Searching for controlled variable

From Phil Runkel on 25 July 98:

        Rick Marken and Bruce Abbott were recently exchanging
remarks about demonstrating control (I guess). Abbott (if I
remember correctly -- these thoughts came to me after I had
discarded the last batch of e-mail) said that many experiments of
the traditional sort located a controlled variable by showing
that a person preferred one object or dimension or activity or
something to another. There is indeed a large literature
studying preferences, and a fairly large literature investigating
the dimensions underlying preferences -- a lot of the work using
factor analysis and nonmetric scaling falls there.

        One can talk about the topic Marken and Abbott were talking
about in the language of that literature. Imagine two orthogonal
axes X and Y and a line slanting across them with a positive
slope. Suppose there is an object A lying some distance out on
that line and another object B lying closer in. Obviously, B is
closer to X than A is, and B is also also closer to Y than A is.
If the person chooses B, you know that the person preferred B to
A -- at least at the moment of choice. You may conclude (if your
hypothesis coaxes you into doing so) that the person made the
choice because of a judgment based on X (a controlled variable),
while the person could equally well have made it on the basis of
Y.

        This kind of reasoning and geometry was beautifully
elaborated by Clyde Coombs in his "Theory of Data" (1964),
especially chapters 3 through 9. Runkel and McGrath wrote about
it in chapters 11 and 12 of "Research on Human Behavior" (1972).
I have not kept up with the recent studies.

[From Bill Powers (980726.0530 MDT)]

From Phil Runkel on 25 July 98:

Suppose there is an object A lying some distance out on
that line and another object B lying closer in. Obviously, B is
closer to X than A is, and B is also also closer to Y than A is.
If the person chooses B, you know that the person preferred B to
A -- at least at the moment of choice. You may conclude (if your
hypothesis coaxes you into doing so) that the person made the
choice because of a judgment based on X (a controlled variable),
while the person could equally well have made it on the basis of
Y.

I don't know if the references you cite mention this, but the preference
phenomenon you talk about can easily lead to a paradox: preferring B to A,
C to B, and A to C. This can happen if different axes are used for each
comparison: that is, I prefer apples to oranges because I like their color
better, oranges to lemons because oranges are sweeter, and lemons to apples
because they make better lemonade. Something similar can happen with
respect to voter preferences.

Preferences imply conflict, because you like both items but for some reason
the experimenter won't let you choose both of them. The basis for a
preference, it seems to me, is always a higher-order perception which can
be controlled by manipulating either of the items involved. The two axes
Runkel mentions are just the two weights assigned to the two lower-order
inputs by each of two control systems. In principle the same argument
applies to multiple (more than 2) dimensions.

Best,

Bill P.

[From Bruce Abbott (980726.1720 EST)]

Bill Powers (980726.0530 MDT) --

Preferences imply conflict, because you like both items but for some reason
the experimenter won't let you choose both of them.

That's not true in general because there are other possibilities. You may
like one option and be indifferent to the other; you may dislike both; you
may be indifferent to both. Even when you like both items, you may like one
of them more, in which case there will be no conflict.

Preferences still exist in the absence of constraint imposed by an
experimenter. I think you meant choice, not preference. Choice may imply
conflict, when one would prefer to have both but cannot.

This thread seems to have started in reply to my assertion that the choice
experiments which I described provide an example of a Test for a controlled
variable. I hope it's not intended as an argument against that assertion.
Is it?

Regards,

Bruce

[From Bill Powers (980727.0539 MDT)]

Bruce Abbott (980726.1720 EST)--

I said:

Preferences imply conflict, because you like both items but for some reason
the experimenter won't let you choose both of them.

That's not true in general because there are other possibilities. You may
like one option and be indifferent to the other; you may dislike both; you
may be indifferent to both. Even when you like both items, you may like one
of them more, in which case there will be no conflict.

I find it hard to understand what you mean by preference that remains the
same under all these conditions -- especially since you distinguish between
preference and choosing. How do I have a preference between two items to
both of which I am indifferent?

Conflict (in PCT) simply means simultaneously desiring two different things
that you can't have at the same time. Liking more or less would be measured
by the relative loop gains, or by the relative settings of the reference
levels. Obtaining the item you like more still leaves the desire for the
other unsatisfied. Are you proposing that this is a state of no conflict?

I guess you're right about the rest. So this phenomenon is given the same
name under all the possible conditions you mention, as well as the conflict
condition. What is it that makes "preference" the same under all these
conditions?

It occurs to me that under behaviorism, the act of indicating a choice
might be the common element, since under behaviorism one doesn't talk about
inner states (i.e., the state of having a preference even before one
behaves in a way that reveals it).

Preferences still exist in the absence of constraint imposed by an
experimenter. I think you meant choice, not preference. Choice may imply
conflict, when one would prefer to have both but cannot.

Is "both" an acceptable answer when one is asked about a preference
involving two items?

This thread seems to have started in reply to my assertion that the choice
experiments which I described provide an example of a Test for a controlled
variable. I hope it's not intended as an argument against that assertion.
Is it?

Does a test for preference involve all the elements of the Test? My
recollection is that preference tests mostly involve asking the person
which of two items would be preferred, and accepting the action that
indicates the preference (uttering the name of the item, or checking a box,
or pointing, and so on). That is not testing for a controlled variable.

You distinguish between a test for preference and a test for choosing, but
it seems to me that it would be difficult to discover a preference unless
the person performed some action indicating one item but not any other.

Anyway, as always seems to happen when we talk about concepts from
conventional psychology, here we are arguing about the meanings of words
again. How boring.

Best,

Bill P.

[From Bruce Abbott (980727.1335 EST)]

Bill Powers (980727.0539 MDT) --

Bruce Abbott (980726.1720 EST)

I find it hard to understand what you mean by preference that remains the
same under all these conditions -- especially since you distinguish between
preference and choosing. How do I have a preference between two items to
both of which I am indifferent?

In the same way that you can have a reference of zero intensity. Preference
is a continuous variable, which may have a value of zero, i.e., no preference.

Conflict (in PCT) simply means simultaneously desiring two different things
that you can't have at the same time.

So if you simultaneously desire NOT having either of two different things
when you must experience one of them, that is not conflict in PCT, correct?
When the choice is either the frying pan or the fire, there can be no conflict?

I guess you're right about the rest. So this phenomenon is given the same
name under all the possible conditions you mention, as well as the conflict
condition. What is it that makes "preference" the same under all these
conditions?

Relative evaluation of the alternatives. A continuum of value, ranging from
negative through zero to positive, with alternatives defined along that
continuum. Alternatives higher on the scale are preferred over those lower
on the scale.

It occurs to me that under behaviorism, the act of indicating a choice
might be the common element, since under behaviorism one doesn't talk about
inner states (i.e., the state of having a preference even before one
behaves in a way that reveals it).

Choice is a way to assess preference. There are others (e.g., ranking).

Preferences still exist in the absence of constraint imposed by an
experimenter. I think you meant choice, not preference. Choice may imply
conflict, when one would prefer to have both but cannot.

Is "both" an acceptable answer when one is asked about a preference
involving two items?

No, but "both" would indicate that the two items are both evaluated at the
positive end of the value scale. "Neither" would indicate that the two
items are both evaluated at the negative end of the value scale. But if
they are equally valued, one would not have a preference _between_ them.

Does a test for preference involve all the elements of the Test? My
recollection is that preference tests mostly involve asking the person
which of two items would be preferred, and accepting the action that
indicates the preference (uttering the name of the item, or checking a box,
or pointing, and so on). That is not testing for a controlled variable.

That is not the preference test that I described. How about addressing the
case in point?

You distinguish between a test for preference and a test for choosing, but
it seems to me that it would be difficult to discover a preference unless
the person performed some action indicating one item but not any other.

You are confused. Choice is a method for assessing preference. Choice is a
behavior; preference is a state of mind.

Anyway, as always seems to happen when we talk about concepts from
conventional psychology, here we are arguing about the meanings of words
again. How boring.

It seems to me that preference (result of an evaluative process) and choice
(a behavior under certain constraints) offer no definitional problems. If
we are arguing about definitions, I would argue that it is because you have
not thought clearly about the distinction.

Is there a place for an evaluative process in PCT?

Regards,

Bruce the boring

[From Bruce Abbott (980727.1530 EST)]

Bill Powers (980727.1348 MDT) --

I guess I shouldn't even try to comment on subjects in
which I have so little interest.

I asked whether an evaluative process has a place in PCT. Is this one of
the subjects on which your interest is lacking? Or does your silence on the
matter simply mean that your answer is no? Please disambiguate.

Regards,

Bruce

[From Bill Powers (980727.1348 MDT)]

Bruce Abbott (980727.1335 EST)--

So if you simultaneously desire NOT having either of two different things
when you must experience one of them ...

Relative evaluation of the alternatives. A continuum of value ...

Choice is a way to assess preference ...

No, but "both" would indicate ...

That is not the preference test that I described...

It seems to me that preference (result of an evaluative process) and choice
(a behavior under certain constraints) offer no definitional problems. If
we are arguing about definitions, I would argue that it is because you have
not thought clearly about the distinction.

Right, I haven't. I guess I shouldn't even try to comment on subjects in
which I have so little interest.

Best,

Bill P.

[From Bill Powers (980727.1630 MDT)]

Bruce Abbott (980727.1530 EST)--

I asked whether an evaluative process has a place in PCT. Is this one of
the subjects on which your interest is lacking? Or does your silence on the
matter simply mean that your answer is no? Please disambiguate.

I don't even know what you mean by an "evaluative process," so I can't
answer your question. Is this something that can be handled within one of
the proposed HPCT levels, or is it a process that needs to be added to the
model?

Best,

Bill P.

[From Bruce Abbott (980727.2015 EST)]

Bill Powers (980727.1630 MDT) --

Bruce Abbott (980727.1530 EST)

I asked whether an evaluative process has a place in PCT. Is this one of
the subjects on which your interest is lacking? Or does your silence on the
matter simply mean that your answer is no? Please disambiguate.

I don't even know what you mean by an "evaluative process," so I can't
answer your question. Is this something that can be handled within one of
the proposed HPCT levels, or is it a process that needs to be added to the
model?

It is the assignment of value by the individual to specific perceptions. I
don't know how (or if) it meshes with HPCT; that's what I'm trying to sort
out. You like this a little, you like that a lot (but only when served ice
cold), you dislike that other stuff over there; you'd happily throw yourself
over a high cliff for an approving smile from Lolita. It seems to apply
across levels, so I suspect that it could not be handled within some
particular HPCT level. It seems related to the concept of reference values,
but I don't see a direct correspondence.

Regards,

Bruce

[From Bruce Abbott (980727.1335 EST)]

Bill Powers (980727.0539 MDT) --

It seems to me that preference (result of an evaluative process) and choice
(a behavior under certain constraints) offer no definitional problems. If
we are arguing about definitions, I would argue that it is because you have
not thought clearly about the distinction.

Is there a place for an evaluative process in PCT?

It appeared to me that what were being chosen, preferred and evaluated in
your examples (which I only skimmed) were usually _outputs_ of living
control systems. If so, the study of outputs has more to do with pre-PCT
psychologies than with PCT.

You are not talking about choosing, preferring and evaluating _controlled
variables_, are you? If so, aren't preference, choice and evaluation
simply part of the routine operation of hierarchical control systems, and
so do not require special names and definitions?

Sincerely, Hank Folson

[From Bill Powers (980728.0821 MDT)]

Bruce Abbott (980727.2015 EST)]

I don't even know what you mean by an "evaluative process,"

It is the assignment of value by the individual to specific perceptions. I
don't know how (or if) it meshes with HPCT; that's what I'm trying to sort
out. You like this a little, you like that a lot (but only when served ice
cold), you dislike that other stuff over there; you'd happily throw yourself
over a high cliff for an approving smile from Lolita. It seems to apply
across levels, so I suspect that it could not be handled within some
particular HPCT level. It seems related to the concept of reference values,
but I don't see a direct correspondence.

I think it is related to the concept of reference values, as you say. A
perception by itself has no value; it is simply a report on the apparent
state of the world (and body). Some particular state (the reference level)
is defined as the state to be attained by behavior if not already in that
state. In general, reference values are set by higher systems, or where
relevant by the intrinsic system. It is the intrinsic system that defines
pleasure as good and pain as bad. Learned perceptions are given values as a
result of the reorganizing process, so in the final analysis the values of
learned perceptions are derived from intrinsic reference levels. Good and
bad perceptions are defined in terms of the amount and direction of action
that results from error. We call a perception good if we will behave to
increase the amount of it or to maintain a positive level of it. A
perception is bad if we will behave to reduce the amount of it. It's
possible that experiences of good and bad perceptions are also influenced
by rates of change.

Intrinsic variables are judged relative to inborn unlearned intrinsic
reference signals. Perceptions in the learned hierarchy are judged relative
to adjustable reference signals. "Values" in the learned hierarchy are thus
subject to change, even on a rather fast time scale. Water that feels "too
hot" may feel pleasantly warm 5 minutes later even though its temperature
has not changed. But pain always hurts, although one can learn, like G.
Gordon Liddy, "not minding" the hurt -- i.e., not trying to do anything to
reduce it.

That's how it seems to work out on July 28, 1998.

Best,

Bill P.

[From Bruce Abbott (980728.1230 EST)]

Hank Folson --

Bruce Abbott (980727.1335 EST)

It seems to me that preference (result of an evaluative process) and choice
(a behavior under certain constraints) offer no definitional problems. If
we are arguing about definitions, I would argue that it is because you have
not thought clearly about the distinction.

Is there a place for an evaluative process in PCT?

It appeared to me that what were being chosen, preferred and evaluated in
your examples (which I only skimmed) were usually _outputs_ of living
control systems. If so, the study of outputs has more to do with pre-PCT
psychologies than with PCT.

I'm really not following you here. Please elaborate.

You are not talking about choosing, preferring and evaluating _controlled
variables_, are you? If so, aren't preference, choice and evaluation
simply part of the routine operation of hierarchical control systems, and
so do not require special names and definitions?

I don't know. Are they? Please explain.

Regards,

Bruce

[From Bruce Abbott (980728.1330 EST)]

Bill Powers (980728.0821 MDT) --

Bruce Abbott (980727.2015 EST)

It is the assignment of value by the individual to specific perceptions. I
don't know how (or if) it meshes with HPCT; that's what I'm trying to sort
out. You like this a little, you like that a lot (but only when served ice
cold), you dislike that other stuff over there; you'd happily throw yourself
over a high cliff for an approving smile from Lolita. It seems to apply
across levels, so I suspect that it could not be handled within some
particular HPCT level. It seems related to the concept of reference values,
but I don't see a direct correspondence.

I think it is related to the concept of reference values, as you say. A
perception by itself has no value; it is simply a report on the apparent
state of the world (and body).

Certainly; perception is just perception. That's why I wrote of the
individual "assigning" values to perceptions.

Some particular state (the reference level)
is defined as the state to be attained by behavior if not already in that
state. In general, reference values are set by higher systems, or where
relevant by the intrinsic system. It is the intrinsic system that defines
pleasure as good and pain as bad.

Is pleasure a perception separate from the variable being perceived? What
is pleasure?

Good and
bad perceptions are defined in terms of the amount and direction of action
that results from error. We call a perception good if we will behave to
increase the amount of it or to maintain a positive level of it. A
perception is bad if we will behave to reduce the amount of it.

This almost exactly parallels Thorndike's definitions of satisfiers and
dissatisfiers. Also, you could substitute "pleasurable" and
"displeasurable" for "good" and "bad" and the statement would still sound
right. Are "good" and "bad" simply our higher-level conscious evaluations
of a more fundamental pleasure/displeasure dimension?

That's how it seems to work out on July 28, 1998.

A clear and helpful discussion. Thanks.

Regards,

Bruce

[From Bill Powers (980728.1622 MDT)]

Bruce Abbott (980728.1330 EST)--

Is pleasure a perception separate from the variable being perceived? What
is pleasure?

I don't know. There does seem to be a "pleasure center" which, when
stimulated, leads to behavior that raises the level of stimulation
indefinitely. I have no idea what part of the control loop is being
stimulated, but it apparently leads to an error that can't be corrected no
matter how much more stimulation occurs. The (electrical) stimulation could
actually be _suppressing_ a perceptual feedback signal, so it's not a
certainty that "pleasure center" is an appropriate description.

Good and
bad perceptions are defined in terms of the amount and direction of action
that results from error. We call a perception good if we will behave to
increase the amount of it or to maintain a positive level of it. A
perception is bad if we will behave to reduce the amount of it.

This almost exactly parallels Thorndike's definitions of satisfiers and
dissatisfiers.

Oh.

Also, you could substitute "pleasurable" and
"displeasurable" for "good" and "bad" and the statement would still sound
right. Are "good" and "bad" simply our higher-level conscious evaluations
of a more fundamental pleasure/displeasure dimension?

I think they are _produced_ by the underlying processes of reorganization,
which are directed by built-in reference signals that define "good" or
"right" states of intrinsic variables.
Best,

Bill P.