What are we arguing about?

CSGnet Archive CSG1999
1

[From Rick Marken (991222.1440)]

Martin Taylor (991222 14:37) --

What I'm trying to show is a major distinction between the
control of analogue variables and the control of categorical/logical
variables, a distinction well demonstrated by your spreadsheet.

Well, then we are interested in different things. There is certainly
a difference between categorical and analog variables but none that
seems to make an interesting difference in the context of my
spreadsheet model.

Along with Bruce Gregory (991221.1741 EST) we can ask why it is that
trouble so seldom happens when we really do put constraints on people's
low-level reference values or on their external feedback loops (which
in control theory has much the same effect).

Because the constraints (internal goals) are not imposed. Most
people set these "constraint" goals (like the goal of following
traffic laws) for their own (higher level) reasons. And they
control for those goals only to the extent that doing so doesn't
interfere with their ability to achieve other goals. Few people
are so committed to following traffic laws that they would not
violate any one of these laws (such as the speed limit) in order
to achieve higher level goals (such as getting their injured
child to the emergency room quickly).

My problem with simple agreement is that although the spreadsheet
clearly shows that _some_ of the control the system had is lost,
it also shows that the probability this matters in any particular
case is very low, and therefore the relevance of this loss of control
to any real-world sitation may also be small.

The demo says nothing about the "probability this [loss of a
control system] matters in any particular case". It shows only
that fixing any reference at an arbitrary value leads to loss
of control. It says nothing about the probability that this loss
of control will "matter" (whatever that means) or not

given that we both acknowledge (and on my side have always
acknowledged) that there exist reference conditions at the third
level that cannot be met when a second-level reference is fixed,
do you agree that the difference is that you find it definitive
that such conditions exist, whereas I emphasise more the fact
that they are very rare in a big hierarchy?

Not quite. From your point of view, the demo shows that any
control problem created by "committing" to any arbitrarily selected
second level reference is _small_. From my point of view, the demo
shows that "committing" to any arbitrarily selected second level
reference makes it impossible for other control systems (higher
(4th) level or intrinsic systems) to use the level 3 systems to
achieve their perceptual goals. Whether or not this is a "small"
matter is something that can only be determined by the
hierarchical control system that has been deprived of this
control system.

Your point of view is that forced commitment to a goal isn't
necessarily a big deal. My point of view is that your point of
view is 1) wrong (you have no idea, in any particular case,
whether the loss of control that results from forced commitment
to a goal will be a big deal or not) 2) disrespectful (because
it assumes that you _do_ know that the loss of control will
be no big deal to the system) and 3) misleading (because it
makes PCT seem complicitous in your advocacy of the judicious
use of coercion -- such as that which is involved in getting
a child to make a commitment).

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates mailto: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

2

[Martin Taylor 991222 14:37]

[From Rick Marken (991222.1000)]

But then you are not _freely_ varying the level 3 references;
you are varying them within the constraint that the values
selected for these reference do _not_ lead to "intrinsic
contradictions". If the spreadsheet were a real person, that
would mean that some goals that the person could easily achieve
prior to the commitment can no longer be successfully achieved
after the commitment.

> Let's see if we can agree on something. That would be nice,
> wouldn't it?

I suppose. But you manage to make things far more complicated
than I think they need to be.

I guess that's in the eye of the beholder. What I'm trying to show is
a major distinction between the control of analogue variables and the
control of categorical/logical variables, a distinction well demonstrated
by your spreadsheet. I'm also trying to make precise (and therefore
ultimately clearer) exactly what are the conditions that matter.

I don't even know what we are
arguing about.

I'm beginning to think I do. It's much the same argument as we have had
several times in different guises over the years. Your thinking tends
to be what I would call "absolutist." If a thing doesn't work perfectly,
it doesn't work at all. I tend to take more of an engineering approach.
If it works pretty well most of the time (as real people do), I think
that's important to know, in addition to the fact that it doesn't work
perfectly all the time. I see shades of grey where you see "not white."
I also believe that any model that predicts all people control all
their perceptions perfectly all the time is far from being relevant
to the real world.

In the present case there are (for me) two issues: one is the purely
scientific question of how the spreadsheet actually behaves under
various conditions. As an earlier message today shows, and a message
I was working on when your arrived extends, the spreadsheet hierarchy
controls pretty well, even if not perfectly, for all reference values
at level 3, when even as many as three of the six level 2 references
are fixed. If my analysis is correct, over 90% of the possible sets of
reference values can be achieved by the perceptual signals at level
3 under those conditions. (Of course the analysis might well be wrong,
but that's a different question).

The second issue for me is how the model(s) apply to living control
systems. How likely is it that in the course of a lifetime the higher
control systems will want to control every single combination of
reference values for the lower-level systems? I think it unlikely in
the extreme. Accordingly, it is of some interest to know how likely
it is that a constraint on the ability of the hierarchy to control
to all sets of reference values will matter. If the model suggests that
even very severe constraint such as fixing half of the low-level
reference values matters less than 10% of the time, even for such a
small hierarchy of such a rectangular shape, I think that's a good
thing for theoreticians and experimenters (and teachers) to know.

Furthermore, we can ask the question you hav been posing, but in reverse.
Along with Bruce Gregory (991221.1741 EST) we can ask why it is that
trouble so seldom happens when we really do put constraints on people's
low-level reference values or on their external feedback loops (which
in control theory has much the same effect). And modelling with the
spreadsheet provides a much better answer than

The spreadsheet demo shows very clearly what
we have always known; coercion (forcing a hierarchical control
system to commit to some arbitrarily selected goal or action)
sucks (in the sense that it deprives the system of some of the
control it had).

The real-world question immediately suggested by that statement is "does
the constraint affect the system of the ability to control to the
reference values wanted by systems above the constrained one(s)."
Modelling suggests that it can, but usually does not.

Simple. Elegant. And nice. If you agree with
this (as I'm sure you do) then we have no disagreement.

Simple, no doubt. Elegant--In what respect? Nice--Meaning what?

My problem with simple agreement is that although the spreadsheet
clearly shows that _some_ of the control the system had is lost,
it also shows that the probability this matters in any particular
case is very low, and therefore the relevance of this loss of control
to any real-world sitation may also be small.

Here's another question of agreement. Do you agree that what we have been
arguing about is based on the difference between your "all-or-none"
approach and my "shades of grey" approach? Or to put it another way,
given that we both acknowledge (and on my side have always acknowledged)
that there exist reference conditions at the third level that cannot be
met when a second-level reference is fixed, do you agree that the difference
is that you find it definitive that such conditions exist, whereas I
emphasise more the fact that they are very rare in a big hierarchy?

Martin

3

[From Bruce Gregory (991222.1930 EST)]

Rick Marken (991222.1440)]

Because the constraints (internal goals) are not imposed. Most
people set these "constraint" goals (like the goal of following
traffic laws) for their own (higher level) reasons. And they
control for those goals only to the extent that doing so doesn't
interfere with their ability to achieve other goals.

What makes you think that commitments are any different? That they _could
be_ any different? (Since we can not establish internal reference levels for
others?)

Bruce Gregory

4

[Martin taylor 991222 22:47]

[From Rick Marken (991222.1440)]

> given that we both acknowledge (and on my side have always
> acknowledged) that there exist reference conditions at the third
> level that cannot be met when a second-level reference is fixed,
> do you agree that the difference is that you find it definitive
> that such conditions exist, whereas I emphasise more the fact
> that they are very rare in a big hierarchy?

Not quite. From your point of view, the demo shows that any
control problem created by "committing" to any arbitrarily selected
second level reference is _small_.

I precisely didn't say that. On many occasions I have said that the
control problem might be, and I quote, "catastrophic". It is the
likelihood of _any_ problem arising that is small. Usually, there's
no problem at all.

From my point of view, the demo
shows that "committing" to any arbitrarily selected second level
reference makes it impossible for other control systems (higher
(4th) level or intrinsic systems) to use the level 3 systems to
achieve their perceptual goals.

We are back to saying three time again, are we? I have clearly demonstrated
that the spreadsheet shows that "committing" to any arbitrarily selected
second-level reference usually makes no difference in the ability of
othr higher control systems (4th level) to use the level 3 systems to
achieve their goals--but also that it can very occasionally make a
big difference.

See, I can say things three times, too. But my "three times" is based
on what the spreadsheet does.

Your point of view is that forced commitment to a goal isn't
necessarily a big deal.

I never said, implied, or suggested such a thing.

What I said was that it may be no deal at all--and in the spreadsheet
usually that's the case--or it may be a big deal. There's no in between,
so far as I have observed. Either we find control, or we find conflict
and ever-increasing output from the integrating output functions. And
when we get conflict and ever-increasing output, the lower systems
can also get screwed up.

My point of view is that your point of
view is 1) wrong (you have no idea, in any particular case,
whether the loss of control that results from forced commitment
to a goal will be a big deal or not)

I have to agree with that, because I have made that point quite forcibly
many times.

2) disrespectful (because
it assumes that you _do_ know that the loss of control will
be no big deal to the system)

If you know the structure of the hierarchy, as you can do with the
spreadsheet, you can know whether it will be a big deal or no deal at
all. With a living control system, you can't know this, but you can
know from similar dealing with the same control system on other occasions,
or with others who seem similar on the surface, whether it is likely to
be a big deal.

Incidentally, there's a difference here between the spreadsheet and the
living control system. Living control systems usually seem able to limit
the damage caused by a control problem, and control failure is often
"no big deal". That doesn't seem to happen in the spreadsheet. In the
spreadsheet, there seems to be nothing between zero problem and
runaway
explosion (catastrophe).

So, I guess I'd speculate that the spreadsheet models a living
control system correctly in showing that usually there will be no
problem at all if a mid-level reference value is committed, but
overstates the impact on a living control system when a problem
does occur.

and 3) misleading (because it
makes PCT seem complicitous in your advocacy of the judicious
use of coercion -- such as that which is involved in getting
a child to make a commitment).

I'm not going to argue this one, other than to say that sometimes one
can guarantee another control system greater overall control of its
perceptions by denying it the ability to bring one of them to its
reference level.

I'm sure you will disagree strongly with the previous paragraph. But if
you do, ask yourself how much control a child has who has been flattened
on the street, as compared with the control that the same child would
have had if it had been coerced by its mother to stay on the sidewalk.
Sometimes, Mother _does_ know best.

Also, I cannot agree that negotiation necessarily implies coercion, even
in the form of mild threat.

I think this message of yours makes it pretty clear that:

the difference [between us] is that you find it definitive
that such conditions [loss of control] exist, whereas I emphasise more
the fact that they are very rare in a big hierarchy.

Martin

5

[From Rick Marken (991222.2050)]

Me:

Most people set these "constraint" goals (like the goal of
following traffic laws) for their own (higher level) reasons.
And they control for those goals only to the extent that doing
so doesn't interfere with their ability to achieve other goals.

Bruce Gregory (991222.1930 EST) --

What makes you think that commitments are any different?
That they _could be_ any different? (Since we can not establish
internal reference levels for others?)

The spreadsheet demo was used to illustrate what would happen
if an external agent _could_ (or did, via coercion) force a
person to commit to keeping some arbitrarily selected perception
in some arbitrarily selected state (arbitrary with respect to the
person's existing hierarchy of wants). It doesn't apply to
normal hierarchical control. Control of traffic laws is normal
hierarchical control of program and logical perceptions in the
service of control of principle (self-defense, cooperation)
and system concept (citizen, driver) perceptions.

Best

Rick

···

--

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

6

[From Rick Marken (991223.0845)]

Martin Taylor (991222 22:47) --

It is the likelihood of _any_ problem arising that is small.
Usually, there's no problem at all.

I guess I just don't buy the whole idea of measuring control
problems in terms of "likelihood". My son dislocated a pinky
finger some years ago. The dislocation creates no control
problem for him in what I estimate to be over 99.99% of his
waking life. But it does affect his ability to control in some
situations that are important to him, like when he's trying to
intercept a pass in a touch football game. Therefore, this
pinky dislocation is _very_ important to him and he's now
willing to pay big bucks to get it fixed.

By your calculations, there is a very low likelihood that the
dislocation will cause any control problem; indeed, I would
estimate that the actual probability of a control problem
since the dislocation occurred (about 15 years ago) is
.0001, since my son rarely played touch football until lately
and the dislocation apparently creates no control problems
when he's playing his favorite sport, tennis. So by your
calculations, it would be "no big deal" if I had decided,
when he was a child, that I liked dislocated pinky fingers
and forced the dislocation myself. In fact, as demonstrated
by the fact that the finger was dislocated by accident, the
"unlikely" consequences of that apparently inconsequential
imposition of my will would have been quite consequential for
one of the most wonderful, kind, loving (and responsible)
people I know.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates mailto: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

7

[From Rick Marken (991223.1010)]

Bill Powers (991223.0847 MDT)--

Rick, I'm very confused and have not yet worked your spreadsheet
demo. You keep saying, flatly, that fixing any reference at an
arbitrary value (I presume you mean at the second level in your
spreadsheet) leads to loss of control (I presume you mean by one
or more third level systems).

Yes.

Martin keeps saying that he can fix any _three_ references at
the second level of the same model with NO loss of control at
the third level. Both statements can't possibly be true. What
is going on here? Is Martin lying? Are you lying? Is God lying?

I think the problem is just one of defining "control". Martin
is not lying; he _can_ fix _three_ level 2 references and
still get all level 3 perceptions to match their _existing_
references. For example, suppose that all six level 3 references
are for the perception to be true (1); so all the level three
references are 1's. Then it's easy to fix three of the level 2
references to constants so that all the level 3 perceptions are
still true. This is because (as you note) each level 3 perception
is based on only two (at most) of the level 2 perceptions. So
judiciously fixing the level 2 references will produce a set
of level 2 perceptions that allow all the level three perceptions
to match their references.

For example, here's a simple example. Here are a set of level
2 references (under the letter indicating their column in the
spreadsheet):

   D E F G H I
112.30 110.00! 7.11 5.00! 2.88 2.00!

The ! after a reference indicates that it is fixed (a constant I
entered into the spreadsheet). Since the level 2 control systems
work just fine, the level 2 perceptions are made to match these
references. So the level 2 perceptions are approximately:

  D E F G H I
112.30 110.00 7.11 5.00 2.88 2.00

The level three perceptions are just looking for ">" relationships
between adjacent perceptions. So the first system at level 3 is
perceiving whether D>E is true (where D is the level 2 perception
corresponding to the reference in D (112.3) and E is the level 2
perception corresponding to the reference in E (110.0)); the second
level 3 system is perceiving whether E>F is true; and so on to
the sixth level 3 system which is perceiving whether I>0 (zero).
As you can see, all of the level 3 perceptions are true given the
level 2 perceptions produced by the references above: D is > E,
E is > F ... and I is >0.

It is this result that led to Martin's conclusion that you can
fix as many as three level 2 references and still have control.
To Martin, control meant that the level three perceptions that
result from fixing the level 2 references continue to match the
_existing_ level 3 references (in this case, all 1's (all true)).

I am disagreeing with this conclusion because when you fix
even just _one_ level 2 reference -- _any_ level 2 reference
to _any_ value_ -- you will find that you can no longer change
the level 3 references (all 1's in the example) arbitrarily and
still get the level 3 perceptions to match those references.
For example, given the level 2 reference settings above (with
E, G and I fixed) the spreadsheet loses control of the level
3 perceptions if the level 3 references are set to:

   D E F G H I
-1.00 -1.00 -1.00 1.00 1.00 1.00

The spreadsheet can eventually get the perceptions to be:

   D E F G H I
-1.00 1.00 1.00 1.00 1.00 1.00

So system B and C can't make it. This is because the
spreadsheet can never get level 2 perceptions E<F and
F<G while keeping all the other relationships in their
reference states. The level 2 references converge to
the following:

   D E F G H I
4.12 110.00! 7.21 5.00! 2.88 2.00!

So D is not > E (as requested) but E>F and F remains

G.

So my conclusion that fixing _any_ level 2 reference
leads to loss of control is based on experiments like
this -- where I change the level 3 references and see
if the model can bring the level three perceptions
to the newly specified reference states. To me, control
means being able to produce a perception that matches
any possible reference specification; for level 3
of the hierarchy there are two possible reference
specifications for each of the six level 3 perceptions;
true (1) or false (-1). So the level three systems
are only in control if the can bring their perception
to either of the possible demanded values -- 1 or -1.

So far, every time I have tried fixing a _single_
level 2 reference at some randomly selected fixed value
I have found some combination of level 3 references
that leads to loss of control of one or more of the level
3 perceptions. When none of the level 2 references
is fixed, the model can produce level 3 perceptions
that match _any_ combination of level 3 references that
I have tried. I believe this is because the level 3
systems are controlling independent aspects of the
level 2 perceptions.

Does this help?

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates mailto: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

8

[From Bruce Gregory (991223.0707 EST)]

Rick Marken (991222.2050)

The spreadsheet demo was used to illustrate what would happen
if an external agent _could_ (or did, via coercion) force a
person to commit to keeping some arbitrarily selected perception
in some arbitrarily selected state (arbitrary with respect to the
person's existing hierarchy of wants). It doesn't apply to
normal hierarchical control. Control of traffic laws is normal
hierarchical control of program and logical perceptions in the
service of control of principle (self-defense, cooperation)
and system concept (citizen, driver) perceptions.

So there is no problem as long as the commitment is not arbitrary with
respect to the person's existing hierarchy of wants. Perhaps the teacher
could first ask, "Do you want to stay in school?" If the child says no, the
child could then be turned over to the authorities charged with dealing with
delinquency. That process seems perfectly reasonable to me.

Bruce Gregory

9

[From Bill Powers (991223.0847 MDT)]

Rick Marken (991222.1440)--

The demo says nothing about the "probability this [loss of a
control system] matters in any particular case". It shows only
that fixing any reference at an arbitrary value leads to loss
of control. It says nothing about the probability that this loss
of control will "matter" (whatever that means) or not

Rick, I'm very confused and have not yet worked your spreadsheet demo. You
keep saying, flatly, that fixing any reference at an arbitrary value (I
presume you mean at the second level in your spreadsheet) leads to loss of
control (I presume you mean by one or more third level systems). Martin
keeps saying that he can fix any _three_ references at the second level of
the same model with NO loss of control at the third level. Both statements
can't possibly be true. What is going on here? Is Martin lying? Are you
lying? Is God lying?

What I need to see, and what I'm sure a lot of other people who aren't
actually running this demo need to see, is the evidence on which each of
you is basing your statements. What reference signals do you fix, and how
do you determine the continued ability (or the inability) of a higher-level
system to control?

Since you and Martin are drawing opposite conclusions from, I assume, the
same set of observations, either you are not doing the same tests or you
have different criteria for what constitutes loss of control. I can't guess
what the explanation is. I suggest that each of you send the other a
spreadsheet modified so that one second-level reference signal is fixed,
and tell the other what you did to determine that control was or was not
lost. I trust you will be varying both reference signals and disturbances.

We can't just leave this discussion in the "'tis so, 'taint so" state. I
don't care what theory you're using; the question is _what do you do and
what happens as a result_? It is NOT POSSIBLE that you both do the same
thing to the same model and something different happens for each of you.

Best,

Bill P.

10

[Martin Taylor 991223 17:17]

[From Rick Marken (991223.0845)]

I guess I just don't buy the whole idea of measuring control
problems in terms of "likelihood". My son dislocated a pinky
finger some years ago. ... if I had decided,
when he was a child, that I liked dislocated pinky fingers
and forced the dislocation myself. In fact, as demonstrated
by the fact that the finger was dislocated by accident, the
"unlikely" consequences of that apparently inconsequential
imposition of my will would have been quite consequential for
one of the most wonderful, kind, loving (and responsible)
people I know.

I can certainly buy that point. And recognize that such things happen to
an awful lot of people--in fact, it's hard to imagine that there is
anyone who would not now be better able to control some important
perception had something different happened (or been chosen) many years
ago.

It's impossible to know for sure whether a person you interact with will
be better able to control something important later if you now choose
to impede their control (by physical or mental means) or if you now
choose to allow them to do what you see as a bad thing they are doing.
  All you can do is work on the basis of likelihood. Is it likely to
be better _for them_ if you intervene or if you leave them alone. And
remember Rabbie Burns "The best laid plans of mice and men gang aft
agley". You are likely to be wrong, whichever choice you make.

I suppose I can't come down hard on either side of this argument. Mainly
I have been arguing one side, because that side has had much less of an
airing than the one you espouse. Had the balance been the other way, I
might have taken your present position. I guess that what I really feel
is that there are occasions when impeding control of some perception by
another person is likely to help that person over time, and occasions
when it won't. Since one can never know for sure, the best one can do
is work on the balance of probabilities and the scale of the expected
consequences.

How much enjoyment has been had by adults who as children were compelled
to sit in front of a piano or blow a clarinet for hours on end? How much
less enjoyment would those people have had if their parents had employed
less coercion. For years I have regretted that I allowed my children to
stop learning musical instruments when they felt that their practice
time was cutting into other things they wanted to do, and have blessed
my parents for being stricter with me. I suspect that now they wish I
had made them keep practicing for a few more years. They, with respect
to music, are like your son with respect to his dislocated finger, They
are unable to control perceptions they now wish to control.

It cuts both ways.

If you don't like the idea of "likelihood" that's fine. All I've been
doing with the spreadsheet is attempting to find out what kind of levels
of likelihood one might expect. To some people, that's a relevant
exercise, I think.

Martin

11

[Martin Taylor 991223 17:42]

[From Bill Powers (991223.0847 MDT)]

Rick, I'm very confused and have not yet worked your spreadsheet demo. You
keep saying, flatly, that fixing any reference at an arbitrary value (I
presume you mean at the second level in your spreadsheet) leads to loss of
control (I presume you mean by one or more third level systems). Martin
keeps saying that he can fix any _three_ references at the second level of
the same model with NO loss of control at the third level.

I think we've sorted this out now.

Since you and Martin are drawing opposite conclusions from, I assume, the
same set of observations, either you are not doing the same tests or you
have different criteria for what constitutes loss of control.

The last is the answer. Rick is correct in saying that there are some
arrangements of third-level reference values that cannot be achieved
when any second-level reference is fixed. He calls this "loss of control."
I note that when I vary "randomly" the third reference values while a
second level reference is fixed (in fact three of them), the hierarchy
almost always (for a long time it was "always") has been able to bring
the third-level perceptions to match their reference values, and sometimes
do it more quickly than when the second level references can vary. I call
this "ability to control."

When we found that we were in fact observing the same thing, the topic
became "what are we arguing about" because that became the significant
question. I think what we are arguing about is the relevance of the
fact that most, but not all, logical reference values can be achieved
by the third level when several of the second-level references are fixed.
It's not the fact itself--that's not in dispute. It's the relevance of
the fact that is at issue.

For myself, but I suspect not for Rick, there is another item of interest,
which is the big difference between analogue and logical control systems
in this respect. If the third level had been controlling analogue
real-valued perceptions, there would have been NO achievable pattern
of reference values at the third level (given that no two real
numbers can
be equal except at the infinitesimal instant when they cross from A>B
to A<B). But since the third level is controlling categorical/logical
perceptions, ALMOST ALL patterns of reference values are achievable.

This result is generalizable beyond the specific spreadsheet, and I have
described some of the mathematical reasons in earlier messages. I think
it's quite important when we think about control at and above the category
level. Rick doesn't think it's important--he seems almost to think it's
a dangerous truth.

Martin

12

[From Rick Marken (991224.0930)]

Martin Taylor (991223 17:17)--

It's impossible to know for sure whether a person you
interact with will be better able to control something
important later if you now choose to impede their control
(by physical or mental means) or if you now choose to allow
them to do what you see as a bad thing they are doing.
All you can do is work on the basis of likelihood.

I would prefer to work on the basis of control theory. When
I choose to impede control (or force compliance) I do it
because what I am perceiving (my kid running into the street,
for example) differs from what I want to perceive.

How much enjoyment has been had by adults who as children
were compelled to sit in front of a piano or blow a clarinet
for hours on end?

I see your point. It makes me want to quickly point out (in
case anyone is still listening) that my spreadsheet demo of
the consequences of forcing commitment to control a particular
perception (like a commitment to not disrupt or to practice
the piano) was a demo of what _would_ happen (loss of control)
if an external agent actually _could_ force such a commitment.
The demo was meant to show what is wrong (from the perspective
of the PCT model of human nature) with adopting the _attitude_
that forcing commitments is the way to teach responsibility.

It's important to remember that it is really impossible to
force such commitments on real human beings. In real life, you
can lead a person (or horse) to a commitment but you can't make
him commit. You can't poke a reference signal value into a
person (or a horse) in the same way that you can poke such a
value into a cell of the spreadsheet. The spreadsheet model
shows what would happen if you _could_ -- and what would
happen would not be in the best interests of the control
system itself.

Parents can't really force their kids to commit to things
like practicing a musical instrument. If the kid really
doesn't want to practive he will figure out ways to deal
with the parent's prodding (which is just a disturbance
to some variables the kid is controlling). I think parents
delude themselves into thinking that they are (or would
have been, if they persisted) providing a service to their
child by forcing then to do things. I think it is good to
expose kids to things that _could_ enrich their lives (if
such things "work for them" in terms of their hierarchy of
control) but I think kids take what they want of what the
parents expose them to and leave the rest. You (Martin)
benefited from forced piano lessons because you could fit
piano playing easily into your existing hierarchy; your
kids didn't benefit from such lessons, not because you
were too lenient but because piano playing didn't fit into
their developing hierarchies of control (same with my kids,
by the way, much to my chagrin -- although my daughter is
a dynamite singer).

Again, what kids learn to control for is what works for them
in the context of their existing hierarchy of control. It's
good to expose kids to perceptual experiences that you, as a
parent, value; perceptions of responsible behavior, beautifully
performed music, hard work, etc. The problem (demonstrated
by my spreadshett hierarchy demonstration) is the _attitude_
or _belief_ that forcing people to commit to control of such
perceptions will _teach_ commitment rather than create intra
or interpersonal conflict.

Best

Rick

···

--

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

13

[From Bill Powers (991225.0153 MDT)]

Rick Marken (991223.1010)--

I think the problem is just one of defining "control". Martin
is not lying; he _can_ fix _three_ level 2 references and
still get all level 3 perceptions to match their _existing_
references. For example, suppose that all six level 3 references
are for the perception to be true (1); so all the level three
references are 1's. Then it's easy to fix three of the level 2
references to constants so that all the level 3 perceptions are
still true. This is because (as you note) each level 3 perception
is based on only two (at most) of the level 2 perceptions. So
judiciously fixing the level 2 references will produce a set
of level 2 perceptions that allow all the level three perceptions
to match their references.

"Judiciously" fixing the reference conditions at level 2 implies being able
to observe the systems individually and finding just the set of reference
conditions that will produce no conflict. Since we can't do that with a
child at the center of this discussion, we can't do it with the model, either.

It seems to me that Martin also has a point, which is that the existing
spreadsheet may not have 6 independently-variable logical conditions at
highest level. In this automated environment it should be possible to see
whether all 2^6 or 64 combinations of 6 reference signals are eventually
satisfied by the system. Martin made the point that by waiting long enough
he observed that arbitrarily fixing three 3rd-order reference signals did
not prevent the system from restoring the previous logical conditions. He
did not state that this was a stable condition -- that is, it occurs to me
to ask, if he had waited even longer, would the system have stayed in the
new state?

There is no danger of this spreadsheet system's learning anything, so it's
possible to cycle through all 64 combinations of reference conditions
simply by stepping a number systematically from 0 to 63 (000000 to 111111
in binary) and connecting each bit to one reference input at the third
level. This can be done by setting the first reference signal to the number
AND 1, the second to the number AND 2, and so on for the number AND 4, AND
8, and so on. I trust you can do AND with the spreadsheet. After each
increment of the number you would have to iterate some number of times, say
10000, to see if a stable state is reached.

The point you raise is that "free control" implies freedom to vary
reference signals, so no conclusions should be drawn if the existing state
turns out not to be disturbed by fixing particular level-2 reference
signals at particular values.

It seems to me that it is necessary to cycle through all possible
combinations of logical reference signals to verify that the existing
spreadsheet model is controlling six independent logical variables. There
is no point to basing arguments on abstract principles (which depend on
assumptions both known and overlooked) when we have the model itself in
front of us and can experiment with it.

Best,

Bill P.

14

[From Bill Powers (991225.0221 MDT)]

Martin Taylor 991223 17:17--

It's impossible to know for sure whether a person you interact with will
be better able to control something important later if you now choose
to impede their control (by physical or mental means) or if you now
choose to allow them to do what you see as a bad thing they are doing.

While this is an interesting point, it doesn't bear on the issue that gave
rise to all this: is the practice of sending a child to the RTC on the
second infraction coercive, or not? I still claim that it is, but in
claiming that I am not by any means saying that the teacher should not do
it. My primary concern is with not perpetuating lie: claiming, in front of
the children, that there is nothing coercive about what the teacher (or the
program) is doing. I'm willing to at least consider that this is a case
where no reasonable person would hesitate to coerce a child, other cases
including snatching a child out of real or possible danger. But let's at
least agree on teaching by example, at the same time, the principle of
truthfulness.

The issue we are engaged in right now is whether requiring (or forbidding)
a specific set of overt actions from a child is likely to create internal
conflict in the child, and therefore likely to result in some form of
resistance. Since requiring actions amounts to requiring the setting of
lower reference signals, it's appropriate to ask the effects of fixing an
intermediate reference signal in a hierarchy of control. Whether the higher
level is logical or any other level is irrelevant.

Telling a child "don't do that" is specifically telling the child to stop
producing an output that is being used to control some perception (since,
according to PCT, no behavior ever takes place except as the means of
controlling one or more perceptions). Unless the child has some
readily-available alternative action which will accomplish the same end
with at least equal ease, the child can't comply with the order, or
request, without giving up some degree of control at the level that was
using the offending action. Generally, at least in traditional schools I
have closely observed, the "disrupting" effect of a child's actions is no
more important than the goal of the actions: if the child passes a note to
another child in a quiet and nearly-undetectable manner, that is not an
acceptable alternative to speaking loudly to the other child to convey the
same information. The child should be paying attention to the teacher, not
trying to communicate by _any_ means with another student during class
time. One of Ed Ford's more important contributions has been to narrow the
definition of disruptive behavior so it is limited to real disruptions.
This has the advantage of interfering less with peripheral control actions
that do not actually interrupt teaching or learning. In RTP schools, I am
told in answer to enquiries, children do talk to each other during class,
but quietly and not in a way that actually disrupts anything. So one answer
to the problem of how to avoid coercion is to take a more relaxed attitude
toward the important of at least some of the rules. This is one way to
teach that obeying rules is not the highest level of human aspiration.

Best,

Bill P.

15

[From Bill Powers (991225.0255 MDT)]

Martin Taylor 991223 17:42--

For myself, but I suspect not for Rick, there is another item of interest,
which is the big difference between analogue and logical control systems
in this respect. If the third level had been controlling analogue
real-valued perceptions, there would have been NO achievable pattern
of reference values at the third level (given that no two real
numbers can
be equal except at the infinitesimal instant when they cross from A>B
to A<B). But since the third level is controlling categorical/logical
perceptions, ALMOST ALL patterns of reference values are achievable.

If control were analog, then there would be no question of having
perceptions _exactly_ equal to reference conditions. Analog control simply
keeps the perception as close as possible in magnitude to the reference
signal.

The truth of your statement that almost all patterns of reference signals
are achievable depends on what the logical perceptual functions are. Is it
easy to think of six mutually orthogonal logical functions of six
variables? Perhaps A>B, B>C, C>D, D>E, E>F, F>0 constitutes such a set, but
are there any in which _each_ function involves _all six_ variables? The
requirement for orthogonality is that _each_ perception can be true or
false without affecting the state of any of the other five perceptions. If
the last condition above were F>A, this would not be true. With 10^19
functions to choose from it might seem easy to find six orthogonal
functions, but I don't see, just at the moment, how to do it.

There is no problem with devising the second level of the spreadsheet;
practically any combination of perceptual weightings from the first level
will produce a solvable set of equations, although some sets may require
extreme outputs to achieve the solution (the "solution" being that every p
is, within some small number, the same as the corresponding r). But with a
logical system, the conditions do not allow for truth or falsity to be
within "some small range" or 1 or 0. As you point out, this makes some
conditions unsuitable for logical control -- such as "equality." Every
programmer who uses floating point numbers discovers this: it is not a good
idea to make a loop terminate on a test like (if x not-equal-to A). It is
unlikely that such a loop would ever terminate. The more precise the
calculations, the less likely it is that the condition would ever be met.

Anyway, "greater-than" is not a logical condition but a statement of
relationship. The appropriate _analog_ question is "How much greater?" The
logical condition arises when we ask whether the statement "A is greater
than B" is true or false. This kind of question is the link between the
analog world (the actual values of A and B, and the subtraction of one
magnitude from another) and the digital world (the categorical existence of
a state of affairs). The magnitude relationship between analog variables A
and B is a magnitude, the difference, and as such is controllable by an
analog control system (as in keeping the cursor one inch to the left of the
target). To create a logical variable we must go through an input function
that has a threshold (with or without hysteresis). It is this physical
device that determines whether there will or will not be a large signal in
the logical channel. And from then on, the logical systems can work
digitally, with no values of any variables between 0 and 1. The only
operations needed are AND, NOT, and OR. I'm not sure there really are any
such variables in the brain, but we can pretend there are just to see what
happens.

Having worked my way through all that, I am led to wonder whether the third
level in Rick's spreadsheet is really a logical level. The perception that
is cast as A > B can also be expressed as A - B, with a positive reference
level. It is subtraction, an analog operation, not logic, by which we
determine whether A is or is not greater than B. By specifying a positive
reference level for this difference, we are saying we want A to be greater
by some amount than B. If we set the same reference signal to a negative
number, we're saying that we want A to be less than B by some amount. If we
make the gain very high and give the output function limits that will be
reached for even very small errors, we will have what looks like a logical
device, but which actually remains a crude (bang-bang) analog system
controlling a relationship (not very well). This would still be true,
although harder to grasp, if the input function also saturated for small
differences between A and B. It would then seem to work like a logical
function, except that the basic input operation would still be the
subtraction of one magnitude from the other -- which is not a logical
function. In a true logical function, variables have no magnitudes other
than 1 or 0. Today is Tuesday AND is it 7:00 PM. That's a logical statement
that does not involve analog comparisons: it's a statement in which the
elements are propositions that can only be true or false.

Best,

Bill P.

16

[From Rick Marken (991225.0800)]

Bill Powers (991225.0153 MDT)--

It seems to me that it is necessary to cycle through all
possible combinations of logical reference signals to verify
that the existing spreadsheet model is controlling six
independent logical variables. There is no point to basing
arguments on abstract principles (which depend on assumptions
both known and overlooked) when we have the model itself in
front of us and can experiment with it.

Right. I'll have to write a macro to do this. I'm pretty
confident that the six level 3 perceptions are independent
variables (I've tried about 1/5 th of the 64 possibilities).
But I should do this systematcially and exhaustively. I'll
try to have the answer before the new millennium.

Bill Powers (991225.0221 MDT) --

Great post.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

17

[From Rick Marken (991227.1610)]

Bill Powers (991225.0153 MDT)--

It seems to me that Martin also has a point, which is that
the existing spreadsheet may not have 6 independently-variable
logical conditions at highest level...

It seems to me that it is necessary to cycle through all
possible combinations of logical reference signals to verify
that the existing spreadsheet model is controlling six
independent logical variables.

I have now demonstrated that the spreadshett is controlling
six independent level 3 variables (whether they ashould be
called "logical" or not is no big deal to me). I wrote the macro
that cycles through all 2^6 (64) possible combinations of level
3 references. The level 3 perceptions can be brought to any
of the 64 combination of level 3 references. I re-ran the macro
(which takes about 15 minutes on my machine at home) with one
level 2 reference fixed (committed). With this level 2 reference
fixed, the spreadsheet was unable to bring the level 3 perceptions
into a match with 4 of the 64 possible combinations of level 3
references. So with respect to the issue we have been debating,
which, as you say [Bill Powers (991225.0221 MDT)], is "whether
requiring (or forbidding) a specific set of overt actions from
a child is likely to create internal conflict in the child, and
therefore likely to result in some form of resistance" the
answer from the hierarchcal spreadsheet model is (as I said
when this discussion began) "yes".

I will send the spreadsheet with the testing macro to
anyone who wants it (I don't think I want to put it up
at my website -- too specialized).

Best

Rick

···

---
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

18

[From Bill Powers (991228.0726 MDT)]

Rick Marken (991227.1610)--

Bill Powers (991225.0153 MDT)--

It seems to me that Martin also has a point, which is that
the existing spreadsheet may not have 6 independently-variable
logical conditions at highest level...

It seems to me that it is necessary to cycle through all
possible combinations of logical reference signals to verify
that the existing spreadsheet model is controlling six
independent logical variables.

I have now demonstrated that the spreadshett is controlling
six independent level 3 variables (whether they should be
called "logical" or not is no big deal to me). I wrote the macro
that cycles through all 2^6 (64) possible combinations of level
3 references. The level 3 perceptions can be brought to any
of the 64 combinations of level 3 references. I re-ran the macro
(which takes about 15 minutes on my machine at home) with one
level 2 reference fixed (committed). With this level 2 reference
fixed, the spreadsheet was unable to bring the level 3 perceptions
into a match with 4 of the 64 possible combinations of level 3
references.

Excellent. However, let's give credit where it's due. It seems to me that
Martin predicted that with a single fixed second-level reference signal,
the third-level systems would be able to continue controlling in 15/16 of
the cases, or in other words all but 4 of 64.

So with respect to the issue we have been debating,
which, as you say [Bill Powers (991225.0221 MDT)], is "whether
requiring (or forbidding) a specific set of overt actions from
a child is likely to create internal conflict in the child, and
therefore likely to result in some form of resistance" the
answer from the hierarchcal spreadsheet model is (as I said
when this discussion began) "yes".

No, it is "no" if you mean by the word "likely" what other people mean.
Chances are 15 out of 16 that no internal conflict would be created, which
makes _that_ the most "likely" case.

I think we all agree that the spreadsheet program is only a first
approximation to the real case. However, support for it will quite likely
increase now that it has proven someone _else_ right. Martin conjectured
that the more systems that are involved at a given level, the more robust
the hierarchy will be against loss of a single intermediate-level control
system. I think your test vindicates his claim. Try it with 12 systems at
each level and see what happens!

I would not yet conclude from these simulation-experiments that causing
children to adopt arbitrary reference levels (assuming that is possible)
would be harmless to the child. But we have to conclude that in the
spreadsheet world, it is not expecially harmful unless the minority of
cases in which a problem does occur are important to the child.

Best,

Bill P.