understanding the hierarchy

CSGnet Archive CSG1999
1

[From Bob Christensen (991016.1702 PT)]

I've been thinking about the hierarchy for some time, observing my private
experiences closely to see how they fit with my understanding of the
hierarchy, and trying to revise my understanding as I go along. Some
experiences seem to fit well with the hierarchy, some don't seem to at first
but then I see how they fit, and some I'm still not clear on. I have also
tried to observe my experiences and come up with my own best model to see if
I rederive the hierarchical structure or if I induce a somewhat different
model. So here is my current understanding. I'm not sure how much of it is
consistent with Bill's hierarchical model, how much of it is just a different
emphasis or way of explaining, and how much of it is different. I think that
most of it is allowed within the parameters of Bill's model but the emphasis
I think is somewhat different.

I think that higher level perceptions are composed of or transformed from
lower level perceptions is the most clear part of the model to me. I don't
recall any disconfirming instances here, and it seems quite sound. What else
would higher level perceptions be made of?

Then there is the part of higher level references categorically setting the
references for lower level references in order to match their perceptions.
Now it has not been easy for me to see just what this implies, because it
might not fit with the way I and probably others categorize intentions with
our natural language notions. It seems to me that CSGnet members often talk
as if everything we do is "in order" to satisfy our highest level references.
If this is what the model implies then this part of the model is quite
disagreeable to me. For instance, If I feel hungry, I think of myself as
engaging in programs and sequences "in order" to reduce this low level
mismatch. I think of hunger pangs as being at the intensity, sensation, and
transition levels. The exact level is beside the point. The point is that
it is a quite low level mismatch. Another example is when your on a road
trip and you experience this pain in you lower abdomen telling you that you
have to go to the bathroom, and you start engaging in programs to see if you
should get off at the next exit (do I see a gas station? ...). As another
example to make clear that this does not just apply to internal body
perceptions, lets say I have a painful splinter stuck in my finger and I have
to engage in programs in order to figure out how to get it out. In these
examples it seems that the "in order to" goes in the reverse direction in the
hierarchy. I have this low level mismatch, I need an event to happen (the
splinter coming out) in order to make the pain (error) go away, I need to
find a sequence in order to have the event happen, I need to find a program
in order to have the sequence occur, hopefully I have good principles so that
I can find a workable program. So how does the hierarchy account for this
apparent reversal?

This is my best explanation. First, lower level error signals or equivalents
are sent up the hierarchy and are combined with other (perceptual) signals to
form higher level perceptions. The way the CS unit is drawn we do not think
of error signals as going upward because they are on the output side. Now
once they go upward, whether you call them error signals or perceptual
signals is definitional. They are error signals because they vary in the
same way an error signal that could correct the mismatch by going down to
outputs does. They are perceptions because they go up to form higher level
perceptions which are on the input side of higher level control systems. Now
whether, these error/perceptual signals are taken off the same signal that
goes to the outputs or there is a separate perceptual function is not
important. In the case of pain, we typically have references for zero pain,
and since there are no negative currents in the nervous system, the pain
perceptual signal can just be fed up to higher levels because it will be
equal in magnitude to zero minus itself. At higher levels, for instance the
event level, a comparison of current state with the desired event may occur
in a separate perceptual function. But why not utilize the comparison
function to double as a perceptual function as well? In any case, the main
point is that lower level mismatches are available to higher level perceptual
functions. I don't think that anything that I have said so far in
inconsistent with what is allowed by Bill's hierarchical model, but maybe
there is some additional explained explanation.

Now the error/perceptual signals get combined and transformed with other
perceptions in order to form higher level perceptions that pertain to my
understanding of the pain signals. For instance, the above pains have
locations in my stomach, lower abdomen, and finger respectively. They are
temporally related to not having eaten, having drank a lot and not having
gone to the bathroom in a while, and having just slid my hand along that
wooden rail. In the case of the splinter, I can see a dark speck when I look
at my finger. At my age I have many memories of previous similar situations
which contribute via imagination to my models of what is going on when I have
one of these pains. Thus, the pain signals and other current perceptions
evoke memories of similar situations which get combined with the current
perceptions to form my model of the current situation. With the model in
imagination I perceive a higher level mismatch which I have learned which
also corrects the pain signals. I find myself imagining eating from a plate
in front of me, standing in front of a urinal, or pulling out the splinter.
Not in one of these current states, I engage in programs to get me there.
Again, I do not think that anything that I have said is inconsistent with the
current model. The capacity to make cross level relationship perceptions is
needed, but this is already allowed. (I think this cross level capacity is
related to the observer sitting to the side of the hierarchy, but that is an
aside for the purpose of this post.) At no point did I say that lower level
references set higher level references. But I do think that higher level
references are often in service of (in order to satisfy) lower level
references. This has clinical implications because when you ask a person why
they do or think something, you may be going up or down in the levels of
perceptions. It also has implications for changing systems level concepts.
System level concepts will change when a person perceives that they cease to
give a differential advantage in serving lower level purposes such as
affiliation, recognition, status, income, tenure, or eternal life.

Now I want to follow up more on the statement that higher level references
are often in service of lower level references. Now I suppose that a
reference at any level can serve more than one purpose, and these purposes
can probably correspond to references at both higher and lower levels. But I
am talking more or less here. The standard answer in PCT about what all
references ultimately serve is to reduce intrinsic error. Many intrinsic
error signals are at the lowest levels, and I think that these intrinsic
error signals or signals that vary in the same way are available to the
hierarchy at the lowest levels (I think Bill has proposed this before). Such
signals can be experienced as discomfort/pain and when decreased as
pleasure/relief/comfort. Since they can be experienced, I think they must
also be able to be perceived and remembered. Hunger sensations covary in
intensity with intrinsic variables relating to blood levels of glucose and
other supplies. The feeling you get holding your breath under water for some
time covaries with oxygen level. When your body temperature drops you feel
cold. Since these signals are available at low levels of the hierarchy, and
since they are analogues of intrinsic errors, then much of the hierarchies
higher level references can be seen as serving to change these lower level
signals available to the hierarchy in the desired direction.

Since lower level perceptions and references evolved and developed first,
higher level perceptions and references probably often become organized out
of a need to better reduce errors at lower levels. Higher level perceptions
provide more comprehensive models of the relationships pertaining to an error
signal. For instance when I have to go to the bathroom on my road trip, I
might remember having drank that extra cup of coffee, I might imagine a
swollen bladder in my lower abdomen and ways to empty it. Once I perceive
higher level relationships, I can use them to form higher level references
which also reduce the lower level errors. In order to satisfy both event A
and B occurring without conflicting, I may have to go up to form a sequence
that allows one to occur after the other. In order to have the sequence
occur I might have to come up with a program. In order to be able to come up
with programs that work, I need to have sound principles. Moral principles
likely evolved partly out of a need to reduce the lower perceptions of costly
conflicting interpersonal interactions.

I was originally going to put this in a different post, but I think it is
appropriate here to make a few comments about nonrandom aspects of
reorganization which simplify the process of hierarchical reorganization (I
think most of this has been proposed before as well). First, the problem of
reorganization is simplified by having low level analogues of intrinsic error
signals available to the hierarchy. It is further simplified by the fact
that we can remember events occurring together and events occurring closely
in time. Memories provide the basis for forming relationship perceptions.
Our ability to induce higher level perceptions would be next to nil without
memory. Third, the search for invariants on the perceptual side is I think
not a totally random process if it is very much random at all. By
associative memory addressing, if event of type B consistently follows event
of type A, than the occurrence of A will evoke B, and this makes it quite
easy for a relationship to be perceived. This relationship forming process
is used to associate events with the intrinsic error analogues available to
the hierarchy (IEA's). If event of type B consistently occurs with IEA of
type A, then a simple output strategy for ceasing IEA of type A is to control
for the cessation of event of type B. Further, if event of type B
consistently occurs before IEA of type A, then a simple output strategy for
preventing IEA of type A is to control for the cessation of event of type B.
(This has tie ins to the substitution process formulated in evolutionary
epistemology) Even if B is not 100% correlated with A, if it is the most
correlated it will have the greatest propensity to be evoked from memory, and
an organism has probably the best chance of error reduction by following the
same strategy of ceasing B if it cannot come up with a more certain strategy.
From the event level sequences and programs need to be found ending in the
cessation of B, and perhaps there is more blind variation with this part of
the process. I think the event level is an important middle level of
perception because it is so readily associated with IEA's. Another
simplifying aspect of reorganization could be that a particular IEA focusses
attention on itself and relevant relationships as when pain focusses our
attention (also proposed before). Now these simplifying processes that
shortcut a more blind reorganization originally evolved via a more blind
process, but I think that most of them are available to human infants.

I think that CSGnet members also often talk as if higher level references
some how have priority over lower level references. I don't think that this
is the case either. What I mean by priority, is that given two mutually
exclusive (conflicting) concerns, the one you act on is the one of higher
priority. This is the one with the higher gain. I do think it is possible
for references at different levels to conflict if they are in different
branches. They just have to compete for the same resources like my mouth,
arms, and awareness. Should I go for the taste sensation of another bowl of
ice-cream or should I adhere to being more healthy? Low level errors may
often have more priority (gain) associated with them because they are
associated with IEA's. Such painful signals can quickly boot other matters
out of awareness. Perhaps it is useful to talk of the errors of the highest
priority in addition to errors of the highest levels. I think an interesting
question is to consider what one's goals are when one is not impinged upon by
low level IEA's such as coldness or hunger (which is most of the time for us
Americans and other 1st world countries). Are there higher level IEA's that
ones goals serve and that promote survival that kick in, or would we just
consider these hierarchical goals that further secure low level IEA's in the
long run. For instance, maybe exploring one's surroundings kicks in after
more basic needs are secured, and this propensity can further secure survival
by coming in handy at future times. Perhaps there are levels of priority
such that when high level priorities are secured, one has greater freedom,
and other more general level intentions kick in sort of like in Maslow's
hierarchy. In any event, I think it is important to consider what the source
or error origination of the highest priority concern is and that it is often
not at the highest level.

This post is longer than I expected. As is often the case, the more I
thought, the more I thought. Please feel free to comment with disagreements
and agreements.

Bob C.

2

[Martin Taylor 991017 0:13]

[From Bob Christensen (991016.1702 PT)]

I've been thinking about the hierarchy for some time, observing my private
experiences closely to see how they fit with my understanding of the
hierarchy, and trying to revise my understanding as I go along.
...
Then there is the part of higher level references categorically setting the
references for lower level references in order to match their perceptions.
Now it has not been easy for me to see just what this implies, because it
might not fit with the way I and probably others categorize intentions with
our natural language notions. It seems to me that CSGnet members often talk
as if everything we do is "in order" to satisfy our highest level references.
If this is what the model implies then this part of the model is quite
disagreeable to me. For instance, If I feel hungry, I think of myself as
engaging in programs and sequences "in order" to reduce this low level
mismatch. I think of hunger pangs as being at the intensity, sensation, and
transition levels. The exact level is beside the point. The point is that
it is a quite low level mismatch.

Etc...

Perhaps the following may help.

What you say is superficially correct, but in the "standard model" it poses
no problem. Let's see why it doesn't.

There is a controlled perception--let's say perception of hunger--for which
a higher level output has provided a reference signal--let's say that I
want to perceive myself as a martyr and can satisfy this reference by
perceiving extreme hunger. In other words, my reference for hunger level
is set to "high" as a consequence of the output of the "perceive me a
martyr" elementary control loop. If I am not feeling hungry (perhaps some
other higher-level system conflicts with the "martyr" system and forces
the "feel hunger" reference to "low") then I might satisfy the "perceive
me a martyr" reference by causing myself some other pain--executing a
program to do so.

Or, my reference level for perceived hunger might be "low" but my actual
perceived hunger might be high. There is indeed error, but there is no
need for the error signal itself to be propagated back up the hierarchy.
If the higher-level system needed hunger to be low, and hunger is high,
then the higher-level unit is experiencing error, and will be varying
its output. That output, part of the higher-level feedback loop, will
cause things to happen that (if successful) will result in the hunger
perception level becoming "low." This allows the higher-level system
in turn to bring its own perception to match its reference level.

Below the high-level system that set (eventually) the reference level
for hunger and the system that controls the hunger perception, there
may be many levels, including program, sequence, event, or others.
If keeping hunger "low" is essential to the correct functioning of the
high-level control system, then all these intermediate levels will
necessarily be generating outputs that eventually result in hunger
becoming "low."

The bottom line is that it may well be the case that in order to bring
one perceptual variable to its reference value, the action of higher
level systems may be required. But these higher-level actions need not
be induced by reference values set directly from the lower-level output
or error signals. They can (and in the standard model, do) come from the
output of some yet higher-level control unit that eventually requires
that the low-level perception take on some value that it does not
currently have.

In effect, the reduction in the "hunger" perception is a _side-effect_ of
the actions of other control systems that have no physical or intrinsic
connection with hunger--such as passing money to a supermarket cashier,
or turning the control on an oven. The external world acts consistently
enough that we can usually rely on these side-effects to bring about
the desired result--the hunger perception matching its reference,
thereby allowing the high-level perception to match its own.

The point here is that if there is a reference value set for the hunger
perception, it is set because that value is what some higher level
control system needs as part of its own perceptual input. If the
higher-level system doesn't get that value for the hunger perception,
its outputs cause things to happen until it does get its desired
value. It is not necessary for the hunger control system itself to
cause the actions that affect the hunger perception. And there's no need
to modify the standard model to allow error signals to propagate
upward (though the possibility is indeed open that there exist control
systems for which the perceptual input comes in part from the error
signals of other control systems. One hypothesis is that such connections
form part of the reorganizing subsystem--but that's not what we are
dealing with, here).

It might be worth noting in this context that it is exactly this kind of
side-effect consistency that allows reorganization to build a perceptual
control hierarchy that successfully sustains the intrinsic variables near
their reference levels. Few, if any, of the perceptions we control all
day have a direct physical connection to such intrinsic variables as
blood CO2 or blood sugar, but our behaviours nevertheless are such as
to maintain those levels where they should be--because we have both
evolved and reorganised so that we control those perceptions that lead
to their maintenance.

Martin

3

[From Bruce Gregory (991017.0804 EDT)]

Martin Taylor 991017 0:13

It might be worth noting in this context that it is exactly this kind of
side-effect consistency that allows reorganization to build a perceptual
control hierarchy that successfully sustains the intrinsic variables near
their reference levels. Few, if any, of the perceptions we control all
day have a direct physical connection to such intrinsic variables as
blood CO2 or blood sugar, but our behaviours nevertheless are such as
to maintain those levels where they should be--because we have both
evolved and reorganised so that we control those perceptions that lead
to their maintenance.

Very fine post Martin. This point took me the longest to get. When I finally
did, the entire picture fell into place.

Bruce Gregory

4

[From Bruce Abbott (991017.1625 EST)]

[Martin Taylor 991017 0:13] --

The bottom line is that it may well be the case that in order to bring
one perceptual variable to its reference value, the action of higher
level systems may be required. But these higher-level actions need not
be induced by reference values set directly from the lower-level output
or error signals. They can (and in the standard model, do) come from the
output of some yet higher-level control unit that eventually requires
that the low-level perception take on some value that it does not
currently have.

In effect, the reduction in the "hunger" perception is a _side-effect_ of
the actions of other control systems that have no physical or intrinsic
connection with hunger--such as passing money to a supermarket cashier,
or turning the control on an oven. The external world acts consistently
enough that we can usually rely on these side-effects to bring about
the desired result--the hunger perception matching its reference,
thereby allowing the high-level perception to match its own.

Martin, I'm afraid that I have to side with Bob Christensen (991016.1702 PT)
on this one. The scenario you present is possible, but unlikely to be
anything like the typical situation, in which systems fairly high in the
hierarchy (e.g., program level) actually become the _means_ by which a
low-level perception is controlled. Reduction in hunger is not a
side-effect of the actions that eventually bring food to mouth, it is in
many cases the _intended consequence_ of these acts, taken as a whole.

How can this fact be harmonized with the hierarchical control structure
proposed in HPCT?

Regards,

Bruce

5

[From Bob Christensen (991017.1441 PT)]

In response to
[Martin Taylor 991017 0:13]

The point of my post was to show some examples of some cases which occur in
which the main purpose of my programs is to come to an event which will
reduce some uncomfortabel feeling such as hunger and to explain how this can
be so. Your case of the Maytr is a different type of case in which the Maytr
self system conflicts with satisfying ones hunger. I could think of other
similar conflicts like when a marathon runner endures much physical pain in
order to finish a race. But, I just wanted to consider the more simple case
of when one is reminded of being hungry by some uncormfortable feelings, and
one just wants to act to change these feelings in the desired direction, and
there are no conflicts from systems concerned with being a martyr or running
a marathon. These simple examples are interesting to me because I think they
show a sort of reversal in the hierarchy (of what is in order to satisfy
what). I don't see how your more complicated example shines any light on
this more simple example except to confound things more which is what I was
trying to avoid.

Or, my reference level for perceived hunger might be "low" but my actual
perceived hunger might be high. There is indeed error, but there is no
need for the error signal itself to be propagated back up the hierarchy.
If the higher-level system needed hunger to be low, and hunger is high,
then the higher-level unit is experiencing error, and will be varying
its output. That output, part of the higher-level feedback loop, will
cause things to happen that (if successful) will result in the hunger
perception level becoming "low." This allows the higher-level system
in turn to bring its own perception to match its reference level.

Below the high-level system that set (eventually) the reference level
for hunger and the system that controls the hunger perception, there
may be many levels, including program, sequence, event, or others.
If keeping hunger "low" is essential to the correct functioning of the
high-level control system, then all these intermediate levels will
necessarily be generating outputs that eventually result in hunger
becoming "low."

There are several levels at which hunger can be perceived. At the lowest
level, I'm just talking about the painful intensity/sensations coming from
one's stomach. Higher level perceptions involve the word "hunger",
associations to other times I've been hungry, understanding that I've not
eaten in a while, and that I need food to resupply my body. Higher level
perceptions are composites/ transformations of many lower level perceptions.
Probably all of the higher level perceptions of hunger incorporate the lowest
level hunger pain signal or some deriviative since it is essential to the
experience. Now this lowest level hunger pain perceptual signal is also like
an error signal. All other things being equal (Martyrdom aside) one can
imagine a more desired value to these painful hunger signals i.e. their
absence, and this implies mismtach/error. As I said in the last post, these
low level hunger signals also covary with intrinsic error associated with
supplies of nutrients. You say there is no need for error to be propagated
back of the hierarchy. I say that these low level hunger pain signals are
essential to perceiving hunger at all levels. Whether you call them
perceptual signals or error signals is besides the point. According to
definitions of the model, anything that goes up to inputs of higher control
systems is called a perception. I think that in general the model allows for
analogues of lower level error signals to be fed to higher levels as
perceptions (or parts of perceptions). There are myriad types of perceptual
functions, why can't one of them incorporate r-p. Now I do think that hunger
has to be perceived at high enough of a level where a reference can result in
actions to reduce all of the lower levels of hunger. If your a grown up,
this might involve driving to a restuarant. If your a baby, it might only
involve crying.

The bottom line is that it may well be the case that in order to bring
one perceptual variable to its reference value, the action of higher
level systems may be required.

Yes, this was one of the main points of the post.

But these higher-level actions need not
be induced by reference values set directly from the lower-level output
or error signals.

I don't think that I argued this. I think that the higher level perceptions
that are associated with higher level references and outputs incorporate
lower level perceptions which incorporate these error/perceptual signals (not
all of the time, but some of the time).

They can (and in the standard model, do) come from the
output of some yet higher-level control unit that eventually requires
that the low-level perception take on some value that it does not
currently have.

I don't think I have a disagreement here but let me ad the following. I
think Bill has suggested that new control systems get organized in the order
of perceptions, then references, then outputs. Once a complex variabel can
be perceived, one can develop a preference, and then output. Organizing the
perception is probably the most difficult part and this is more what I was
talking about. Furthermore, the perception must be composed/transformed from
lower order perceptions.

In effect, the reduction in the "hunger" perception is a _side-effect_ of
the actions of other control systems that have no physical or intrinsic
connection with hunger--such as passing money to a supermarket cashier,
or turning the control on an oven. The external world acts consistently
enough that we can usually rely on these side-effects to bring about
the desired result--the hunger perception matching its reference,
thereby allowing the high-level perception to match its own.

I wouldn't call the reduction in hunger perception a side-effect.
Side-effects are things we do not care about and that do not matter to
whether a control system ceases its output. Without the reduction in hunger
perceptions, the higher level systems would conintue to search for actions
until this hunger reduction occurred.

The point here is that if there is a reference value set for the hunger
perception, it is set because that value is what some higher level
control system needs as part of its own perceptual input.

Going back one step farther, the higher level system became organized in such
a manner in order to reduce intrinsic error who's analogue is availible to
the hierarchy at a low level. The availability of this low level signal
simplifies the abilitiy to organize such a successful high level system.

If the
higher-level system doesn't get that value for the hunger perception,
its outputs cause things to happen until it does get its desired
value.

Okay

It is not necessary for the hunger control system itself to
cause the actions that affect the hunger perception. And there's no need
to modify the standard model to allow error signals to propagate
upward

I don't think that I argue this, or at least I would not word it this way.
As I said before, I think IEA's are available to the hierarchy. Second, in
gerneral I think it is possible for higher level perceptions to incorporate
inputs that are analogues of lower level errors. If I'm driving to a
destination, I'm able to perceive my location, that I a not at my
destination, and the difference. That I am able to perceive the difference
says to me that I have a perceptual function that compares the two.

(though the possibility is indeed open that there exist control
systems for which the perceptual input comes in part from the error
signals of other control systems. One hypothesis is that such connections
form part of the reorganizing subsystem--but that's not what we are
dealing with, here).

In the way I stated directly above, I think hierarchy has similar capacities
which are incorporated in some systems.

It might be worth noting in this context that it is exactly this kind of
side-effect consistency that allows reorganization to build a perceptual
control hierarchy that successfully sustains the intrinsic variables near
their reference levels. Few, if any, of the perceptions we control all
day have a direct physical connection to such intrinsic variables as
blood CO2 or blood sugar, but our behaviours nevertheless are such as
to maintain those levels where they should be--because we have both
evolved and reorganised so that we control those perceptions that lead
to their maintenance.

If I understand you correclty, what you mean by side-effect consistencies,
is that there are lawlike regularities in the physical and socially
constructed world such that acting on the world in one way has indirect
desired effects. For instance, paying the grocery clerk contributes to me
being able to satisfy my hunger. This is what I meant by learning
relationships and models. Granted the higher level models are useful in many
specific cases. I would not use the word side-effect because althought, the
effects may be indirect, they are consistent and intended.

Bob C.

6

[From Bruce Gregory (991017.1757 EDT)]

Bruce Abbott (991017.1625 EST)

Martin, I'm afraid that I have to side with Bob Christensen
(991016.1702 PT)
on this one. The scenario you present is possible, but unlikely to be
anything like the typical situation, in which systems fairly high in the
hierarchy (e.g., program level) actually become the _means_ by which a
low-level perception is controlled. Reduction in hunger is not a
side-effect of the actions that eventually bring food to mouth, it is in
many cases the _intended consequence_ of these acts, taken as a whole.

How can this fact be harmonized with the hierarchical control structure
proposed in HPCT?

I agree that Martin's choice of the term "side-effect" was unfortunate. Each
of the steps he describes are intended outcomes in a sequence intended to
reduce hunger. The test would show this. No need to harmonize this with
anything, it's orthodox HPCT.

Bruce Gregory

7

[From Bruce Gregory (991017.1830 EDT)]

Bob Christensen (991017.1441 PT)

Here is the model I use. The hierarchy exists to control hunger (and other
intrinsic variables). Hunger is not a controlled variable. Hunger is not
part of the hierarchy. Most of the actions taken by the hierarchy to control
hunger happen in the absence of hunger.

Bruce Gregory

8

[From Bruce Abbott (991017.2105 EST)]

Bruce Gregory (991017.1757 EDT) --

I agree that Martin's choice of the term "side-effect" was unfortunate. Each
of the steps he describes are intended outcomes in a sequence intended to
reduce hunger. The test would show this. No need to harmonize this with
anything, it's orthodox HPCT.

In orthodox HPCT, higher-level control systems set the references of
lower-level control systems as the means by which the higher-level systems
control their CVs.

But in the situation we are discussing, an intensity-level (hunger) control
system sets the references of various program- or sequence-level control
systems as the intensity-level system's means of controlling its CV (hunger
level). That's _also_ orthodox HPCT? How so?

Bruce Gregory (991017.1830 EDT)]

Here is the model I use. The hierarchy exists to control hunger (and other
intrinsic variables). Hunger is not a controlled variable. Hunger is not
part of the hierarchy. Most of the actions taken by the hierarchy to control
hunger happen in the absence of hunger.

Hmmm. Hunger is not a controlled variable, but the hierarchy exists to
control hunger, which is not controlled. I must be misunderstanding you.
Please clarify. It sounds to me like you are reverting to Martin's "side
effect" explanation. I don't buy that. I think that when I'm hungry, I
might very well light up the oven for the ultimate purpose of satisfying my
hunger.

Bruce A.

9

[Mark Lazare (101799)]

In a message dated 10/17/1999 7:18:05 PM US Mountain Standard Time,
ABBOTT@IPFW.EDU writes:

<< Hmmm. Hunger is not a controlled variable, but the hierarchy exists to
control hunger, which is not controlled. I must be misunderstanding you. >>

If I may take a shot at this -- Hunger is not a conrolled variable -- no more
that anxiety or depression is a controlled varaible. Hunger is a emotion or
feeling that is an error signal. Hunger is an error signal telling you, your
system is crashing you need fuel, some carbon based matter. Engery is your
controlled Varrable not the hunger -- Hunger and the lack thereof will let
you know if you need to eat or not.

Mark Lazare.

10

[From Bob Christensen (991017.2103 PT)]

[Bruce Gregory (991017.1830 EDT)]

Here is the model I use. The hierarchy exists to control hunger (and other
intrinsic variables).

Okay

Hunger is not a controlled variable.

If hunger is controlled, it is a controlled variable. It does not make sense
to say otherwise.

Hunger is not
part of the hierarchy.

It can be experienced. It can be perceived. It can be remembered. Therfore I
think that it enters the hierarchy at the lowest levels just like the pain
from a splinter.

Most of the actions taken by the hierarchy to control
hunger happen in the absence of hunger.

Some of them do happen in the presence of hunger. When I get hungry I start
a program sending me to the kitchen to scan the refridgerater and pantry. If
theres nothing I like, I might dial for Pizza.

Bob C.

11

[Martin Taylor 991018 1:05]

I think maybe Bruce Gregory understood my post, but it seems neither
Bruce Abbott nor Bob Christensen did. But perhaps even Bruce Gregory
did not. I'll try again.

[From Bruce Abbott (991017.2105 EST)]

Bruce Gregory (991017.1757 EDT) --

I agree that Martin's choice of the term "side-effect" was unfortunate.

I disagree--see the end of this message.

Each
of the steps he describes are intended outcomes in a sequence intended to
reduce hunger. The test would show this. No need to harmonize this with
anything, it's orthodox HPCT.

My point was that the steps taken have the _effect_ of reducing hunger,
but they are part of a control loop at a higher level. That higher-level
control loop has a perceptual signal influenced by the perceived level
of hunger, and it has a reference value for its perception that cannot
be satisfied unless the perceived level of hunger is low. The acts, such
as paying a supermarket cashier or turning the control knob on the oven
do not include the value of the hunger perception in their perceptual
signals, nor do they directly affect the level of hunger.

What affects the perception of hunger is the ingestion of food. What
leads to the ingestion of food is the execution of some program as a
consequence of the output of the control system that set the reference
level for hunger to be low rather than high. That higher-level system has
succeeded in controlling its perception by means that have developed
over time by reorganization. So long as its perception is near its
reference, those acts need not take place. But one buys things at the
market when one is _not_ hungry, and one prepares the oven before
it is time to eat. These acts are _not_ acts to correct an error in
the perception of level of hunger higher than the reference value. They
are acts to correct an error in a higher level system with a longer
time-frame.

In orthodox HPCT, higher-level control systems set the references of
lower-level control systems as the means by which the higher-level systems
control their CVs.

But in the situation we are discussing, an intensity-level (hunger) control
system sets the references of various program- or sequence-level control
systems as the intensity-level system's means of controlling its CV (hunger
level). That's _also_ orthodox HPCT? How so?

That's not orthodox HPCT, and I hope my discussion above argues that it
probably doesn't happen, either.

Bruce Gregory (991017.1830 EDT)]

Here is the model I use. The hierarchy exists to control hunger (and other
intrinsic variables). Hunger is not a controlled variable. Hunger is not
part of the hierarchy. Most of the actions taken by the hierarchy to control
hunger happen in the absence of hunger.

The last sentence is true, but that doesn't mean that the level of
perceived hunger is uncontrolled. There is certainly a difference between
the reference level for hunger under everyday cirumstances, under
circumstances when one is trying to conserve a limited food supply on
a trek, and in the circumstance that one is trying to perceive oneself as
a martyr. And if there are differences in reference levels, it is likely
that the perception is a controlled one, especially if one's actions
differ for a given level of perceived hunger under the three different
sets of circumstances.

It sounds to me like you are reverting to Martin's "side
effect" explanation. I don't buy that. I think that when I'm hungry, I
might very well light up the oven for the ultimate purpose of satisfying my
hunger.

Yes, you very well might. But you wouldn't even be trying to "satisfy"
your hunger if that satisfaction did not satisfy some higher-level
controlled perception, would it now? We aren't in an S-R situation where
perception of hunger MUST lead to immediate eating if food is available.
Imagine getting hungry during an important meeting! What would the boss
say if you just left to get a snack?

···

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

I think you have missed the critical importance of side-effects in the
operation of the hierarchy generally. The whole reorganization system
depends on it--almost all controlled perceptions having no direct
physical connection with the intrinsic variables, and the intrinsic
variables not entering into the value of the controlled perception.
One sustains one's blood sugar by getting along with the neighbours, by
pleasing the boss, by using an umbrella in the rain, by voting for a good
party, etc. etc.

The end result is that one is able to eat when blood sugar is low and
one feels hungry--if one wants to do so. But it's an awfully long stretch
to argue that controlling one's perception of "boss pleased" or "feel dry"
is part of the output of a hunger-control system.

The world _is_ moderately consistent in its operation, and many higher
level control systems work by getting lower-level control systems to act
(i.e. by setting reference levels for low-level controlled perceptions)
in such a way that the side-effect of their control influences different
low-level perceptions. In our constrained world, such side-effects can
hardly be avoided. In a randomly organized set of control systems, they
are almost certainly unhelpful.

But we are not randomly organized--we are _reorganized_.

Reorganization keeps what works, and allows what doesn't to fade
away and change. So the small proportion of helpful and consistent
side effects is kept (according to orthodox HPCT reorganization theory).
It takes a bit of convoluted thinking to avoid this happening while
sticking to HPCT when there are many parallel control units active at
any level of the hierarchy.

So I reject Bruce Gregory's statement that my choice of the term
"side-effect" was unfortunate. It was deliberate and it was accurate.
But the side-effect was not of the control system that "desired" hunger
to be low. It is of the intermediate control systems that do not include
the perception of hunger-level as part of their perceptual input function,
but that nevertheless have the effect of enabling hunger-reduction.
Hunger reduction is a direct effect of the act of eating, and the control
loop for hunger-level is in the direct line of some other high-level
system.

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

I hope all this isn't too incoherent. I should have gone to bed before
starting it!

Martin

12

[From Bob Christensen (991017.2144 PT)]

Bruce Abbott (991017.2105 EST)

In orthodox HPCT, higher-level control systems set the references of
lower-level control systems as the means by which the higher-level systems
control their CVs.

But in the situation we are discussing, an intensity-level (hunger) control
system sets the references of various program- or sequence-level control
systems as the intensity-level system's means of controlling its CV (hunger
level). That's _also_ orthodox HPCT? How so?

I did not hypothesize that lower level control systems set the references for
higher level systems. I hypothesized that lower level errors (or analogues
there of) are incorporated into higher level perceptions starting from the
lowest levels. These error analogue inputs to higher levels are instrumental
in organizing the higher level systems because one can relate them to other
perceptions which often occur with them. In this way one can come to better
understand how the error analogue inputs arise and how one might cease them.
Once the higher level systems are organized the error analogue inputs, when
present, result in errors at the higher levels in comparison to references
that are already set. Then, the error signals at this high level drive
outputs. The net effect can be that when the error analogue inputs are
present they do affect the references set at higher levels because they
result in the error at the highest level (which may be a few levels up) that
changes references at all lower levels (often along branches commonly thought
of as the voluntary nervous system).

If this scenario is correct, I think that it means that it does not make
sense to say that lower order references are always to satisfy higher order
references, or that lower references are always the means to satisfying
higher order references. It's just not necessarily the case. There are
probably often times that one wants an event to occur, but in order to make
it occur one has to see many higher level relationships, and employ a
program. The program ceases when the event occurs, not when some higher
level principle is satisfied. The principle is used to organize the program
to begin with, probably when one was frustrated, and probably when error
analogue inputs from the event level were present.

Bruce Abbott

I think that when I'm hungry, I
might very well light up the oven for the ultimate purpose of satisfying my
hunger.

I concur.

[Mark Lazare (101799)]

If I may take a shot at this -- Hunger is not a conrolled variable -- no more
that anxiety or depression is a controlled varaible. Hunger is a emotion or
feeling that is an error signal. Hunger is an error signal telling you, your
system is crashing you need fuel, some carbon based matter. Engery is your
controlled Varrable not the hunger -- Hunger and the lack thereof will let
you know if you need to eat or not.

Hunger is a variable because it can range from no hunger (satiation) to
extreme hunger. Hunger is controlled via eating usually near the reference
for no hunger. Hunger is a controlled variable. Hunger is perceived and it
is an error in the sense that it is correlated with low "fuel" in the blood.
That's why I called it an intrinsic error analog input to the hierarchy.
Energy is a high level concept. Animals do not have the concept of energy or
models of glucose floating in their blood stream. They feel the
uncomfortable low level perception of hunger and they act to reduce it
through their species specific routines. Emotions are perceptions and
representations of errors as well. But emotions are different from hunger in
this way. Emotions often result when ones hierarchical references are
frustrated or disturbed. Frustrated when stuck in traffic, worried about
performance on tomorrow's presentation. Hunger simply arises through some
mechanism that reflects as you say, low "fuel" in the blood.

The main point of my original post was that higher level references can serve
to change a lower level perception like hunger to a desired state.

Mark Lazare (101799)]

MOL for Disturbing Views---

The Hows and Whys, related to heirarchies. In the MOL, if you want to find
out a motive, goal, desire, proposes, or intention, you ask the question;
WHY?.

Okay.

This is going up a level.

I don't think this is necessarily the case. Comments like this are one of
the reasons I made my original post.

If you what to find and appropriate output, you as HOW?

Okay

This will take you down a level.

I don't think this is necessarily the case either.

Where is the Reference coming from at the highest level in the hierarchy?
Any conjectures anyone?

Maybe it is not originating from the highest level.

Bob C.

13

[From Bruce Gregory (991018.0626 EDT)]

Bruce Abbott (991017.2105 EST)

But in the situation we are discussing, an intensity-level
(hunger) control
system sets the references of various program- or sequence-level control
systems as the intensity-level system's means of controlling its
CV (hunger
level). That's _also_ orthodox HPCT? How so?

No, this is definitely not orthodox HPCT. It appears to be S-R, but without
a model I can't be sure.

>Bruce Gregory (991017.1830 EDT)]

Here is the model I use. The hierarchy exists to control hunger (and other
intrinsic variables). Hunger is not a controlled variable. Hunger is not
part of the hierarchy. Most of the actions taken by the hierarchy
to control
hunger happen in the absence of hunger.

Hmmm. Hunger is not a controlled variable, but the hierarchy exists to
control hunger, which is not controlled.

I used "control" in two different senses, I apologize. Hunger signals a
problem that the hierarchy was constructed to minimize.

I must be misunderstanding you.

No, you seem to understand me quite well.

Please clarify. It sounds to me like you are reverting to Martin's "side
effect" explanation.

Correct. But unlike him, I think the word "side effect" can be misleading.
He and I are using the same models.

I don't buy that. I think that when I'm hungry, I

might very well light up the oven for the ultimate purpose of
satisfying my
hunger.

True. But hunger is not "causing" you to light the oven. This, I believe, is
the sense in which Martin says that alleviating hunger is a "side-effect" of
lighting the stove. You are intentionally lighting the stove because in the
past this has been part of a sequence of actions that led to a reduction in
hunger. In the same way, I drive to work, but getting to work does not cause
me to enter the garage.

Bruce Gregory

14

[From Bruce Gregory (991018.0639 EDT)]

Bob Christensen (991017.2103 PT)

[Bruce Gregory (991017.1830 EDT)]

> Here is the model I use. The hierarchy exists to control hunger
(and other
> intrinsic variables).

Okay

>Hunger is not a controlled variable.

If hunger is controlled, it is a controlled variable. It does not
make sense
to say otherwise.

You are using the word "controlled" in two different ways. Hunger is not a
perception that serves as input to a control loop that maintains this
perception at a desired level.

> Hunger is not
> part of the hierarchy.

It can be experienced. It can be perceived. It can be remembered.
Therefore I
think that it enters the hierarchy at the lowest levels just like the pain
from a splinter.

O.K. We disagree.

>Most of the actions taken by the hierarchy to control
> hunger happen in the absence of hunger.

Some of them do happen in the presence of hunger. When I get
hungry I start
a program sending me to the kitchen to scan the refrigerator and
pantry. If
there's nothing I like, I might dial for Pizza.

Dialing the telephone does not reduce hunger. This is what Martin means when
he says the reduction in hunger is a side-effect of dialing the phone. You
dial the phone intentionally, but this is part of a sequence of actions
constructed to maintain intrinsic error at a low level. I'll admit this
point is not obvious. You'll have to trust me that Martin and I mean
something when we say this. What we mean may not yet be apparent, but we
have an intention.

Bruce Gregory

15

[Martin Taylor 991018 1:05]

>>Each
>>of the steps he describes are intended outcomes in a sequence
intended to
>>reduce hunger. The test would show this. No need to harmonize this with
>>anything, it's orthodox HPCT.

My point was that the steps taken have the _effect_ of reducing hunger,
but they are part of a control loop at a higher level. That higher-level
control loop has a perceptual signal influenced by the perceived level
of hunger, and it has a reference value for its perception that cannot
be satisfied unless the perceived level of hunger is low. The acts, such
as paying a supermarket cashier or turning the control knob on the oven
do not include the value of the hunger perception in their perceptual
signals, nor do they directly affect the level of hunger.

I agree.

What affects the perception of hunger is the ingestion of food. What
leads to the ingestion of food is the execution of some program as a
consequence of the output of the control system that set the reference
level for hunger to be low rather than high. That higher-level system has
succeeded in controlling its perception by means that have developed
over time by reorganization.

Indeed.

So long as its perception is near its
reference, those acts need not take place. But one buys things at the
market when one is _not_ hungry, and one prepares the oven before
it is time to eat. These acts are _not_ acts to correct an error in
the perception of level of hunger higher than the reference value. They
are acts to correct an error in a higher level system with a longer
time-frame.

Yes, that is the point I took.

>>Bruce Gregory (991017.1830 EDT)]
>
>Here is the model I use. The hierarchy exists to control hunger
(and other
>intrinsic variables). Hunger is not a controlled variable. Hunger is not
>part of the hierarchy. Most of the actions taken by the
hierarchy to control
>hunger happen in the absence of hunger.

The last sentence is true, but that doesn't mean that the level of
perceived hunger is uncontrolled. There is certainly a difference between
the reference level for hunger under everyday circumstances, under
circumstances when one is trying to conserve a limited food supply on
a trek, and in the circumstance that one is trying to perceive oneself as
a martyr. And if there are differences in reference levels, it is likely
that the perception is a controlled one, especially if one's actions
differ for a given level of perceived hunger under the three different
sets of circumstances.

I'm not certain. Let me think about this for a while.

Yes, you very well might. But you wouldn't even be trying to "satisfy"
your hunger if that satisfaction did not satisfy some higher-level
controlled perception, would it now? We aren't in an S-R situation where
perception of hunger MUST lead to immediate eating if food is available.
Imagine getting hungry during an important meeting! What would the boss
say if you just left to get a snack?

Yes, my point as well.

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

I think you have missed the critical importance of side-effects in the
operation of the hierarchy generally. The whole reorganization system
depends on it--almost all controlled perceptions having no direct
physical connection with the intrinsic variables, and the intrinsic
variables not entering into the value of the controlled perception.

Yes, this is my understanding.

One sustains one's blood sugar by getting along with the neighbours, by
pleasing the boss, by using an umbrella in the rain, by voting for a good
party, etc. etc.

Yes.

The end result is that one is able to eat when blood sugar is low and
one feels hungry--if one wants to do so. But it's an awfully long stretch
to argue that controlling one's perception of "boss pleased" or "feel dry"
is part of the output of a hunger-control system.

Yes.

The world _is_ moderately consistent in its operation, and many higher
level control systems work by getting lower-level control systems to act
(i.e. by setting reference levels for low-level controlled perceptions)
in such a way that the side-effect of their control influences different
low-level perceptions. In our constrained world, such side-effects can
hardly be avoided. In a randomly organized set of control systems, they
are almost certainly unhelpful.

These may be side-effects, but they are intended side-effects.

But we are not randomly organized--we are _reorganized_.

Reorganization keeps what works, and allows what doesn't to fade
away and change. So the small proportion of helpful and consistent
side effects is kept (according to orthodox HPCT reorganization theory).
It takes a bit of convoluted thinking to avoid this happening while
sticking to HPCT when there are many parallel control units active at
any level of the hierarchy.

So I reject Bruce Gregory's statement that my choice of the term
"side-effect" was unfortunate. It was deliberate and it was accurate.

I think it may be misleading even though accurate. But since you explain
what you mean _perhaps_ the misunderstanding can be minimized. Paying the
cashier is an intended outcome in a series of intended outcomes. It does not
directly reduce hunger.

But the side-effect was not of the control system that "desired" hunger
to be low. It is of the intermediate control systems that do not include
the perception of hunger-level as part of their perceptual input function,
but that nevertheless have the effect of enabling hunger-reduction.
Hunger reduction is a direct effect of the act of eating, and the control
loop for hunger-level is in the direct line of some other high-level
system.

Yes. I think our differences are only terminological save for question
whether hunger is a controlled variable.

Bruce Gregory

16

[From Dick Robertson,991018.0715CDT]

Lazare, Mark Crisis counselor, Phoenix AZ wrote:

[Mark Lazare (101799)]

In a message dated 10/17/1999 7:18:05 PM US Mountain Standard Time,
ABBOTT@IPFW.EDU writes:

<< Hmmm. Hunger is not a controlled variable, but the hierarchy exists to
control hunger, which is not controlled. I must be misunderstanding you. >>

If I may take a shot at this -- Hunger is not a conrolled variable -- no more
that anxiety or depression is a controlled varaible. Hunger is a emotion or
feeling that is an error signal. Hunger is an error signal telling you, your
system is crashing you need fuel, some carbon based matter. Engery is your
controlled Varrable not the hunger -- Hunger and the lack thereof will let
you know if you need to eat or not.

Mark Lazare.

Actually, I doubt that we have any sensors that sense energy directly. It is
known that we have sensors for various sugars, O2 level, CO2 level and so forth.

Best, Dick R.

17

[Martin Taylor 991019 9:32]

[Martin Taylor 991018 1:05]

Bruce Gregory, apparently Mon, 18 Oct 1999 06:51:26 -0400

The world _is_ moderately consistent in its operation, and many higher
level control systems work by getting lower-level control systems to act
(i.e. by setting reference levels for low-level controlled perceptions)
in such a way that the side-effect of their control influences different
low-level perceptions. In our constrained world, such side-effects can
hardly be avoided. In a randomly organized set of control systems, they
are almost certainly unhelpful.

These may be side-effects, but they are intended side-effects.

They are side-effects of the actions of the intermediate-level control
systems. They are intended effects of a higher-level control system. You
can't take an organism-wide view and say that anything the _organism_
intends to happen is not a side-effect. Whether something is a side-effect
or not depends on the structure of a single control loop. If the effect
is in the feedback path between a control unit's output and its _own_
perception, it is not a side effect. If the effect is on something that
changes without directly affecting that unit's perception, it is a
side-effect.

But we are not randomly organized--we are _reorganized_.

Reorganization keeps what works, and allows what doesn't to fade
away and change. So the small proportion of helpful and consistent
side effects is kept (according to orthodox HPCT reorganization theory).
It takes a bit of convoluted thinking to avoid this happening while
sticking to HPCT when there are many parallel control units active at
any level of the hierarchy.

So I reject Bruce Gregory's statement that my choice of the term
"side-effect" was unfortunate. It was deliberate and it was accurate.

I think it may be misleading even though accurate. But since you explain
what you mean _perhaps_ the misunderstanding can be minimized. Paying the
cashier is an intended outcome in a series of intended outcomes. It does not
directly reduce hunger.

And _therefore_ hunger reduction is one possible side-effect of paying
the cashier. There are many other side-effects of this same action,
including a perception of having less money in the wallet, perceiving
the cashier saying "thank you," perceiving weight on the arms shortly
thereafter... None of these enter into the control loop for perceiving
myself to be paying the cashier.

I seem to remember Rick having sent a few messages to this same effect
some years ago, but I'm not intending to go back in the archives to
look them up.

Yes. I think our differences are only terminological save for question
whether hunger is a controlled variable.

Mostly that's true, but the terminological difference stems from the
meaning of "side-effect". I use it only to indicate an effect of the
output of a control loop that does not influence the perception controlled
in that loop. You seem to use it the mean an effect that does not influence
_any_ perception controlled within the same hierarchy. I don't think your
usage is technically well-defined.

···

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

Side Note:
There are cases where the definitionof "side-effect" might get blurred,
but that's usually the case when we use language to describe physical
systems. Under Ted Nelson's dictum "everything is deeply intertwingled,"
there may be very slight cross-influences everywhere, so that it would
be more accurate to describe a side-effect as an effect of a control
unit's output that has _almost_ no influence on the controlled perception.
But at the level of precision common in these discussions, "no influence"
is as accurate and easier to talk about.

Martin

18

[From Bruce Nevin (991018.1005 EDT)]

Martin Taylor 991018 1:05--

···

At 01:47 AM 10/18/1999 -0400, Martin Taylor wrote:

Reorganization keeps what works, and allows what doesn't to fade
away and change.

Suppose reorganization is a continual background fact of life, and what is
relevant is something in control systems that resists it, protecting
control loops that are working and preventing them from fading away and
changing. This would perhaps solve the localization problem.

  Bruce Nevin

19

[From Bruce Gregory (991018.1022 EDT)]

Martin Taylor 991019 9:32

They are side-effects of the actions of the intermediate-level control
systems. They are intended effects of a higher-level control
system. You
can't take an organism-wide view and say that anything the _organism_
intends to happen is not a side-effect. Whether something is
a side-effect
or not depends on the structure of a single control loop. If
the effect
is in the feedback path between a control unit's output and its _own_
perception, it is not a side effect. If the effect is on
something that
changes without directly affecting that unit's perception, it is a
side-effect.

O.K. That's clear.

And _therefore_ hunger reduction is one possible side-effect of paying
the cashier. There are many other side-effects of this same action,
including a perception of having less money in the wallet, perceiving
the cashier saying "thank you," perceiving weight on the arms shortly
thereafter... None of these enter into the control loop for perceiving
myself to be paying the cashier.

Here's my problem. I do not pay the cashier in order to have less money
in my pocket, which I agree is a side-effect. Nor do I pay the cashier
in order for her to say thank you (although I may expect her to do so),
this too I agree is a side-effect. I do not pay the cashier to feel a
weight in my arms--I'm perfectly happy for someone else to carry the
groceries to the car. I do pay the cashier as part of a plan called
"doing the weekly grocery shopping." This is an intended outcome.
Calling it a side-effect seems misleading to me.

Mostly that's true, but the terminological difference stems from the
meaning of "side-effect". I use it only to indicate an effect of the
output of a control loop that does not influence the
perception controlled
in that loop. You seem to use it the mean an effect that does
not influence
_any_ perception controlled within the same hierarchy. I
don't think your
usage is technically well-defined.

No, I use it to mean that which is not part of the intended outcome. I
think I understand your usage but I find it unilluminating. Killing you
is, if I understand you properly, a side-effect of pulling the trigger
on my pistol. I tend to think of pulling the trigger as part of the plan
to achieve the outcome of killing you.

Bruce Gregory

20

[From Bruce Nevin (991018.1111 EDT)]

Martin Taylor 991017 0:13--

Or, my reference level for perceived hunger might be "low" but my actual
perceived hunger might be high. There is indeed error, but there is no
need for the error signal itself to be propagated back up the hierarchy.
If the higher-level system needed hunger to be low, and hunger is high,
then the higher-level unit is experiencing error, and will be varying
its output. That output, part of the higher-level feedback loop, will
cause things to happen that (if successful) will result in the hunger
perception level becoming "low." This allows the higher-level system
in turn to bring its own perception to match its reference level.

So we may mistakenly think we see reorganization when another loop is
controlling the same perception (hunger), and this other loop has other
means for controlling that perception.

We are familiar with the scenario of two loops controlling the same input,
it is what happens in conflict. What is less familiar is the possibility of
the two loops being at different levels, perhaps widely different. We often
control lower-level perceptions by means of higher-level control-- control
intensity of sound by turning down the volume control knob, control
intensity of light by pulling down the sun visor in a car or putting on
sunglasses. Perhaps this explains why MOL experience often doesn't seem to
fit the implicit 11-step geometry of "going up a level."

The perceptual signal is passed up the hierarchy, not the error signal.
However, it is not the same perceptual signal that is being controlled at
the higher level, right? We assume that multiple perceptual inputs are
combined in the input function at the higher level to construct a
perceptual signal of a different kind.

It is possible that the lower-level perceptual signal is not passed up the
hierarchy. At the lower level, a nutrient like glycogen has slipped below
the reference level; at a higher level hunger is perhaps a combination of
stomach pangs, weakness, lightheadedness, etc., and these perceptions that
are controlled at the higher level are symptoms of error at lower levels.
(Low blood sugar to the nervous system and muscles has physiological
effects perceived as weakness and lightheadedness. The stomach attempts to
digest as means of controlling glucose level, but there is no food there,
and the result is stomach pangs.)

  Bruce Nevin

···

At 12:45 AM 10/17/1999 -0400, Martin Taylor wrote: