[Martin Taylor 2010.08.03.16.43]
[From Bruce Gregory (2010.08.03.1433 EDT)]
[Martin Taylor 2010.08.03.11.33]
I'd say PCT is more about "why" than about "how". The reference signal is "why" a control system acts. The control system's output system doesn't care how it influences the perception, only that it does, and does so in a direction to reduce the discrepancy between the reference vale and the perceptual value. A PCT analyst may want to figure out just how this influence occurs, to determine its bandwidth, the loop gain and delay, and so forth, but the control unit has no access to that kind of information. "Why" the reference value is thus and so is that the perception being controlled is going to be used somewhere in some other control, whether higher in the hierarchy (HPCT) or because of its influence on some other variable through the environment.
That "why" seems to me to be pretty empty. Or perhaps I should say that explains too much. No matter what I do, I do in order to satisfy some some control loop higher in the hierarchy or because of the influence of my actions on some other variable through the environment. That's rather like saying that any activity in the universe is the result of the interaction of quantum fields. O.K., now what?
Think of the child's hierarchy of questions "Q. Why is the sky blue"
“A. Because blue light scatters more from the air than does read or
green light” “Q. Why does that happen” “A. Because blue has a
shorter wavelength than red or green” “Q. Why does blue have a
shorter wavelength” “A. It’s the way our eyes work. We call it blue
when we see shorter wavelength light, and red when we see longer
wavelength light” “Q. Why do we call it those names” “A. It’s
convenient. We could call it anything we wanted, so long as other
people could understand us” “Q. Why …” (and so on). “Why” is
fundamental, and in PCT, the answer to “Q. Why do you do that” is
always the same as the everyday world answer “A. I do that in order
to achieve this”. The difference from the everyday answer is that
the PCT view always carries the implication of “could you achieve
this another way” and “by doing it this way, what side effects might
there be” (the “unintended consequences” of economics). And, in most
cases, the PCT view also suggests that the stopping point of the
“why” series is likely to be “because that’s what I believe to be
right” (a controlled perception at a rather high level in Bill’s
proposed hierarchy).
I don't buy your quantum analogy. A better one might come from
Newton’s laws of mechanics. For example: (Newton) To every action
there is an equal and opposite reaction; (PCT) A disturbance to a
controlled perception results in a compensating action. (Newton) A
body continues its state of motion unless affected by external
forces. (PCT) An undisturbed perception maintains its value unless
its reference value changes.
The point is that from almost trivially true elements, a complex set
of consequences can be derived by introducing specific boundary
conditions. The PCT “why” is indeed universal for living things, and
it is trivially simple. By itself, as you say, it doesn’t explain
too much. Even without PCT, people ask each other “Why did you do
that?” An understanding of PCT doesn’t change that any more than an
understanding of Newton’s laws change people’s understanding that if
you kick a stone it moves in the direction of the kick.
Understanding Newton’s laws allows you to consider the forces
involved in the kick. Understanding PCT induces one to ask “If this
is why you did that, could you have got as good a result by doing
something else?” In PCT-speak, one might ask what perceptions enter
into the one being controlled, and how the actions involved in
controlling those other perceptions might interact with (possibly
conflict with) the control of yet other perceptions being controlled
by yourself or by other people. In this sense, PCT formalizes the
kinds of questions you might ask.
Rick reduces the questions you might ask to just one: "What
perceptions are being controlled". I disagree with that limitation,
but I do think that it is useful to be able to categorize the kinds
of questions, and more importantly to be able to say what it would
mean to answer them. Looking at just the simple circuitry of a
single control unit, you can ask a question about each individual
component. Rick’s question turns into “what is the perceptual input
function”. To that question we can add others, of which the more
important might include “what is the reference value for that
perception”, “What influences would disturb that perception”, "What
are the loop characteristics (e.g. gain and transport lag), “what is
the environmental feedback path”, “What side effects might occur”
(which in part can be translated as “how efficient is the control”,
since if all the output energy is used only to influence the
controlled perception, none is left for side effects; of course,
that never happens in practice, but to set up that ideal leads to
consideration of how to control one’s perceptions cleanly, reducing
unexpected conflicts).
Let's look at a simple experiment in behavioral economics. It has been found that a higher percentage of employees enroll in a retirement plan if they must opt out than if they must opt in. The lesson here is clear. If you want more of your employees to participate in a retirement plan, make it a requirement that they must opt out if they do not wish to participate. The simplest interpretation of these findings is that employees need a reason to act. Absent such a reason, the do not act by opting out. What, if anything, does PCT have to add to this story? I can't think of anything, but I am admittedly incompetent.
I mentioned "tolerance", but I think you misunderstood...
There's a whole PCT science in just the analysis of this kind of group interaction. For example, what is the influence of tolerance? Tolerance has a specific PCT meaning -- a level of difference between the reference and the actual value of some controlled perception that is reported as "zero error" to the output function -- but its implication is the same as "tolerance" in everyday life.
BG: Really? Now if we only knew how to report zero-error to the output function, look how many of problems of the world would disappear. The Middle East mess would vanish. I can see myself explaining it to Netanyahu, "Look Ben (we are on a first-name basis), its all a matter of getting your systems to report a "zero error" to your output functions. I've explained this to the Palestinians and they are on board with the plan."
I'm not sure in what way you misunderstood, but in everyday
language, if Netanyahu could tolerate more the presence of
Palestinians in association with Jews and perceived Palestinians as
tolerating more the existence and legitimacy of Israel, would not
conflict be reduced? That’s just an everyday intuitive statement.
The PCT concept would be best described in diagrams:
![tolerance.jpg|711x321](upload://dEHtvVF5ggxMN1W4flNlSsfcyc1.jpeg)
The left diagram shows the output from the comparator function as it
is usually considered (a simple subtractor), whereas the right one
shows the output from a comparator that reports zero error for small
deviations between reference and perception – a “tolerant”
comparator function. If a control system has such a comparator
function, it generates no change in output when its controlled
perception is slightly disturbed. If two such “tolerant” control
units are in apparent conflict, their conflict will not escalate
when either’s perception is disturbed by small extraneous
influences, whereas with “zero tolerance” control systems in
conflict, any extraneous disturbance, no matter how small, will
result in exponentially increasing output, each control unit’s
output being a disturbance to the other’s perception.
What is the influence of timing? If the effects of one action are slow to reach the perceptual inputs of another group member, are the group-level effects qualitatively different than if the communications are faster? How do those timings interact with the time-scales of control at the higher levels within individuals? Would the financial melt-down of 2008 have occurred if it took days or weeks for communications between institutions or between branches in different countries?
BG: The answer to your question is known even to those who do not understand PCT. If we reinstated the pony express financial collapses would take longer to develop. But develop they would: _This Time is Different: Eight Centuries of Financial Folly_ by Reinhart and Rogoff.
I've heard of the book, but not read it. Does it explain the cases
when financial collapses did not happen even though the conditions
might seems similar to those cases when they did? Remember that the
2008 collapse just might not have happened if Lehman Bros had been
bailed out that critical weekend. Since we can’t rerun the past, we
won’t know, but some people think we might have avoided or at least
mitigated the collapse.
BG: I seems to me that MOL-type interviews are characterized by exactly the same sort of fuzziness that Bill deplores in conventional psychology.
Interesting. Maybe you could describe what you understand to be an
MOL-type interview, and let Bill comment on the accuracy of your
description?
In fact studies demonstrate when they are led to believe that they chose an option that they did not express preference for, many subjects have no trouble justifying the choice they did not make. (Asked to choose between A and B, they indicated a preference for A but were given B. They later generated a variety of reasons why B was superior)
Yes, that's true. Quite interesting are the cases in which it is not
true. What’s different about those people, or is it just those
people in those specific situations?
My own PCT-type top-of-the-head suggested explanation for this
effect (Rick will disagree) is that one of the perceptions people
control for is self-image, a perception with many contributing
perceptions, among which may well be a perception of one’s own
competence. If that perception is controlled, then a perception that
choice A was in some way wrong acts as a disturbance, a disturbance
that can be countered in imagination by developing reasons
(perceptions of, among other things, environmental affordances) that
they would now choose B instead.
We learn from the time we are young children the importance of justifying our actions. If I was you a series of "non-directive" questions, you will quickly learn what kinds of answers please me and provide them (if pleasing me is important).
Yes. Reorganization will do that. In fact, it is the basis for my
theory of language and cultural acquisition that I presented at the
CSG conference in Durango, 1993.
Despite Bill's enthusiasm for introspection, I know of no evidence that it is anything more than story-telling.
I think Bill would probably agree with that, in the absence of
experimental data. He has often said that his hierarchical levels
are “story-telling” about his own introspections, and very recently
he said that one of his hopes for the future development of PCT
would be to test the very concept of levels, let alone the nature of
different levels. He certainly is quite free with the levels when
making his own models.
I am confident that students will tell you why they made the bids they made. I have no confidence that their answers tell us anything more than that the students believe they are giving plausible reasons for their actions.
Yes. But one might reasonably ask why they feel the need to do that.
What perception are they controlling in doing so? How might one
disturb that perception in a way that would not disturb some
plausible other candidate perception? In other words, could you
perform “The Test” to do so?
For some insight into this are of research, you might check out Dick
Robertson et al (et al includes Dave Goldstein on this list)
“Testing the Self as a control system: Theoretical and
Methodological Issues” in the PCT issue of the International Journal
of Human Computer Studies (1999, v50), or directly ask Roberston or
Goldstein on this list.
The viewpoints are not compatible except in the sense that the analyst's viewpoint will include all the observations available from the observer's viewpoint. Neither is wrong, but the analyst's viewpoint, if the analysis is correct, is more powerful. If the analysis can predict the observations, the model becomes more credible, but even a post-hoc analysis can have some value.
BG: Unfortunately, I find this distinction of limited value in the real world. If a model does not predict, it can "explain" anything.
I would rephrase that to say that after the fact, it is always
possible to produce a model that will explain whatever was observed.
It is not true that a pre-formed model can explain all possible
data. With a pre-formed model, all you have to work with are the
model parameters. If you can fit the data by varying the parameters,
the data don’t disconfirm the model. It could have predicted if you
had known the correct parameters. But you can’t always fit data by
parameter variation within a model, and if you can’t then the model
is wrong. If you pre-set the parameters, you predict, but then if
the data don’t fit, you don’t know whether it’s the model that’s
wrong or the parameter values.
Only once in my professional life have I fit someone else's
experimental data using my model with parameter values from yet
other people’s experiments on different perceptions (Numerical
prediction of a simple figural aftereffect as a function of the
contrast of the inspection figure. Psychol. Rev., 1963, 70,
357-360). It’s quite satisfying, but very rare, to be able to do
that. PCT at least offers the prospect of being able to model the
control of one perception and then using some parameter of that
model in predicting the behaviour in controlling another perception.
I feel certain that those who favor "intelligent design" can provide a satisfying, to them at least, explanation of anything we can discover in the natural world.
They seem to have that ability, but it's not difficult, since their
model is universal and parameter-unbounded.
I guess I am just an observer rather than an analyst. I predict that if you give your employee's the option of opting out of your retirement program, more of them will join the program than if you require them to opt into the program.
Yes. That's a prediction you can get from the tolerance model
without needing numerical parameters. All it requires is that the
tolerance bound for a person with a reference not to be in the
program is greater than zero, since that would result in no
“opt-out” action if the error is small enough – as it might be for
some people. If the option is “opt-in” a zero-tolerance control
system would result in only those people with a reference to be in
the program acting, and a non-zero tolerance would result in fewer
of those acting. Either way, if the relevant control systems have
non–zero tolerance, you get the result you predict (from
intuition?).
I would be impressed if your analytical approach identified which employees would enter the program before I asked them.
I would not expect to identify all of them correctly, because I
don’t know of a way of assessing tolerance bounds (which might even
depend on whether the person had a good breakfast), but if one used
MOL to determine what perceptions entered into their perception of
the retirement program, and asked about their reference level for
(or if they controlled for) those perceptions, one might be able to
guess fairly well.
After the fact, I fear I would have to agree with Pooh-Bah. "Merely corroborative detail, intended to give artistic verisimilitude to an otherwise bald and unconvincing narrative."
I think you have to use that approach more selectively than you seem
to do. Ask whether the explanation could have been available in
large part beforehand, whether what is not predicted is a matter of
parameter variation or model structure, and if parameter variation,
could the parameters have been estimated from other data gathered
independently of those needed to fit the model. The Pooh-Bah
approach is justified if you have to devise a new model to fit the
data, or if there is no way in principle to find the fitting
parameters other than by looking at the data to be fit.
Anyway, it's important to keep track of what viewpoint you are using when discussing observations, and the viewpoint can differ between group and individual views of the same phenomenon.
BG: O.K. Put me down as an observer waiting to be convinced that the analyst has something more valuable to add to the story. (I will not dwell on the value added to the financial disaster by the quants of Wall Street.)
Fair enough. The observer's observations are always welcome, but
technical criticism (an analyst’s tool) is also welcome.
Martin