[From Bill Powers (971224.1546 MST)]
Bruce Abbott (971224.1540 EST)--
I can wait to see the actual numbers. Take your time.
Nonsense, for the reasons given in the part of my post you chose NOT to
quote.
Well, you said the first 7 generalizations were drawn from Ebbinghaus'
obserations of himself. Did his generalizations apply to the results of
every trial? Did he have to average over trials for any of them? When they
have been "replicated" by others, how many of the subjects showed the same
effects, and how many did not? You don't expect me to accept Ebbinghaus'
observations just because of who he was, do you? Or to assume that everyone
is like Ebbinghaus?
I'm sorry, but on matters like this I can't just take your word for how
firm these facts are. I'm sure you wouldn't lie to me, but I would at least
like to know the numbers on which you base your acceptance of these
phenomena as real and important.
You wanted to squelch the discussion that was underway at the time and "talk
about PCT." I asked a question about PCT research. For some reason, Rick
wanted to talk about "conventional" research. I didn't. I still don't.
...
Now, can we get back to talking about PCT?
OK.
1. The fact of control.
Through numerous experiments with different controlled variables, we can
show that behavior is related to controlled variables and disturbances in
exactly (quantitatively) the way that is meant by the word control in PCT.
This is true of essentially every person tested (95% or better) and on
every trial of every test of the generalization with any individual (95% or
better).
2. The fact of control of perception (or at least inputs).
Control is established by a formal experimental procedure. The role of
input is established by several means. The simplest is to eliminate inputs
and see what the effect on control is (it is abolished, for every control
process tested and for every person tested). A more sophisticated approach
is to distort the inputs through which the person is able to perceive the
controlled variable, and show that the change in the formerly controlled
variable is quantitativly what is predicted from control theory. If the
input has a constant added to it, for example, the controlled variable is
maintained at the same constant amount in the other direction from its
original state. Again, these generalizations hold for more than 95% of
people tested, and for any individual, in more than 95% of the trials.
3. The behavioral illusion.
It can be shown that when stimuli appear to cause responses via the
organism (the S-O-R interpretation), it is possible that there is an
undetected controlled variable that is affected both by the stimulus and by
the response. As a result, an actual proximal stimulus (controlled
variable) may show no significant correlation with either the distal
stimulus or the response, although the distal stimulus that acts via the
proximal stimulus does show a highly significant negative correlation with
the response. This shows, in cases where a controlled variable can be
identified, that the organism does not just "mediate" between stimulus and
response. The observed relationship is actually mediated by properties of
the environment. These observations can be replicated as in the other cases
with more than 95% success, with any randomly-chosen subjects.
4. The stability of control parameters.
When matching a control model to behavior, we find that after ten or twenty
practice trials the best-fit parameters come to asymptotic values, after
which they remain close to the same values over periods of as long as 10
years (in Tom Bourbon's longitudinal series). No subjects have been
observed to show sudden or large changes in these parameters, once the
initial practice has occurred. The parameters can vary substantially from
one subject to another.
5. The achievement of consistent results by variable actions.
First noted in 1895 by James, this phenomenon has been demonstrated
countless times with human beings, in a wide variety of instrumented
control experiments. Whether caused by additive disturbances or changes in
feedback parameters, unpredictable changes in the controlled outcome of
behavior are uniformly resisted by the production of exactly those changes
in behavior needed to counteract the disturbances.
6. The integral relation of changes in the controlled variable to changes
in output action.
When the feedback connection is proportional, changes in the controlled
variable away from the observed reference state result in proportional
changes in the _rate of change_ of the output action. Thus the state of the
action variable is proportional to the time integral of the difference
between the value of the controlled variable and its observed reference level.
7. "Mind reading."
Given that an action may have many simultaneous kinds of effects on the
environment, we can apply the basic principles of control to find out which
effects (if any) are intended by the actor and which are uncontrolled
side-effects. While Marken's demos of this phenomenon are designed to make
a correct detection with a minimal amount of data, and thus are sometimes
wrong, with more extended data it is essentially always possible to detect
the intentionally controlled effect and distinguish it from unintentional
effects, by objective means.
8. Control of behavior.
When a controlled variable has been identified, disturbing it can be a
means of controlling the actions that are being used as the means of
controlling the variable. As long as the controller continues to
demonstrate control of the same variable, this method of the experimenter's
controlling action always works with quantitative accuracy.
9. Countercontrol.
When an experimenter is using disturbances to control the actions of a
human controller as in 8., that controller can switch to controlling the
action of the experimenter, by giving up control of the former controlled
variable and instead selecting a reference condition for the way the
experimenter applies disturbances.
10. Nested (hierarchically related) control processes.
It can be shown that a physical impulse applied to an outstretched arm can
serve simultaneously as a disturbance of arm position (which is corrected
on the time-scale appropriate to low-level control systems) and as a signal
agree upon as the occasion for changing the reference position of the arm.
In every human subject, the disturbance is resisted in a time shorter than
is required to interpret the disturbance as a signal and voluntarily change
the arm position. No exceptions to this ordering of events have been observed.
11. The voluntary variation of reference conditions ("the experimental
investigation of purposes").
By matching a model to behavior under conditions where the reference level
can be shown to be constant, it is possible to deduce any subsequent
variations in the controlled variable that are due to changes in the
reference condition. I have demonstrated this deductive process for one
subject, myself; there is no reason it should not work for anyone else.
12. The properties of joint control of shared controlled variables.
I refer here to Tom Bourbon's experiments and those of his students, in
which cooperative tasks were studied. It was shown that the performance
could be exactly predicted from control models for each person interacting
with each other and the experimental apparatus. The predictions were just
as accurate for any mix of computer models and real subjects interacting in
pairs. They were just as accurate for any selection of human participants.
13. Determination of the degrees of freedom of multidimensional control.
In experiments with two-dimensional control, Marken found that independent
control systems working in x and y coordinates could be modeled with a
simple pair of independent control systems, while a model working in polar
coordinates could not reproduce the behavior as accurately.
14. In experiments with control involving the loss of present-time
feedback, it was shown that model-based control could not explain what was
observed (unless the model was so poor as to be useless).
15. Individual differences in the control of sound pitch.
People controlling the pitch of a sound were investigated. Most people
could control the pitch accurately (within 5%) for a period of one minute
after the tone to be reproduced was sounded. About a quarter of the people
(roughly) could not control the pitch at all. All of these people said they
could not sing or play an instrument -- they considered themselves
"tone-deaf" (and so did their teachers!). The distinction between pitch
controllers and noncontrollers was on the order of five standard deviations
by a measure of "stability factor". It is not known whether the
noncontrollers could ever learn to control pitch.
There are more, but this indicates the kinds of empirical discoveries that
have been made in PCT. They all relate, of course, to control behavior.
If you divide the number of discoveries in PCT by the number of workers
actively experimenting each year with control phenomena, you will come up
with a number like 5 to 10 discoveries per investigator. If you divide the
number of (legitimate) discoveries made in psychology by the number of
active experimental psychologists, I suspect that the number would be
considerably smaller -- not to mention that the discoveries in conventional
psychology are nothing like as reliable or reproducible. And that is the
point of my asking you for the numerical details about the list of
phenomena you present as established facts. If any comparison is to be
made, it has to be in terms of a like quality of results. A fact that is
true 80 percent of the time is vastly less important than a fact that is
true 95%, or 99%, of the time. If your conclusions rest on the conjunction
of three facts with 80% probability of being true, the probability that the
conclusion is true will be just about at the chance level. With a
truth-probability of 99%, the conclusion would still be 97% likely to be
true. The quality of facts makes an enormous difference in our ability to
reason logically about them.
Best,
Bill P.