Attention and activation

CSGnet Archive CSG2015
1

[From Rupert Young (2015.02.15 20.15)]

Anyone?

···

On 05/01/2015 22:29, Rupert Young wrote:

[From Rupert Young (2015.01.05 22.30)]

(Rick Marken (2014.12.10.1045) (PCT robots)

RM: The other is what the beacon following robot does when there is no beacon present? Does it just go towards the brightest part of the room? Or does it stop? I'm interested because I've never really understood how a control system is taken "off line" and it looks like the beacon follower might only do its following when there is a beacon around.

I think this is an important question that we need to deal with, of how do control systems become active or inactive, and one I have been mulling over for a while. The beacon follower, with regard to distance, only works properly if the reference is zero. Then, when the perceptual signal is not present (it is zero) the output will also be zero. If, however, the reference is non-zero, 10 say, then when the perceptual signal is not present the error will be 10, and then robot will continually reverse, whereas the control system should deactivate. In this case there is no distinction between 0 as a signal value and 0 as a signal absence (null). So, there seems to be a need for the control system to become inactive. Technically this is not a great problem, but I am interested in how this would be viewed in PCT.

Traditionally this has been seen as the issue of attention, I think (actually Visual Attention was my initial topic on my PhD, before I discovered PCT). What is it that brings a system into attention? Does attention equate to a control system being active? What has happened to a previously active system? Is there a difference between a system which has achieved its goal and an inactive system (both of which have no output)? How does an inactive system become active (attention shift)? Can you have an active system when the perceptual input is null? Does the onset of a perceptual signal activate a control system? Does the onset of a reference signal activate a control system? Does the onset of a perceptual signal activate a reference signal? I will try to address some of these questions as I go on.

The attached shows a standard discussion of attention (though does include a vague inkling of PCT, see match condition on p68) but raises some interesting points on which we can expand for a PCT analysis. (From The Cognitive Approach to Conscious Machines by Pentti O. Haikonen).

Haikonen defines two types of attention, sensory and inner, and two subdivisions, voluntary and involuntary. So what would these mean in PCT terms?

Sensory involuntary - a shift in attention due to the onset of a perceptual signal such as movement in peripheral vision, or the cocktail party effect. This suggests the pre-existence of relevant control systems. For the former it could be a low-level, continually active system with the goal of equalising retinal motion, resulting in foveating the "stimulus". The latter is slightly different as it seems to require high-level context (memory) which provides significance. In this case then, is there a continually active system monitoring for the presence of one's name? This would imply that there would need to be active systems for everything of significance to a person. Or could the input itself in some way trigger activation of a control system, which results in attention?

Sensory voluntary - requires a goal for attending to a particular perception. For example, the thirsty man entering a room where there are a number of objects would want to attend to a bottle of water. So, he would have an existing goal for achieving water. But would he have to attend to each object in turn or would the water stand out from the scene due to the active water-desiring control system? That is, would the control system for a pie, say, be inactive as he is not hungry?

Inner involuntary - a shift in attention due to the onset of a reference signal. That is, activation of a control system due to top-down context rather than changes in perceptual input. Consider the situation of a dog with a juicy bone, but nearby is a man with stick. The man could be a threat with a stick to beat the dog or a friend throwing a stick to fetch. So the dog could either stay eating the bone or, depending upon context, run away or run after the stick. In both cases the perceptual input is the same, but the context depends upon whether or not the dog is afraid of the stick man; whether the run away system or the run after stick system is activated.

I'm not sure how useful these classifications are considering all involve references and perceptions (inner and sensory), thus "Sensory voluntary" seems to be the same as "Inner voluntary". The voluntary and involuntary division seems to indicate whether what you are currently controlling is interrupted or control passes to another system, perhaps in a sequence.

So, attention seems to describe that one system (at a particular level) is active to the exclusion of others (at the same level). And shift in attention describes that a system which was previously inactive becomes active, and the system that was active becomes inactive. And that shift (and activation) can be triggered by either a sensory onset or a reference onset.

Perhaps, then for a system to be active it is necessary for both the reference and perceptual signals to be present. We could then describe a number of different states of a control system, in terms of activation. In this context I mean inactive to indicate no output.

Inactive (dead) - Neither reference or perceptual signals are present.

Inactive (dormant) - Perceptual signal present, but not reference. When a reference signal becomes present this becomes active and the perceptual signal comes under control.

Inactive (monitoring) - Reference signal present, but not perceptual. This monitors for a certain perception but does nothing until that perceptual signal becomes present and then springs into action, e.g. when the beacon is turned on or there is a loud bang.

Active (controlling) - a system that is continuously controlling its perception according to its reference, and protects against disturbances.

Active (finished) - a system where the perception had been brought into line with the reference, but where there will not be any further change to perception or reference. In this case there will be no output, or no change in output.

So, there are a couple of questions here. One is how are inactive systems physically realised, when only one of either reference or perception is present?

And, two, what happens to a system that was active when it is interrupted? For example, if you are in a bar playing pool and you stop when a meaningful news item comes on the TV, what happened to your pool playing system? Is the reference switched off or suppressed? How does it switch back on when the news item is over?

Regarding one of the questions from the earlier list, "Does the onset of a perceptual signal activate a reference signal?" I can envisage how this might occur with a two-level system. Suppose they are both inactive, the top one monitoring and the lower one dead. Then a perception occurs which feeds up to the top system, which then becomes active producing output which sets the reference for the lower system which also becomes active.

Hmm, some rather muddled thoughts, so I'd welcome comments from others.

--
Regards,
Rupert

2

[From Rick Marken (2015.02.15.1710)]

···

Rupert Young (2015.02.15 20.15)–

RY: Anyone?

RM: This is a great topic and I will try to come up with some ideas, if I can. I’ve just scanned over you’re suggestions but they look interesting.

RM: I’ll just quickly say, given the present discussions, that even though your robotics research doesn’t involve the TCV it is certainly very worthwhile because it answers questions about how we might model various aspects of control once we get to modeling real behavior. This topic is a particularly good example because people do seem to stop controlling perceptions (or do they?) and figuring out how to implement these kinds of things in a robot will help us understand how to do it when we want to try to model that kind of behavior in real living systems.

Best

Rick

On 05/01/2015 22:29, Rupert Young wrote:

–
**Regards,
Rupert
**

[From Rupert Young (2015.01.05 22.30)]

  (Rick Marken (2014.12.10.1045) (PCT robots)



  RM: The other is what the beacon following robot does when there

is no beacon present? Does it just go towards the brightest part
of the room? Or does it stop? I’m interested because I’ve never
really understood how a control system is taken “off line” and it
looks like the beacon follower might only do its following when
there is a beacon around.

  I think this is an important question that we need to deal with,

of how do control systems become active or inactive, and one I
have been mulling over for a while. The beacon follower, with
regard to distance, only works properly if the reference is zero.
Then, when the perceptual signal is not present (it is zero) the
output will also be zero. If, however, the reference is non-zero,
10 say, then when the perceptual signal is not present the error
will be 10, and then robot will continually reverse, whereas the
control system should deactivate. In this case there is no
distinction between 0 as a signal value and 0 as a signal absence
(null). So, there seems to be a need for the control system to
become inactive. Technically this is not a great problem, but I am
interested in how this would be viewed in PCT.

  Traditionally this has been seen as the issue of attention, I

think (actually Visual Attention was my initial topic on my PhD,
before I discovered PCT). What is it that brings a system into
attention? Does attention equate to a control system being active?
What has happened to a previously active system? Is there a
difference between a system which has achieved its goal and an
inactive system (both of which have no output)? How does an
inactive system become active (attention shift)? Can you have an
active system when the perceptual input is null? Does the onset of
a perceptual signal activate a control system? Does the onset of a
reference signal activate a control system? Does the onset of a
perceptual signal activate a reference signal? I will try to
address some of these questions as I go on.

  The attached shows a standard discussion of attention (though does

include a vague inkling of PCT, see match condition on p68) but
raises some interesting points on which we can expand for a PCT
analysis. (From The Cognitive Approach to Conscious Machines by
Pentti O. Haikonen).

  Haikonen defines two types of attention, sensory and inner, and

two subdivisions, voluntary and involuntary. So what would these
mean in PCT terms?

  Sensory involuntary - a shift in attention due to the onset of a

perceptual signal such as movement in peripheral vision, or the
cocktail party effect. This suggests the pre-existence of relevant
control systems. For the former it could be a low-level,
continually active system with the goal of equalising retinal
motion, resulting in foveating the “stimulus”. The latter is
slightly different as it seems to require high-level context
(memory) which provides significance. In this case then, is there
a continually active system monitoring for the presence of one’s
name? This would imply that there would need to be active systems
for everything of significance to a person. Or could the input
itself in some way trigger activation of a control system, which
results in attention?

  Sensory voluntary - requires a goal for attending to a particular

perception. For example, the thirsty man entering a room where
there are a number of objects would want to attend to a bottle of
water. So, he would have an existing goal for achieving water. But
would he have to attend to each object in turn or would the water
stand out from the scene due to the active water-desiring control
system? That is, would the control system for a pie, say, be
inactive as he is not hungry?

  Inner involuntary - a shift in attention due to the onset of a

reference signal. That is, activation of a control system due to
top-down context rather than changes in perceptual input.
Consider the situation of a dog with a juicy bone, but nearby is a
man with stick. The man could be a threat with a stick to beat the
dog or a friend throwing a stick to fetch. So the dog could either
stay eating the bone or, depending upon context, run away or run
after the stick. In both cases the perceptual input is the same,
but the context depends upon whether or not the dog is afraid of
the stick man; whether the run away system or the run after stick
system is activated.

  I'm not sure how useful these classifications are considering all

involve references and perceptions (inner and sensory), thus
“Sensory voluntary” seems to be the same as “Inner voluntary”. The
voluntary and involuntary division seems to indicate whether what
you are currently controlling is interrupted or control passes to
another system, perhaps in a sequence.

  So, attention seems to describe that one system (at a particular

level) is active to the exclusion of others (at the same level).
And shift in attention describes that a system which was
previously inactive becomes active, and the system that was active
becomes inactive. And that shift (and activation) can be triggered
by either a sensory onset or a reference onset.

  Perhaps, then for a system to be active it is necessary for both

the reference and perceptual signals to be present. We could then
describe a number of different states of a control system, in
terms of activation. In this context I mean inactive to indicate
no output.

  Inactive (dead) - Neither reference or perceptual signals are

present.

  Inactive (dormant) - Perceptual signal present, but not reference.

When a reference signal becomes present this becomes active and
the perceptual signal comes under control.

  Inactive (monitoring) - Reference signal present, but not

perceptual. This monitors for a certain perception but does
nothing until that perceptual signal becomes present and then
springs into action, e.g. when the beacon is turned on or there is
a loud bang.

  Active (controlling) - a system that is continuously controlling

its perception according to its reference, and protects against
disturbances.

  Active (finished) - a system where the perception had been brought

into line with the reference, but where there will not be any
further change to perception or reference. In this case there will
be no output, or no change in output.

  So, there are a couple of questions here. One is how are inactive

systems physically realised, when only one of either reference or
perception is present?

  And, two, what happens to a system that was active when it is

interrupted? For example, if you are in a bar playing pool and you
stop when a meaningful news item comes on the TV, what happened to
your pool playing system? Is the reference switched off or
suppressed? How does it switch back on when the news item is over?

  Regarding one of the questions from the earlier list, "Does the

onset of a perceptual signal activate a reference signal?" I can
envisage how this might occur with a two-level system. Suppose
they are both inactive, the top one monitoring and the lower one
dead. Then a perception occurs which feeds up to the top system,
which then becomes active producing output which sets the
reference for the lower system which also becomes active.

  Hmm, some rather muddled thoughts, so I'd welcome comments from

others.

–

Richard S. Marken, Ph.D.
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

3

[From Rick Marken (2015.02.24.0930)]

···

Rupert Young (2015.02.15 20.15)

  RM: The other is what the beacon following robot does when there

is no beacon present? Does it just go towards the brightest part
of the room? Or does it stop? I’m interested because I’ve never
really understood how a control system is taken “off line” and it
looks like the beacon follower might only do its following when
there is a beacon around.

  RY: I think this is an important question that we need to deal with,

of how do control systems become active or inactive, and one I
have been mulling over for a while. The beacon follower, with
regard to distance, only works properly if the reference is zero.
Then, when the perceptual signal is not present (it is zero) the
output will also be zero. If, however, the reference is non-zero,
10 say, then when the perceptual signal is not present the error
will be 10, and then robot will continually reverse, whereas the
control system should deactivate. In this case there is no
distinction between 0 as a signal value and 0 as a signal absence
(null). So, there seems to be a need for the control system to
become inactive. Technically this is not a great problem, but I am
interested in how this would be viewed in PCT.

RM: Well, I haven’t been able to think of anything useful on this. But I was thinking that maybe control systems never become inactive in the sense that something outside the system would turn a system off or on. That would seem to require an awful lot of capability in terms of the switching system, which would have to know the basis on which each of the thousands of control systems that make up a living organism should be switched in or out.

RM: My guess is that all our control systems are always “on” and seem to go off only when the variables they control are at levels that don’t require any action (as is the case with your beacon follower). I have a feeling this may have to do with the fact that neural signals are always positive so that you never get a situation like the one you describe above where you get a negative error signal when the perceptual signal is below the reference. Perhaps putting an offset into the comparator that keeps the error positive until the deviation between perception and reference exceeds the offset at which point the error becomes truly 0 and can’t go negative. I haven’t tried to simulate this yet but it if it works it would be a solution to the problem, maybe. I’ll try it as soon as I get a chance.

Best

Rick

Richard S. Marken, Ph.D.
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble

4

[From Rupert Young (2015.02.25 19.30)]

(Rick Marken (2015.02.24.0930))

RM: Well, I haven't been able to think of anything useful on this. But I was thinking that maybe control systems never become inactive in the sense that something outside the system would turn a system off or on. That would seem to require an awful lot of capability in terms of the switching system, which would have to know the basis on which each of the thousands of control systems that make up a living organism should be switched in or out.

RY: I don't think that such a capability is necessary. I think that all that is required is that a reference signal is absent (is zero) so that there is no output. As there is no output then the references for lower systems will also be zero, in effect switching them off. This morning I went swimming so all my associated lower control systems were active and I thrashed my arms around and kicked my legs for a while. In the afternoon I was sitting in front of a computer and my arm thrashing and leg kicking systems were not "active", I would say because their references were zero, because my higher level swimming reference was also zero.

As you say neural signals are always positive so the output of a comparator will always be zero if the reference is zero even if the perception isn't (an active inhibitory signal); in brains r - p = 0, if r = 0 and p > 0.

Maybe this is what you are saying?

There is then the question of how a system that wasn't active becomes active (the reference becomes > 0) when you see motion out of the corner of your eye, or with the cocktail party effect. It seems that previously "inactive" controls systems can be triggered into activity by the onset of a reference signal or by a perceptual signal.

Regards,
Rupert

5

[Martin Taylor 2015.02.25.13.47]

[From Rupert Young (2015.02.25 19.30)]

(Rick Marken (2015.02.24.0930))

RM: Well, I haven't been able to think of anything useful on this. But I was thinking that maybe control systems never become inactive in the sense that something outside the system would turn a system off or on. That would seem to require an awful lot of capability in terms of the switching system, which would have to know the basis on which each of the thousands of control systems that make up a living organism should be switched in or out.

RY: I don't think that such a capability is necessary. I think that all that is required is that a reference signal is absent (is zero) so that there is no output.

[MT] Rupert, having a reference value of zero just means that if the perceptual value is zero, so is the error signal. It's not at all the same as the control unit being switched off. For the control unit to be switched of by turning someting to zero, that something must be the output gain, and even then the lower-level reference values to which it contributes would still see a zero-value reference contribution, not an absence of contribution.

[MT] So yes, we do need a way to deal with the problem of activation of control units, and as yet there is no concensus as to how that might be done. As Rick points out, for that to happen independently for every control unit in a living hierarchy seems energetically wasteful. You need that capability for the actual control those systems do when they are active and varying continuously under the influence of each other's variations. An efficient system would not need it for a binary change of sate that happens infrequently on the time scale of actual control.

There is then the question of how a system that wasn't active becomes active (the reference becomes > 0) when you see motion out of the corner of your eye, or with the cocktail party effect. It seems that previously "inactive" controls systems can be triggered into activity by the onset of a reference signal or by a perceptual signal.

[MT] Yes. Early in my involvement with PCT I assumed the existence of a non-control "Alerting" system that would play this triggering role (it didn't actually come out of PCT; it came out of a now 40-year-old analysis of the perceptual requirements for using the masses of data then and now coming back from space). I assumed that an "Alerting" unit would have a perceptual function like those for control units (and might even be the PIF of an inactive control unit). When that perceptual function produced a supra-threshold output, it would send the trigger signal somewhere to exchange the activation state of some control units, keeping the number of actually active units within the available degrees of freedom for the environmental feedback path availability. Because the Alerting signals were not controlled, there would be no limit on how many could be functioning at any one moment. It wasn't fully worked out, and I haven't done anything about it since about 1992, but I think something of the kind is needed.

Martin

6

[From Rupert Young (2015.02.27 20.30)]

(Martin Taylor 2015.02.25.13.47)

RY: I don't think that such a capability is necessary. I think that all that is required is that a reference signal is absent (is zero) so that there is no output.

[MT] Rupert, having a reference value of zero just means that if the perceptual value is zero, so is the error signal. It's not at all the same as the control unit being switched off. For the control unit to be switched of by turning someting to zero, that something must be the output gain, and even then the lower-level reference values to which it contributes would still see a zero-value reference contribution, not an absence of contribution.

RY: If it is zero (the neuron has no activation) then the inhibitory signal from the perceptual function will have no effect on the comparator, as neurons can't be negative. So the error will always be zero no matter what the value of the perception. A similar effect could be achieved by zero gain, but that seems to require capabilities which are not the normal part of the operation of control systems.

There is then the question of how a system that wasn't active becomes active (the reference becomes > 0) when you see motion out of the corner of your eye, or with the cocktail party effect. It seems that previously "inactive" controls systems can be triggered into activity by the onset of a reference signal or by a perceptual signal.

[MT] Yes. Early in my involvement with PCT I assumed the existence of a non-control "Alerting" system that would play this triggering role (it didn't actually come out of PCT; it came out of a now 40-year-old analysis of the perceptual requirements for using the masses of data then and now coming back from space). I assumed that an "Alerting" unit would have a perceptual function like those for control units (and might even be the PIF of an inactive control unit). When that perceptual function produced a supra-threshold output, it would send the trigger signal somewhere to exchange the activation state of some control units, keeping the number of actually active units within the available degrees of freedom for the environmental feedback path availability. Because the Alerting signals were not controlled, there would be no limit on how many could be functioning at any one moment. It wasn't fully worked out, and I haven't done anything about it since about 1992, but I think something of the kind is needed.

RY: Perhaps it could be done with a two-level system where the higher system has a present reference but absent perception and the lower system has an absent reference but present perception. And both have no error, because they are not active. For a system to be active both signals must be present, in my definition. When the higher system perception becomes present then so does the error and the output, which then produces a reference for the lower system which starts controlling. The higher system could be thought of continually monitoring for a perception (to be present). So, with the cocktail party effect the higher system is continually monitoring for the mention of one's name which triggers an output, and reference for the lower level orientation system.

Regards,
Rupert

7

[Martin Taylor 2015.02.28.10.02]

[From Rupert Young (2015.02.27 20.30)]

(Martin Taylor 2015.02.25.13.47)

RY: I don't think that such a capability is necessary. I think that all that is required is that a reference signal is absent (is zero) so that there is no output.

[MT] Rupert, having a reference value of zero just means that if the perceptual value is zero, so is the error signal. It's not at all the same as the control unit being switched off. For the control unit to be switched of by turning someting to zero, that something must be the output gain, and even then the lower-level reference values to which it contributes would still see a zero-value reference contribution, not an absence of contribution.

RY: If it is zero (the neuron has no activation) then the inhibitory signal from the perceptual function will have no effect on the comparator, as neurons can't be negative. So the error will always be zero no matter what the value of the perception. A similar effect could be achieved by zero gain, but that seems to require capabilities which are not the normal part of the operation of control systems.

So you are saying that a control system could be designed to control for perceiving A to be 3 units more than B, but could not control for A to be any particular amount less than B? Or to control for A to be equal to B?

Your point about neurons not having a negative firing rate has been clear since the very beginning of PCT. There are several ways that real physical (biological or engineered) systems can get around this, but the one usually mentioned is that two half-systems operate in parallel, one in which the error is r-p and the error signal is an excitatory input to the output function, the other in which the error signal is p-r and is an inhibitory input to the output function. Another way to deal with it is to incorporate a resting bias, so that "zero" error is represented by a particular firing rate.

Remember also that in the PCT-standard control unit, the output function includes an integrator, which means that zero error signal does not mean zero output. It means zero change in output (apart from the effect of the leak that also is usually included).

Anyway, the control system must be able to control for a perception equal to a zero reference value, and not turn off if the reference value is zero.

As for zero gain not being a normal part of the operation of control systems -- how do you know that? Granted, gain controls are not drawn in the standard diagram of the hierarchy, but that doesn't mean they are not there. Nor does it mean Bill didn't consider them, as you may see if you go back a long way in the archives. He just left it to others such as (I think) Tom Bourbon to study. Bill's personality was one that liked to go one step at a time, building always on a firm substrate, and the demands of reducing the millions of degrees of freedom per second in the sensory input to the tens of degrees of freedom per second in the muscular output were not something he wanted to get into, as I found out very early in my acquaintance with PCT. That doesn't necessarily mean he thought it was wrong to consider such questions, but it does mean he didn't want to do it himself.

That's all rather by the way. My point is simply that a value of zero is not the same as the absence of a value. What shifting activation requires is the absence of output that might interfere with control by active systems, not the absence of perception or of reference value, or even of error. Those are important always, because something drastic might happen in a perception not currently being actively controlled that would require immediate action if the organism is to survive.

Martin

8

[From Rupert Young (2015.02.28 20.30)]

(Martin Taylor 2015.02.28.10.02]

RY: If it is zero (the neuron has no activation) then the inhibitory signal from the perceptual function will have no effect on the comparator, as neurons can't be negative. So the error will always be zero no matter what the value of the perception. A similar effect could be achieved by zero gain, but that seems to require capabilities which are not the normal part of the operation of control systems.

So you are saying that a control system could be designed to control for perceiving A to be 3 units more than B, but could not control for A to be any particular amount less than B? Or to control for A to be equal to B?

I am saying that in a natural (neural) control system there would be no output if the reference signal has no activation even though the perception signal was activated.

Your point about neurons not having a negative firing rate has been clear since the very beginning of PCT. There are several ways that real physical (biological or engineered) systems can get around this, but the one usually mentioned is that two half-systems operate in parallel, one in which the error is r-p and the error signal is an excitatory input to the output function, the other in which the error signal is p-r and is an inhibitory input to the output function.

And, I guess, you could have some systems which are only r-p and others which are only p-r. The latter actually would be the opposite of the arrangement I've been mentioning except that it would be the onset of the perceptual signal which switches it on. In other words you could have two types of systems one where the reference must be active and one where the perception must be active, for there to be output.

Remember also that in the PCT-standard control unit, the output function includes an integrator, which means that zero error signal does not mean zero output. It means zero change in output (apart from the effect of the leak that also is usually included).

Sure, but a unit could be modelled without an integrator, could it not?

Anyway, the control system must be able to control for a perception equal to a zero reference value, and not turn off if the reference value is zero.

It depends what you mean by zero. In real neural systems an inactive neuron has zero, err, activity and may not a goal signal. Whereas in the computer models we generally use zero is just another value on a particular scale, just as 0 Celcius is a value on the temperature scale.

As for zero gain not being a normal part of the operation of control systems -- how do you know that?

I don't know. But I don't think it is necessary to achieve the functionality that I am trying to get to. Mind you I don't recall coming across discussions of gain control, except as part of reorganisation. If you could point me towards them I 'd be interested to read them.

That's all rather by the way. My point is simply that a value of zero is not the same as the absence of a value. What shifting activation requires is the absence of output that might interfere with control by active systems, not the absence of perception or of reference value, or even of error. Those are important always, because something drastic might happen in a perception not currently being actively controlled that would require immediate action if the organism is to survive.

What I am saying is that the absence of output can be achieved by the absence of perception or of reference within control systems, but that these systems can be activated by the onset, once again, of those signals. And by absence I mean neurons which are not active.

Regards,
Rupert

9

[From Rupert Young (2015.03.01 16.00)]

I don't think your reply went to the list.

(Martin Taylor 2015.02.28.16.41)

I am saying that in a natural (neural) control system there would be no output if the reference signal has no activation even though the perception signal was activated.

Then how could the top level of the hierarchy produce any output? And if it did not, the reference inputs at the level below would have no input, so than how could they produce output? And then the level below that ...

They wouldn't, that's the point. One scenario I could envisage this is with the control of a sequence. For example, in the morning you may be controlling the sequence of having breakfast, taking a shower and going to work. While you are having breakfast your shower-taking control system reference is inactive, and without output. The same would apply to all the lower systems required for taking a shower. So, an entire sub-hierarchy below that higher system is switched off. Once breakfast is over the shower reference becomes active and all the necessary lower systems then spring into activity.

Some lower systems, of course, are common to different sub-hierarchies and can be activated from different higher systems. So, control systems for hands can be used for cooking bacon or for washing hair.

Regards,
Rupert

10

[Martin Taylor 2015.03.01.11.45]

[From Rupert Young (2015.03.01 16.00)]

I don't think your reply went to the list.

(Martin Taylor 2015.02.28.16.41)

I am saying that in a natural (neural) control system there would be no output if the reference signal has no activation even though the perception signal was activated.

Then how could the top level of the hierarchy produce any output? And if it did not, the reference inputs at the level below would have no input, so than how could they produce output? And then the level below that ...

They wouldn't, that's the point. One scenario I could envisage this is with the control of a sequence.

No, I was referring you to the top of the hierarchy, which by definition has no reference input. We typically follow Bill Powers, and take this as being exactly the same as having a reference input always equal to zero. My point was that if your equation of zero reference value with no output were correct, the ENTIRE hierarchy of control would never be able to act at all.

For example, in the morning you may be controlling the sequence of having breakfast, taking a shower and going to work. While you are having breakfast your shower-taking control system reference is inactive, and without output. The same would apply to all the lower systems required for taking a shower. So, an entire sub-hierarchy below that higher system is switched off. Once breakfast is over the shower reference becomes active and all the necessary lower systems then spring into activity.

Yes, all that sequence of control systems is switched on and off. The only question is the mechanism by which this happens. My point was and is that if you accept the hierarchy as Bill described it, prohibiting within-level loops in which control units at a given level contribute to each other's reference values, then setting a reference value of zero is not equivalent to switching the control system off.

The only reason, in an evolutionary sense, for having any kind of switch-off mechanism is that you can't change output a anywhere near the rate at which the totality of your controllable perceptions change, not by several orders of magnitude. You have to have a way of switching the output serving one control unit off while still being able to see whether the perception that unit controls is getting dangerously far from its reference value.

Example: You are a hunter-gatherer thousands of years ago, picking berries to eat. You are not actively controlling for being far from a hungry tiger. You see a flash of yellow through the trees. Suddenly you are actively controlling for being far from a hungry tiger, and not controlling for that nice juicy berry to be in your mouth. You must all along have been able to perceive that tiger, and to have a reference value for perceiving yourself to be far from it. But that control system was not active until you saw (or thought you saw, because it might have just been a trick of the light) the tiger.

Some lower systems, of course, are common to different sub-hierarchies and can be activated from different higher systems. So, control systems for hands can be used for cooking bacon or for washing hair.

Yes, indeed.

Martin

11

[From Rupert Young (2015.03.01 21.30)]

(Martin Taylor 2015.03.01.11.45]

[From Rupert Young (2015.03.01 16.00)]

They wouldn't, that's the point. One scenario I could envisage this is with the control of a sequence.

No, I was referring you to the top of the hierarchy, which by definition has no reference input. We typically follow Bill Powers, and take this as being exactly the same as having a reference input always equal to zero. My point was that if your equation of zero reference value with no output were correct, the ENTIRE hierarchy of control would never be able to act at all.

This could mean that there is no reference function in which case the output may be a function of just the perception so my point wouldn't affect this. Anyway there is much of PCT which is only described at a general, and somewhat vague, level. In reality the operation of control systems is likely to be much more varied and messy than the idealised systems usually under discussion. When trying to sort out the detail of how systems work I don't think we should restrict ourselves to fit in with areas of the theory which are only vaguely defined, as long as its still perceptual control.

Attention, as far as I can determine, is an area which is not well accounted for in PCT.

For example, in the morning you may be controlling the sequence of having breakfast, taking a shower and going to work. While you are having breakfast your shower-taking control system reference is inactive, and without output. The same would apply to all the lower systems required for taking a shower. So, an entire sub-hierarchy below that higher system is switched off. Once breakfast is over the shower reference becomes active and all the necessary lower systems then spring into activity.

Yes, all that sequence of control systems is switched on and off. The only question is the mechanism by which this happens. My point was and is that if you accept the hierarchy as Bill described it, prohibiting within-level loops in which control units at a given level contribute to each other's reference values, then setting a reference value of zero is not equivalent to switching the control system off.

I don't believe I mentioned "within-level loops". As mentioned already I think there is a difference between inactive references and zero value references.

You must all along have been able to perceive that tiger, and to have a reference value for perceiving yourself to be far from it. But that control system was not active until you saw (or thought you saw, because it might have just been a trick of the light) the tiger.

That sounds like a system that was continuously active. There was no error because the perception was aligned with the reference. The sight of the tiger disturbs the system resulting in error, and output action.

With the phenomenon of attention there seems to be cases where systems are switched off (inactive references), but PCT does not have a mechanism to explain it.

Rupert

12

[From Rick Marken (2015.03.01.1335)]

···

Rupert Young (2015.03.01 21.30)–

Martin Taylor (2015.03.01.11.45)–

RY: They wouldn’t, that’s the point. One scenario I could envisage this is with the control of a sequence.

MT: No, I was referring you to the top of the hierarchy, which by definition has no reference input. We typically follow Bill Powers, and take this as being exactly the same as having a reference input always equal to zero. My point was that if your equation of zero reference value with no output were correct, the ENTIRE hierarchy of control would never be able to act at all.

RY: This could mean that there is no reference function in which case the output may be a function of just the perception so my point wouldn’t affect this. Anyway there is much of PCT which is only described at a general, and somewhat vague, level.

RM: This is a very interesting topic to me and I agree that there is a lot in PCT that is vaguely described and this whole topic of bringing control systems in and out of use is definitely one of them. I’m still not completely convinced that bringing systems into and out of use is really a necessary capability. And I think this is a place where the robotics work could really make a great contribution. So I wonder if we could start back at the beginning because I don’t think I really understood the beam follower robot. First what is the beam follower controlling for when it follows the beam? And why does it stop controlling when the beam goes away?

Best

Rick

–
Richard S. Marken, Ph.D.
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble