[From Bruce Abbott (980413.1935 EST)]
Rick Marken (980413.1430) --
I don't think the book Bill read was a disturbance that started
reorganization; I think it just allowed him to gain a perspective
on an existing conflict (that was already creating error) that
allowed him to see an acceptable solution (ptting up with the
pain of craving for a _finite_ time).
Dear Ann Landers-Marken:
Where in HPCT do I find "gain a perspective" and "see an acceptable solution"?
(signed)
Floundering in Fort Wayne
[From Bruce Gregory 9980414.1730 EDT)]
Rick Marken (980414.1410)
Me:
> I don't think quitting smoking is an example of hierarchical
> control.Bruce Gregory 9980414.1405 EDT) --
> What may be involved here is control at the program level.
That's still normal hierarchical control; control of a program
perception. It doesn't explain why it is so difficult to stop
smoking.
Why is it so hard for some people to learn to serve a tennis ball properly?
I suppose there might be some conflict involved, but I suspect it is more
like not knowing the sequence of perceptions to control. it is always
frustrating when you want to go from A to B and don't know the route that
will accomplish this.
> "What series of perceptions do I need to control to make X
> happen?"I think people often assume that there is a programmatic solution
(a plan, a diet, etc) to a conflict. In fact, there is none.
I see you are being your usual optimistic self.
As long
as the incompatible goals that cause the conflict exist, there is
no series of steps that will solve any conflict (achieve the
incompatible goals that create the conflict).
I hear what you say, but I am not totally convinced. It seems to me that
what you call "going up a level" may involve program level change. Since
"going up a level" and its attendant virtues lie outside HPCT, I'd prefer to
look for less esoteric explanations--ones that might possibly be modeled.
The idea that there are programmatic solutions to conflicts is what,
I believe, makes it difficult for people to "go up a level" and see
their conflicts from a point of view that makes a solution possible.
Yeah, well you know how we optimists are...
> In the hypothetical "Bill" example reading the book provides
> these new perceptions.This suggests that reading the book causes a change in some
program level perceptual input functions. Such a change in a
perceptual input function is what is needed to provide new program
level perceptions . It's hard for me to see how reading a book
could change perceptual input functions.
What exactly do you think reading a book (or an e-mail for that matter) does
do? I can think of many tests at the program level that yield one result
before reading and e-mail message an other afterward. If I couldn't, I'd be
hard pressed to send and read e-mail.
Anyway, PCT does not
assume that books can change perceptual input functions so
as to provide new perceptions. Changing perceptual input functions
is the job of the reorganizing system.
I don't think I'm calling for creating new perceptual input functions. But
perhaps I am. If so, I waste a lot of time reading books like B:CP. It sure
doesn't _feel_ like a waste of time.
Best Offer
[From Rick Marken (980414.1410)]
Me:
I don't think quitting smoking is an example of hierarchical
control.
Bruce Gregory 9980414.1405 EDT) --
What may be involved here is control at the program level.
That's still normal hierarchical control; control of a program
perception. It doesn't explain why it is so difficult to stop
smoking.
"What series of perceptions do I need to control to make X
happen?"
I think people often assume that there is a programmatic solution
(a plan, a diet, etc) to a conflict. In fact, there is none. As long
as the incompatible goals that cause the conflict exist, there is
no series of steps that will solve any conflict (achieve the
incompatible goals that create the conflict).
The idea that there are programmatic solutions to conflicts is what,
I believe, makes it difficult for people to "go up a level" and see
their conflicts from a point of view that makes a solution possible.
In the hypothetical "Bill" example reading the book provides
these new perceptions.
This suggests that reading the book causes a change in some
program level perceptual input functions. Such a change in a
perceptual input function is what is needed to provide new program
level perceptions . It's hard for me to see how reading a book
could change perceptual input functions. Anyway, PCT does not
assume that books can change perceptual input functions so
as to provide new perceptions. Changing perceptual input functions
is the job of the reorganizing system.
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[From Rick Marken (980414.1550)]
Bruce Gregory 9980414.1730 EDT)--
Why is it so hard for some people to learn to serve a tennis ball
properly?
My guess is that, in most cases, it is difficult because people
have a hard time figuring out which perceptual variables to control,
at what level to control them and how to control them. The
difficulties involved in learning a tennis serve are quite different
than the difficulties involved in "learning" to stop smoking (I
base this on personal experience with having "learned" to do both).
Me:
As long as the incompatible goals that cause the conflict exist,
there is no series of steps that will solve any conflict (achieve
the incompatible goals that create the conflict).
Bruce:
Since "going up a level" and its attendant virtues lie outside HPCT,
I'd prefer to look for less esoteric explanations--ones that might
possibly be modeled.
We have modeled conflicts like those involved in smoking. There
is no way to solve such conflicts by programmatic means; the models
show us this. You can also demonstrate this to yourself in my
"Cost of Conflict" demo at
http://home.earthlink.net/~rmarken/ControlDemo/Conflict.html. As long as
you want to control _both_
the x and y position of the cursor you are in conflict; there is
no clever set of steps, no program, that will allow you to control
the cursor in both dimensions simultaneously. The same is true with
smoking. As long as control systems in you want to control both
your coughing and your nicotine level you are in conflict; you
can't control both variables simultaneously. Control of coughing
requires that smoking be controlled near 0; control of nicotine
level requires that smoking be controlled at, say, 5 packs a day.
The net reference for smoking will, therefore, be something like
2 1/2 packs a day; neither the cough nor the nicotine control
system is happy, but both are working against each other to keep
smoking going at about 2 1/2 packs a day.
This is what modeling tells us; if there is a conflict like this
in the hierarchy then there is no way that the hierarchy itself
can "fix" the conflict. At least, I can think of no way. If you
think that control of a program perception can solve such a
conflict then please show me how -- in terms of a working,
hierarchical model.
Because I see no way for the hierarchy itself to solve conflicts
that exist within itself, and because I see people (including
myself) solve conflicts all the time -- sometimes with great ease --
I believe that there is something about people that lets them do
this. My own experience with solving conflicts leads me to believe
that consciousness is involved in this process. Again, I don't know
how consciousness "works"; all I know is that the HPCT model can't
solve it's own conflicts. Reorganization, which we _can_ model,
can solve some conflicts. But I can solve most of my little conflicts
"instantly" when I can see them from a new conscious perspective;
this may be "reorganization" but it is not the kind of reorganization
we have modeled; it seems to involve a change of conscious perspective
on the problem; that's why I talk about going "up a level" as a way
to solve conflicts. You might not like this "mushy" approach to
solving conflict but it works for me and it is consistent with
modeling, which shows that a hierarchy of control systems cannot
rid itself of conflict.
What exactly do you think reading a book (or an e-mail for that
matter) does do?
I think it evokes imaginings that are based on my existing
perceptual input functions.
I don't think I'm calling for creating new perceptual input
functions. But perhaps I am. If so, I waste a lot of time reading
books like B:CP.
I think some books can lead us to develop (via reorganization) new
perceptual input functions: new ways of perceiving. But I think
it takes time and effort to develop these new perceptual functions.
And the book itself doesn't do it, of course; we do it ourselves.
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[From Bruce Gregory (980415.0306 EDT)]
Rick Marken (980414.1550)
Bruce Gregory 9980414.1730 EDT)--
Why is it so hard for some people to learn to serve a tennis ball
properly?My guess is that, in most cases, it is difficult because people
have a hard time figuring out which perceptual variables to control,
at what level to control them and how to control them.
Yes, this is my point, too.
The
difficulties involved in learning a tennis serve are quite different
than the difficulties involved in "learning" to stop smoking (I
base this on personal experience with having "learned" to do both).
Yes, addictive behavior obscures the real point I wanted to make, which was
the one you just made.
Me:
As long as the incompatible goals that cause the conflict exist,
there is no series of steps that will solve any conflict (achieve
the incompatible goals that create the conflict).Bruce:
Since "going up a level" and its attendant virtues lie outside HPCT,
I'd prefer to look for less esoteric explanations--ones that might
possibly be modeled.We have modeled conflicts like those involved in smoking. There
is no way to solve such conflicts by programmatic means; the models
show us this.
This is what modeling tells us; if there is a conflict like this
in the hierarchy then there is no way that the hierarchy itself
can "fix" the conflict. At least, I can think of no way. If you
think that control of a program perception can solve such a
conflict then please show me how -- in terms of a working,
hierarchical model.
A reasonable request. The first place I'll look is whether the persistence
of incompatible goals may not be an artifact of the fact that loops at the
same being level are unable to influence each other, a point Martin has
raised in the past. Or have I got this wrong?
Because I see no way for the hierarchy itself to solve conflicts
that exist within itself, and because I see people (including
myself) solve conflicts all the time -- sometimes with great ease --
I believe that there is something about people that lets them do
this. My own experience with solving conflicts leads me to believe
that consciousness is involved in this process. Again, I don't know
how consciousness "works"; all I know is that the HPCT model can't
solve it's own conflicts. Reorganization, which we _can_ model,
can solve some conflicts. But I can solve most of my little conflicts
"instantly" when I can see them from a new conscious perspective;
this may be "reorganization" but it is not the kind of reorganization
we have modeled; it seems to involve a change of conscious perspective
on the problem; that's why I talk about going "up a level" as a way
to solve conflicts. You might not like this "mushy" approach to
solving conflict but it works for me and it is consistent with
modeling, which shows that a hierarchy of control systems cannot
rid itself of conflict.
This is a very clear way of stating my major disatisfaction with the HPCT
model. In order to model behavior an unmodeled element must intervene to
overcome the limitations of the model. You probably recall that Newton
thought God had to intervene to solve long-term instabilities in the solar
system. Such requirements sugest that we haven't yet thought things through
properly--to me at least.
What exactly do you think reading a book (or an e-mail for that
matter) does do?I think it evokes imaginings that are based on my existing
perceptual input functions.
And what exactly do these imaginings do?
I don't think I'm calling for creating new perceptual input
functions. But perhaps I am. If so, I waste a lot of time reading
books like B:CP.I think some books can lead us to develop (via reorganization) new
perceptual input functions: new ways of perceiving. But I think
it takes time and effort to develop these new perceptual functions.
And the book itself doesn't do it, of course; we do it ourselves.
My hunch is that reorganization is a convenient culprit to round up with the
usual suspects (and about as likely to be guilty). Mary noted in a post of a
few weeks ago that most learning does not really require us to develop new
abilities. Further, we hardly seem to employ random variations. The real
problem, in my view, is how we make very non-random changes in the
organization of our perceptual functions.
Best Offer
[From Rick Marken (980415.1100)]
Me:
This is what modeling tells us; if there is a conflict like this
in the hierarchy then there is no way that the hierarchy itself
can "fix" the conflict.
Bruce Gregory (980415.0306 EDT)
The first place I'll look is whether the persistence of incompatible
goals may not be an artifact of the fact that loops at the
same being level are unable to influence each other, a point
Martin has raised in the past. Or have I got this wrong?
Conflict ("persistence of incompatible goals") is not an artifact.
It's what happens in a control hierarchy when two or more control
systems at the same level try to control the same (or similar)
perceptions. I strongly suggest that you read (or re-read) my
spreadsheet model paper (_Mind Readings_, p 133) while running
the hierarchical spreadsheet model that can be found at
http://home.earthlink.net/~rmarken/demos.html
The spreadsheet model at the demo site is a little different from
the one described in the paper but I think you can still figure
out how to use it. You can create a conflict in the hierarchy
very easily by changing the perceptions controlled by the level
2 systems. These perceptions are specified by the numbers in the
first 6 rows of column V of the spreadsheet. If you change the
numbers in the first 2 rows so that they are both 7, say, then systems
1 and 2 at level 2 will be trying to control the same perceptual
variable and there will be conflict. The conflict will be seen
(when you recalculate the spreadsheet) as the failure of either system
1 or 2 to get the perception it wants; this is because higher level
systems are setting "incompatible goals" for these two systems; the
same perceptual variable can't be in two different states at the
same time. So the level 3 systems that use these level 2 systems
don't get what they want either.
This is a very clear way of stating my major disatisfaction with
the HPCT model. In order to model behavior an unmodeled element
must intervene to overcome the limitations of the model.
1. You can't model a behavior (like conflict resolution) if you
have no data to model.
2. The reorganization system is a working part of the HPCT model and
it does resolve conflicts. So this is not really an unmodeled element
of the model; it's just untested.
Such requirements sugest that we haven't yet thought things through
properly--to me at least.
They suggest to me that we haven't got much relevant data against
which to test a model.
Me:
I think it [reading] evokes imaginings that are based on my existing
perceptual input functions.
Bruce:
And what exactly do these imaginings do?
Like perceptions, imaginings are controlled for various reasons.
I don't think imaginings do anything; it's what we do with
imaginings that's interesting; we entertain ourselves, plan,
solve problems, all kinds of things.
My hunch is that reorganization is a convenient culprit to round
up with the usual suspects...
I'm shocked. Shocked!
Actually, the reorganization system was designed to model the
apparently random trial and error activities that are observed
before an organism manages to get some variable (food intake, say)
under control. It's not a catch all; it's a model of how organisms
learn to control.
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[From Bruce Gregory 9980415.1750 EDT)]
Rick Marken (980415.1100)
Me:
>This is what modeling tells us; if there is a conflict like this
>in the hierarchy then there is no way that the hierarchy itself
>can "fix" the conflict.Bruce Gregory (980415.0306 EDT)
> The first place I'll look is whether the persistence of incompatible
> goals may not be an artifact of the fact that loops at the
> same being level are unable to influence each other, a point
> Martin has raised in the past. Or have I got this wrong?Conflict ("persistence of incompatible goals") is not an artifact.
I meant to emphasize "persistence". I have no doubt that conflict is real.
What I was questioning is whether your conclusion this conflict _must_
persist and cannot be "fixed" by the hierarchy.
It's what happens in a control hierarchy when two or more control
systems at the same level try to control the same (or similar)
perceptions. I strongly suggest that you read (or re-read) my
spreadsheet model paper (_Mind Readings_, p 133) while running
the hierarchical spreadsheet model that can be found athttp://home.earthlink.net/~rmarken/demos.html
The spreadsheet model at the demo site is a little different from
the one described in the paper but I think you can still figure
out how to use it. You can create a conflict in the hierarchy
very easily by changing the perceptions controlled by the level
2 systems. These perceptions are specified by the numbers in the
first 6 rows of column V of the spreadsheet. If you change the
numbers in the first 2 rows so that they are both 7, say, then systems
1 and 2 at level 2 will be trying to control the same perceptual
variable and there will be conflict. The conflict will be seen
(when you recalculate the spreadsheet) as the failure of either system
1 or 2 to get the perception it wants; this is because higher level
systems are setting "incompatible goals" for these two systems; the
same perceptual variable can't be in two different states at the
same time. So the level 3 systems that use these level 2 systems
don't get what they want either.
Yes, I see this. But doesn't this reasoning apply only to situations where
two higher level systems are trying to control the same perceptual variable?
In other words, doesn't this demo show that there are some "pathological"
organizations that the hierarchy cannot correct? It is not obvious how
common such organizations are, however, in the real world. I suppose that
one danger of reorganization is that it might create such arrangements. This
possibility suggests that random reorganization may not be a common
mechanism.
By the way, I am perfectly prepared to believe that smoking involves a
reference setting that is _not_ accessible to higher levels, i.e., it is the
"top" of a hierarchy. In such a case you may _want_ to stop smoking, but
there is no effective way to accomplish this. However, you can use the "sour
grapes" mechanism to decrease the gain associated with wanting to stop
smoking and reduce the conflict in this way. or so I would think.
me:
> My hunch is that reorganization is a convenient culprit to round
> up with the usual suspects...I'm shocked. Shocked!
Actually, the reorganization system was designed to model the
apparently random trial and error activities that are observed
before an organism manages to get some variable (food intake, say)
under control. It's not a catch all; it's a model of how organisms
learn to control.
I only meant that it is a convenient culprit because it is not at all
obvious, to me at least, that random trial and error processes are involved
in learning many control processes. To use your example, when you imagine
that doing something will lead to an outcome and then do it and it works, it
is not obvious that random trial and error plays a fundamental role.
Best Offer
[From Bill Powers (980415.1830 MDT)]
Bruce Gregory 9980415.1750 EDT)--
I only meant that it is a convenient culprit because it is not at all
obvious, to me at least, that random trial and error processes are involved
in learning many control processes. To use your example, when you imagine
that doing something will lead to an outcome and then do it and it works, it
is not obvious that random trial and error plays a fundamental role.
Good observation. Planning is not reorganization.
Best,
Bill P.
[From Bruce Nevin (980416.1236 EDT)]
Bruce Gregory (980415.0306 EDT)
The first place I'll look is whether the persistence of incompatible
goals may not be an artifact of the fact that loops at the
same [...] level are unable to influence each other, a point
Martin has raised in the past. Or have I got this wrong?
This is a claim that e.g. a perception at level n (a signal of type n put
together by a perceptual input function from multiple inputs) cannot be
copied and made an input to the PIF of another comparator at the same level
n.
Is there some principled reason for this? That is, is the claim that if
such a connection were to result from reorganization, it would be abandoned
(reorganization would continue) because it necessarily cannot work to
reduce error under any circumstances?
If it is an empirical observation at lower levels the claim is less sure.
Bruce Nevin
[From Rick Marken (980416.0910)]
Bruce Gregory 9980415.1750 EDT)
What I was questioning is whether your conclusion this conflict
_must_ persist and cannot be "fixed" by the hierarchy.
A conflict exists whenever two or more control systems try to keep
the same perceptual variable in two or more different states _at
the same time_.. A conflict can be "persistant" (like the smoking
conflict) or "transient" (like Bruce Nevin's (980415.1517 EDT) nice
"throwing" example). Transient conflicts happen all the time. They
occur because circumstances (like the one Bruce Nevin describes) end
up requiring that perfectly "normal" control systems work against
each other.
For example, consider the monkey reaching for food in a narrow neck
bottle; when the monkey grasps the food it can't remove it's hand
from the bottle. Circumstances (the food in the narrow neck bottle)
have created a conflict between two perfectly normal control systems --
the system controlling for grasping the food and the system controlling
for removing the hand from the bottle. Circumstances have made it
necessary for these control systems to have the perception of the
configuration of the hand in two different states at the same time;
the "grasp food" system needs to have hand in a "grasp" shape; the
"remove hand from bottle" system needs to have the hand in an
"elongated" shape.
Transient conflicts usually go away when circumstances change;
the bottle breaks, for example. Or they go away because the conflicted
individual is able to "go up a level". Apparently monkeys cannot go
up a level above the food in bottle conflict (so I hear) and will remain
with their hand stuck in the bottle for quite some time; a human would
probably be able to solve ("fix") this confllict in a few seconds. I
don't believe that these little transient conflicts are solved by the
hierarchy itself; I think they are solved when the individual notices
he or she is _in_ a conflict and revises his or her goals based on
"higher level" considerations.
But if you think that the hierarchy itself can "fix" conflicts then I
would like to see a model of how it's done. One of the virtues
of modeling is that it makes you think about the detailed mechanisms
that would have to be involved in doing things like "fixing" a conflict.
So far, my experience with modeling hierarchical control organizations
leads me to believe that there is no way to build the hierarchy _itself_
so that it can solve conflicts when they come up. But maybe I'm wrong;
I really would like to see a model of how this might be done.
Yes, I see this. But doesn't this reasoning apply only to situations
where two higher level systems are trying to control the same
perceptual variable? In other words, doesn't this demo show that
there are some "pathological" organizations that the hierarchy
cannot correct?
I hope my discussion above answers these questions. First, conflict
_always_ involves two or more higher level control systems specifying
incompatible goals for the same lower level perceptual variable.
Second, this is not the result of pathology; conflict is almost
certain to happen to some extent when perfectly normal control systems
operate in the context of other perfectly normal control systems; the
more normal the control systems (the better the control systems the
worse the conflicts will be. The only way to eliminate conflict is to
stop controlling. Conflict is _the_ problem for systems (like living
organisms) that are made up of and interact with many control systems.
By the way, the only thing "pathological" about any conflict is that
you can't control what you want to control. The smoking conflict is
pathological only because the person can't control for, say, nicotine
level _and_ not coughing. If the person wasn't controlling for not
coughing and, thus, was able to control for nicotine level just fine
(by smoking as much as needed) then there would be no conflict (and
no pathology). The smoking conflict was "pathological" only because it
prevented the control systems involved (nicotine and cough control)
from controlling, not because there is anything intrinsically wrong
with smoking (or coughing).
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[From Bill Powers (980416.1109 MDT)]
Bruce Nevin (980416.1236 EDT)--
This is a claim that e.g. a perception at level n (a signal of type n put
together by a perceptual input function from multiple inputs) cannot be
copied and made an input to the PIF of another comparator at the same level
n.
Never tried it. In the HPCT model, copies of perceptual signals always go
to inputs at a higher level. We've tried that one, and it works. Seems to
me that doing as you say would either create a duplicate of the first
system (outputs affect same inputs) or produce conflict (what would assure
that the reference signals would be the same?). But if you can devise a
working model with that arrangement, it would take precedence over verbal
reasoning.
Is there some principled reason for this? That is, is the claim that if
such a connection were to result from reorganization, it would be abandoned
(reorganization would continue) because it necessarily cannot work to
reduce error under any circumstances?
The only principle I see right off is that if you have two control systems
at the same level controlling the same perception, any difference in the
reference signals would result in conflict (the perceptual signals would
have to be in two different states at the same time). If there is conflict,
errors would be large, reorganization would be likely, and that
organization would tend to be changed.
Best,
Bill P.
[From Bruce Gregory (980416.1352 EDT)]
Rick Marken (980416.0910)
Transient conflicts usually go away when circumstances change;
the bottle breaks, for example. Or they go away because the conflicted
individual is able to "go up a level". Apparently monkeys cannot go
up a level above the food in bottle conflict (so I hear) and will remain
with their hand stuck in the bottle for quite some time; a human would
probably be able to solve ("fix") this conflict in a few seconds. I
don't believe that these little transient conflicts are solved by the
hierarchy itself; I think they are solved when the individual notices
he or she is _in_ a conflict and revises his or her goals based on
"higher level" considerations.
I agree.
But if you think that the hierarchy itself can "fix" conflicts then I
would like to see a model of how it's done. One of the virtues
of modeling is that it makes you think about the detailed mechanisms
that would have to be involved in doing things like "fixing" a conflict.
I have a few thoughts. They need to jell before I know whether they might
lead anywhere.
So far, my experience with modeling hierarchical control organizations
leads me to believe that there is no way to build the hierarchy _itself_
so that it can solve conflicts when they come up. But maybe I'm wrong;
I really would like to see a model of how this might be done.
Me too! When I've gotten a little further along, I no doubt call on you for
advice.
I hope my discussion above answers these questions. First, conflict
_always_ involves two or more higher level control systems specifying
incompatible goals for the same lower level perceptual variable.
Second, this is not the result of pathology; conflict is almost
certain to happen to some extent when perfectly normal control systems
operate in the context of other perfectly normal control systems; the
more normal the control systems (the better the control systems the
worse the conflicts will be. The only way to eliminate conflict is to
stop controlling. Conflict is _the_ problem for systems (like living
organisms) that are made up of and interact with many control systems.
Yes, I see this now. This exchange has been very helpful. I appreciate your
thoughtful and thought provoking insights.
Best Offer