
Again, I am having real trouble with this one. As I understand it (and
of course that can very well be wrong), a perception that is controlled
can not have uncontrolled perceptions as a part of THAT control loop.
If it did, any change in the uncontrolled perception would result in a
maximum error for the loop.

Tom:
Try thinking about tracking. Perceptions of cursor and target are just
that.

Me:
Ok, I suppose that I see the relationship and understand the meaning of
the term "uncontrolled perceptions". What still bothers me a little is
that in terms of the control loop, it seems that the perception of the
absolute cursor position is incidental to the control action.

Tom:
That's right. _If_ the person intends to control the position of the
cursor relative to that of the moving target, _then_ the person ignores
the absolute position of the disturbed cursor. Given the person's
intended result, there is no way for her or him to also control the
absolute position of the cursor, which really means to hold its position
constant, relative to the screen coordinates. ...

I don't think that this was the issue... exactly. I accept Mary's
example with the cake as a perfect (unambigious) example or how an
uncontrolled perception can be relevent to control.

Using what I now think I understand:

In the tracking experiment, the absolute "target" (my mistake in my
previous postings) position is perceived but is an uncontrolled
perception. The difference position is the controlled perception. This
example differs from Mary's in one significant way:

In the case of the tracking experiment, while the subject might perceive
the absolute position of the target, such a perception is not necessary
to the control action, only the difference (which is perceived). In the
cake example, the "disturbance" (low atmospheric pressure) is not
directly perceived at all but is a perception because it is known (now,
after many failures by many people to perceive cakes like they had
perceived at one time when on the coast). However, successful control
requires control system response to the perceived value of the
uncontrolled perception as opposed to response to the perceived
difference as in the tracking experiement even though the uncontrolled
perception is clearly the cause of any error in position between the
target and the cursor (not produced by the control system itself).

Tom:
Bill Powers keeps reminding people that in PCT modeling we must
maintain a distinction between a _disturbance_, defined as an
independent source of influence on variables perceived by a control
system, and the _effects_ the disturbance has on the perceived
variables.

Me:
I must be missing something in this one. If the disturbance to a
variable is related to that variable through a known function then
observation of the effects of the disturbance on a perception relates
back to the disturbance itself.

I think that I am beginning to understand where the difficulties are in
this one:

A disturbance affects a controlled variable. If the control system
successfully counteracts the effect then there is NO information
available concerning the disturbance contained in THAT perception.

Now it is true that we might perceive something about a disturbance but
such a perception will be due to input from uncontrolled perceptions of
another variable that is also affected by the disturbance (and not
incidentally fully counteracted by the original control loop).

Thus, my example of the smart dog telling his friends about the chase.
His uncontrolled perceptions would include such things as absolute
postions, velocity and the like SO it is possible for him to infer things
about the disturbances introduced by the cat's control system to the
common variable using those OTHER perceptions. If his ONLY perception
was the relative position between himself and the cat and he successfully
maintained that relationship at its reference value then he would be
completely unable to describe the chase.

Tom:

No problem, so long as you realize that you are talking from the
position of a modeler who knows all of the functions and can see the
tables of disturbance values ...

So you see, I was not trying to talk from the position of an "all
knowing" control system designer.

Me:
If the dog's reference was to maintain a specific distance (greater than
zero) then the dog would not perceive the effects of the disturbance
caused by the cat except to the extent that the control action failed.
That is, if the dog did indeed maintain this reference distance then it
would not know what the cat was doing to disturbe the perception.

Tom:
I don't follow you here. If the dog has _no_ reference for "distance to
cat," then nothing the cat does (no relative position of the cat) can
disturb the dog's perception of distance. On the other hand, if the dog
has a reference for a specific non-zero distance, then any deviation
from that distance is error.

Does this now make sense? If the dog is attempting control of the
relative postions but is not able to achieve the control then the dog IS
able to perceive something about the disturbance to the extent that
control fails (saying it the other way around). Such a perception may
not be either particularly accurate but it would exist (or necessarily
useful).

Tom:

By the way, could you make any sense of what I said earlier (Tom Bourbon
[940720.1653]) about the nervous systems of the cat and dog being the
control systems that control their perceptions by acting through their
respective environments, which happen to include their respective ...

Yes, but it is an interesting shift of boundry as the output function of
the cat clearly includes it's own body. I don't particularly care one
way or the other as long as there is a common understanding in such
discussions. If the boundry excludes muscles, claws, etc. the a control
system failure (being caught in this case) might shift a part of the
source of the failure to the "environment" as opposed to physical
limitations of the output function. Is there a difference? No, not
really but PCT can be confusing enough without making such a distinction
(an unnecessary one at that).

Me:
Hummm.... how about a Post Script: "Effects of a disturbance would
include actions other than those that affect the variable in the
perceptual control loop, yes?

Tom:

In other words, in addition to pushing my car around, the wind might
also be blowing down billboards alongside the road?

This makes me realize also that my line of query and statement has been a
bit missleading in other ways. Talking about perceiving a disturbance
(though failure to control or through uncontrolled perceptions) is just
a matter of dealing with the fact that people are so often certain that
they can do so. I am not trying to claim that such perceptions are
necessary for control and indeed can certainly see where the opposite
could often be true.

I also accept that such perceptions concerning the disturbance can also
be quite incorrect AND the reason that they can be so incorrect is that
it is not possible to perceive a disturbance to a variable, that is
controlled, directly. Thus, when we think that we "know" what the
disturbance is all about, we are kidding ourselves BECAUSE we must be
looking at variables that are not controlled and are then "possibly"
assuming that the disturbance is the only thing that is effecting them
AND that this (these) uncontrolled perceptions really are related in an
unambigious manner to the disturbance to the perception that we ARE
controlling.

Me:
Thus, the effects of a disturbance can possibly be perceived when such
multiple effects exist and at least one is influencing an uncontrolled
perception AND the behaviour do not incidentally "control" the
uncontrolled perception. That is the multiple effects are at least no
more than somewhat loosely coupled to the controlled perception.

Tom:
Help. You are losing me here. How about an example or two?

Do you still want an example or is the whole postion that I was trying to
take now clear? I realize that I am at least as guilty as anyone else on
the net of not expressing myself clearly enough to be understood.

-bill

9

<[Bill Leach 940721.19:57 EST(EDT)]

[Martin Taylor 940721 11:00]

It covers the two modes I called "search" and "explore."

I believe that may a problem is in assigning names to this appearent
behaviour. In PCT there is no control system difference represented
except that the references are different. In one the reference(s) are to
"search" and in the other "explore". I can't imagine that PCT has any
trouble with this but I can see that considering them to be special cases
would be a practical problem for the theory.

As with any control system, the output that leads to the action of
moving the gaze direction can be initiated because of external
disturbance or internal change of reference value. Either can affect the
error signal in the control system that perceives the gaze direction.
An alerting sound from a particular direction could be such a
disturbance.

I have not seen a "seal of approval" for this yet but as long as you are
implying the existance of a reference to be associated with the
disturbance that you explicitely mention, I am inclined to say that the
preceeding paragraph is "excellent" PCT.

The phenomenon would become really interesting if it challenged the
theory. But it does not.

No, I don't think it does. I never did think it did. As I have said
before, I think of it as one of those many places where PCT makes
beautifully clear something that was separately described, unexplained,
and somewhat vague beforehand. What has puzzled me is why this fact has
causes so much disturbance to the PCT old hands.

You might well have just identified the "problem" that you have
experienced in your many presentations here on CSG-L (one shared by many
of the rest of us BTW).

The phenomenon you describe is "not interesting to PCT". However, you
are interested in the phenomenon and have found that PCT explains the
phenomenon and of course find THAT interesting.

Now then Martin, if YOU can conceive of a way to apply the TEST to
alerting phenomenon and obtain PCT relevent data, I would say that the
"interest" would rise sharply.

Also, I personally take Bill's comment above in exactly that light. The
discussion of alerting has been quite interesting to me (and probably
many others) but it is not in an of itself "interesting to PCT".

-bill