[From Bill Powers (940930.0655 MDT)]
Martin Taylor (940929.1030)--
... access to an IBM-compatible PCT (Personal Computer Terminal). That
is not always the case.
The Mac is a wonderful machine for the user. But behind those simple
icons, windows, dialogue boxes, sliders, etc. is a lot of very complex
programming. The Apple Computer philosophy has always been to throw the
burden on the programmer rather than the user, with the idea that a
program developer will presumably be acquainted with the state of the
Apple art, and will be compensated for the extra work by being able to
make money from writing user-friendly programs that obey the rules for
Mac programming. Programmers who work under Windows on PCs are in much
the same position.
Fortunately, one is not forced to work under Windows on a PC, so
programming can be done in a simple way using helpful environments like
Borland Turbo C, with the whole machine at the programmer's disposal and
no other programs being run at the same time to cause problems with
memory sharing, timing, and so on. On the Mac, however, one has no
choice: conform to the time-sharing rules, or don't program at all
except in some (much slower) higher-level language like Excel or
Hypercard. Or Basic, which seems limited to me.
Rick has tried to find a C compiler for the Mac that will work as well
as Turbo C, without any luck. His Pascal compiler defeated us when we
tried to port a PC program to his Mac. There is obviously some way to
write C programs for Macs, but we haven't found any that makes
transporting PC programs to the Mac feasible FOR US.
It would be great if some smart, generous, young Mac Whiz were to help
out on this, even if only to steer us to a C or Pascal programming
environment for the Mac that is as user-friendly as Turbo C. If the
Center for the Study of Living Control Systems were in operation we
could hire a couple (hiring just one would be cruel).
Until something like that happens, my programs will be written for PCs,
not because PCs are better than Macs but because programming them is
within my capabilities.
···
------------------------------------------------------------------------
Jeff Vancouver (940929) --
I'll send this direct to you also in case you signed off already.
We'll be sorry to see you go, as you are a link to the mainstream world.
I hope your experiences with PCT so far have not spoiled anything for
you!
I hope you had a chance to read Chuck Tucker's wonderful post of 940929.
Chuck is in much the same position you are in, not being a programmer
and not having had any previous experience with modeling. He is the
perfect example of how non-technical CSG members work with those who can
write programs and set up models. Chuck decided that he would simply run
our models and get some experience with how they work; as can be seen in
his post, this has led to a tremendous increase in understanding of PCT,
and the development of abilities to see the possibilities of PCT
research of many kinds. He can now spot previous research in which there
were approaches compatible with PCT even if their data and analyses were
not presented in a way we could use directly. This is going to lead,
eventually, to a whole methodology that sociologists can use to apply
PCT in their own fields -- without ever requiring them to become
modelers or programmers.
I sympathize completely with your need to pursue tenure; what a world!
But I hope you will stay in touch with Chuck, because he can help you
see how to apply PCT without having to be a techie.
-----------------------------------------------------------------------
Chuck Tucker (940929) --
RE: replication: pulleys, strings, weights, and springs
That was an inspiring post, Chuck. I hope you will write some more on
the advantages of replication. It's not just fulfilling a formal
requirement of science; it's a way of getting to understand what the
other guy is talking about, by walking a mile in his shoes. There is
nothing that can substitute for the direct experience of the control
phenomenon, for seeing how the results just keep repeating and repeating
-- and how well a simple theory can predict them.
So here is an essay on replication, inspired by you.
In introductory physics, students have to run through a whole series of
elementary experiments. They replicate Galileo's experiments by rolling
balls down inclined planes and timing pendulums of different lengths.
They hook up pulleys and weights and measure how far the load moves when
the string at the other end is pulled. They hang weights on springs and
measure the stretch of the spring. One after the other, they repeat all
the basic old experiments which were once at the cutting edge of
physics, done by people who would later have their woodcut portraits
reproduced in 20th-century textbooks. The point of all this is not just
to salute the past, but to show _why_ the basic physical laws, and the
equations that describe them, are considered so important: THEY WORK. A
student can use simple algebra to make predictions, do the experiment,
and see that the equations predict _what actually happens_. For some
students like me, these simple demonstrations open up a new world. They
redefine what "understanding" means. They tell us that nature is
comprehensible.
I don't think that psychologists and sociologists have ever really
believed that human nature is comprehensible. Most of them have never
experienced a prediction that actually describes, in advance and
accurately, what a human being is going to do. It doesn't seem to bother
them that a prediction of behavior may be violated by a large number of
individuals in a study; they don't really expect to see predictions work
accurately or for everyone. All they hope for is to detect a trend, a
suggestion of a relationship, a pale ghost of a natural law that can be
seen only in averted vision.
And why are they satisfied with this kind of knowledge? I claim that it
is simply because _they have never encountered any other kind_. I think
that one reason for Skinner's strong following was that he came up with
an experiment in which results were highly predictable: the cumulative
record marched up the paper, showing scallops, reversals, recoveries,
and other phenomena in a highly predictable way. People could set up a
Skinner box, replicate published schedules of reinforcement, and watch
the same traces develop. People who got used to seeing such regularity
in behavior thenceforth wanted nothing to do with the statistical
vagueness of standard psychology. Once you have seen what true
regularity looks like, there's no going back. Skinner had everything he
needed to put psychology on a new track except a model that could
explain this behavior. Neither he nor his followers understood modeling;
they did curve-fitting instead.
I think the key to what is wrong in the behavioral sciences is the
general lack of replication. In the first place, replicating a typical
behavioral experiment would be time-consuming and costly, because in the
effort to discover small hints of effects, experimenters must use large
numbers of subjects; in trying to get the most out of their efforts,
they vary many conditions, hoping that at least one factor will stick up
out of the noise far enough to be called significant. In the second
place, I don't think that experimenters really believe that if they did
try to replicate another person's results, they would succeed. There are
too many uncontrolled variables, sampling errors, undocumented details
of procedure, differences in available equipment, differences in
physical setting, differences in experimenter attitude. Anyone who stops
to think of all the variations that could take place in attempting a
replication would be discouraged from the start. How could I ever
actually _reproduce_ another person's experimental procedure ewven if I
could afford the time and expense? Where could I obtain a comparable
population of subjects? And anyway, with the results depending on a slim
preponderance of one kind of behavior over another, it doesn't seem
likely that I would be lucky enough to have the _same_ relationships
just barely reach p < 0.05.
I'm just guessing, of course. Others who know experimental psychology
and psychologists better may not agree with me. But the fact remains
that there is almost no replication of experimental results in the
literature of the behavioral sciences. Experiments are done once, and
from then on the statistical findings, if interesting enough in their
implications, are taken as facts.
This tells me that behavioral scientists, for the most part, have never
had the experience of a truly successful replication. And without that
experience, it is impossible to get any sense of reality about the
phenomenon revealed by the replication. Knowledge takes on an abstract
aura, a sense of floating in intellectual space without an anchor. All
things become possible, no fact seems thrust upon one as an inescapable
aspect of nature. Understanding nature becomes a matter of plausibility,
of possibility, of mathematical and logical manipulations with no ties
to observation.
What's important about running the PCT demos is that each demo is in
fact a replication of an experiment with human nature. The outcome is
predicted, and what is predicted happens every time. The human
participants are under no constraints to behave as they do; they hold a
control stick in their hands, and are free to move it in any way they
please. Yet they move it just as the theory predicts they will move it,
creating patterns of movement that could be drawn in advance. These
replications reveal the phenomenon of control; it is right there in
front of you, and you can make it repeat any time you wish. The whole
point is for the person doing the replications to begin to get the sense
of comprehending a basic phenomenon of behavior, to realize that there
is something about human behavior, even if it is a small and unimportant
something, that can be understood just as thoroughly as we understand
how pulleys, strings, weights, and springs work together.
There was a time in the history of science when the most important
things than anyone could understand about nature were at the level of
pulleys, strings, weights, and springs. All the vast array of
developments from these beginnings had to wait upon mastery of these
simple facts of nature. There was no short-cut -- to take a short-cut,
one has to know where one is going. At the time of these first
developments, all of physics lay uninvented and unsuspected in an
unimaginable future. All that could be done was to keep working with
what could be known, to build a base on which more knowledge could be
constructed.
There were always those who wanted to jump ahead of this process, to see
if they could guess how it would all come out and avoid going through
the tedious simple steps. They thought, perhaps, that there would be
glory in guessing right long before the plodders of science got there.
But in doing so, they misconstrued science itself; what matters is not
guessing at the right answer, but developing a method for getting to it
that can be applied over and over to develop more right answers, right
answers in a superabundance that makes a single lucky guess look
trivial. In my opinion, psychology has looked for the lucky guess, the
impressive flashy result, the insight that cuts through the details and
offers explanations out of the blue. It has neglected the pulleys,
strings, weights, and springs, and in doing so has built a house on
sand.
While attempting to apply PCT to larger aspects of human behavior is
interesting and might in some cases even be useful, I think it is a
mistake to give too much importance to that side of our work. There's a
great temptation to join all the others who think that they can meet
human needs (or make a buck) by talking confidently about things they
don't really understand. The true state of affairs is that we are still
at the level of pulleys, strings, etc. and have nothing of any permanent
value to say about higher levels of organization.
I have no objection to trying to apply the principles of PCT in any way
that seems interesting or useful. But I do not choose to be part of
that, and I hope that growing numbers of PCTers will choose as I do, to
continue to look at basic phenomena and keep working on the knowledge
engine that will some day utterly replace everything that people think
of today as a science of behavior. Nothing we can say about the higher
levels today will be of any important whatsoever in the long run. It
will all look like mysticism, superstition, superficiality, and magic,
cloaked in the kind of language we have heard before from scholastics,
seers, alchemists, and the Priests of Isis. We are at the start of a
long task, which will end with a science in which all results are
replicable, and replication is taken for granted as a natural
requirement of science -- replication within the limits of measurement.
I know that this is possible. I have seen it done, and can do it again
on demand. Knowing that, how can I possibly settle for anything less?
---------------------------------------------------------------------
Bruce Buchanan (940929.14:00 EDT)--
Glad you's finally obtained B:CP. I hope it's not a let-down after all
the advance publicity. As you intimate, it would probably be best to
defer further discussions on phenomena of control until you've read it.
I'm also glad that you find CSG-L invigorating. We find your comments
invigorating, too; you bring new ideas and good sense to our
discussions.
----------------------------------------------------------------------
Martin Taylor (940929.1030)--
Those were some great facts about perception; you've mentioned some of
them before, but I didn't realize there were other examples, too. As you
know, I am always impressed when the data points lie on the theoretical
curve.
It strikes me that these phenomena could easily be made part of a PCT
experiment. Suppose the subject saw some shaded bumps and hollows on the
screen, with the apparent depth being variable and affectable by a
handle and a disturbance. Every time the depth went through zero there
would be a chance for an ambiguity (and maybe even when it is not zero).
Of course a reversal would flip the sign of the feedback, leading at
least for a moment to a runaway condition (as in Rick's experiments with
reversals). We could see the reversals without needing any report from
the subject. We could also insert step-disturbances by actually
reversing the bumps and hollows, a matter of changing the simulated
shading. A "tracking" experiment would also work, with a target bump or
hollow that the participant matches by acting on a second bump or
hollow.
This would also work with wire-frame figures rotating on the screen.
Would putting such figures in a control loop make the flips less random?
Modeling perceptual functions is, of course, much harder than modeling
control processes. We have to go a level deeper into the processes, like
trying to guess how the brain extracts position information from an
image of a target and a cursor on the retina. Maybe by using a control
task in which we can see the reversals unambiguously we will be able to
see the effect of higher-level hypothesis-changing, and even devise an
experiment in which control depends explicitly on applying the right
hypothesis. One mode of perceptual control, completely unexplored so
far, would involve a higher system supplying missing data for a lower
system to make an otherwise ambiguous perception unambiguous. This would
be a first experiment with the imagination connection (or at least the
phenomenon -- maybe that connection isn't the only workable model). I'm
reminded of the problem of controlling size in the absence of distance
information; one has to assume a distance in order to perceive a size.
What controls the "assuming?"
The point I am making is that both in principle and in practice, not
all research contributing to the understanding of perceptual control
systems need involve studying the control system in action.
True. But I think that doing the same experiments again as part of an
explicit control task might (a) serve to replicate the original
findings, and (b) give us a better measure of the time of reversal than
a verbal report or unrelated key-press can do. Who knows, maybe the data
would fit the n(n-1) curve even more closely.
Complaints against statistics should be directed against improper uses
of statistics (e.g. any "significance test," or a generalization from a
trend found in a set of everages to an implied average of a set of
trends).
I agree completely. And Phil Runkel has laid out other mistaken uses of
statistics to steer away from. The only other quarrel I have with proper
applications of statistical models is when correlations like 0.3 are
treated as if they had been 1.0.
-----------------------------------------------------------------------
Best to all,
Bill P.