[From Bruce Nevin (970128.0952)]
From Rick's Dancer paper (Conclusion, first para):
"... the relationship between IV and DV reveals something about the nature
of the connection between the organism and aspects of the environment that
it controls."
Is this what Martin is saying about the value of psychophysical research
for PCT? That this part of the loop through the environment is important to
study so that we can model it correctly?
Naively,
Bruce
[From Rick Marken (980128.1120)]
Bruce Nevin (970128.0952)
From Rick's Dancer paper (Conclusion, first para):
"... the relationship between IV and DV reveals something about
the nature of the connection between the organism and aspects of
the environment that it controls."Is this what Martin is saying about the value of psychophysical
research for PCT? That this part of the loop through the
environment is important to study so that we can model it
correctly?
No. Martin thinks psychophysical research is important because
it tells us about the relationship between what we think are the
basic dimensions of physical (e) variation (intensity, frequency,
etc) and their perceptual correlates (p); that is, psychopysics is
aimed at discovering the function (f()) relating p to e (p = f(e)).
I agree that psychophysical research _would be_ important for this
reason because, if it provided _valid_ results, psychophysics would
tell us something about the mapping of simple controlled environmental
variables to their perceptual correlates. But I doubt the validity
of most psychophysical research because it is done in the context of
an input-output model of behavior; that's where I disagree with
Martin regarding psychophysics; not on it's importance but on its
_validity_.
What is observed in psychophysical experiments is what, in PCT,
we would say is the relationship between disturbances (d) and the
outputs (o) that are involved in preventing those disturbances from
having an effect on some controlled variable. Psychophysicists
believe that the relationship between d and o that they observe
in their experiments is a measure of a characteristic of the organism
(the organism's perceptual function, in the case of psychophysical
research). My statement in the "Dancer..." paper that you quote
is saying that this is not the case; the relationship between
d (IV) and o (DV) that you observe in any experiment (including
psychophysical experiments) on a control system is _not_ a measure
of a characteristic of the systemn (such as its perceptual function);
rather, it is a measure of a characteristic of the the environmental
connection between the system's _output_ (o) and the variable
it is controlling via those outputs. That is, what you are seeing
in a psychophysical experiment is a _behavioral illusion_.
I think it is extremely important to understand the behavioral
illusion if you want to understand why PCTers have so much
trouble listening to people tell them about all the important
discoveries made by conventional research. If you haven't done it
yet, I recommend going through the "Behavioral Illusion" demo
at my control demos site:
http://home.earthlink.net/~rmarken/demos.html
Also, you might want to try the version of the rubber band demo
that I describe in the "Dancer..." paper. Bill's discovery of
the "behavioral illusion" has to be the mind-blower of the
millenium -- or the downer of the millenium, if your career is
based on doing conventional behavioral research;-)
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[Martin Taylor 980129 22:50]
Rick Marken (980128.1120)
Martin thinks psychophysical research is important because
it tells us about the relationship between what we think are the
basic dimensions of physical (e) variation (intensity, frequency,
etc) and their perceptual correlates (p); that is, psychopysics is
aimed at discovering the function (f()) relating p to e (p = f(e)).
I'm mildly interested in why you say this. Is it because you think it
true in spite of all I have written to the contrary? I hate to put words
to the other possibility that lurks unbidden in my head.
I don't understand why you say it, whichever reason is correct. Surely it
would be better to take issue with the things I write, with which you
legitimately disagree, rather than asserting that I believe the things
I take pains to say I don't believe.
For the record, I am not sure that for any given physical variable
there exists a function that relates it to a perceptual signal--at least
not a function that is time-invariant.
Also for the record, the canonical version of PCT says that such functions
do exist. They are defined by the perceptual input functions--in fact
they _are_ the perceptual input functions. The reason I say I am not
convinced they are time-invariant (as would be required if we were to
believe Rick's constant assertion that the only goal of any PCT-valid
research is to find "the controlled perception") is what I said a month
ago in a discussion with Bill P and amplified in later messages--
+Martin Taylor 971231 17:45
+The question arises as to whether the perceptual input functions operate
+the same way when the resulting perception is being controlled as when it
+isn't. This issue is not ordinarily considered within HPCT, since normally
+the perceptual input function is taken to be whatever it is, and only the
+magnitude of its output is controlled. But it is an issue, one that might
+invalidate the uncritical use of the results of psychophysical studies
+to assess the elements of related control loops.
Martin
[From Bill Powers (980130.0517 MST)]
Martin Taylor 980129 22:50--
For the record, I am not sure that for any given physical variable
there exists a function that relates it to a perceptual signal--at least
not a function that is time-invariant.
I'm sure you're right about this, but is it what you meant to say?
Electrical charge is a physical variable, but I know of no perception that
corresponds to it. In fact, I'm pretty sure there is no perception that
correponds to most of the idealized mathematical variables defined in physics.
Since we don't perceive everything that is in the physics-model of the
world, for PCT purposes we have to go at it the other way around. The
question is, to what in the physics model does a given perception
correspond? In terms of human experience, it is the perception that is
given, not the physics.
Most human perceptions have no direct counterpart in the world of physics.
Taste, color, timbre, and other such perceptions are functions of multiple
variables and of time, so there is no unique physical situation that goes
with a given value of any perceptual signal. Many different mixtures of
wavelengths, for example, can excite exactly the same color sensation (if
that were not true, it would be impossible to adjust a television set or
the properties of color film to produce realistic-looking colors).
Also for the record, the canonical version of PCT says that such functions
do exist. They are defined by the perceptual input functions--in fact
they _are_ the perceptual input functions. The reason I say I am not
convinced they are time-invariant (as would be required if we were to
believe Rick's constant assertion that the only goal of any PCT-valid
research is to find "the controlled perception") is what I said a month
ago in a discussion with Bill P and amplified in later messages--
+Martin Taylor 971231 17:45
+The question arises as to whether the perceptual input functions operate
+the same way when the resulting perception is being controlled as when it
+isn't. This issue is not ordinarily considered within HPCT, since normally
+the perceptual input function is taken to be whatever it is, and only the
+magnitude of its output is controlled. But it is an issue, one that might
+invalidate the uncritical use of the results of psychophysical studies
+to assess the elements of related control loops.
This is a good point, but what it casts doubt upon is the psychophysics,
not PCT. The question is, which do we consider the primary evidence: the
experienced world, or the physical world? The world of physics is deduced,
or at least hypothesized. But all human beings, in trying to understand the
world, have to start with the experienced world. It is from that world that
physicists arrive at the deduced world of physics. It is the "psycho" part
of psychophysics that contains the givens of experience.
Consider a simple perception like the distance between two objects. As we
know, human perceptions do not create distance measures like those in
physics. The way physicists get around this is through the use of
"measuring rods", so that all the human being has to judge is whether the
ends of the rods (or marks on them) coincide with the positions of the
objects. The judgment of distance is made only for _zero_ distance. This
removes the nonlinearities that develop between physical measurements and
subjective perceptions of distance for non-zero distances.
However, when we control distances without the use of measuring rods, it is
the perceived distance, not the physically-measured distance, that is
controlled. And this is where we lose the correspondence between perception
and physics. I can reproduce the same perception I had before, but there is
no guarantee that in doing so I am producing the same situation as a
physicist would deduce it. Nor, without getting into the brain and locating
and measuring the perceptual signal, is there any way to establish a
relationship between the magnitude of the experience and the physical
measurement. There is no measuring rod that is common to the world of
physics and the world of experience.
To me, this means that the basic aim of psychophysics is impossible to
achieve through experiments with intact organisms. In psychophysical
experiments we are always comparing one perception with another perception,
not one perception against a physical measure.
Best,
Bill P.
[From Rick Marken (980130.0750)]
Me:
Martin thinks psychophysical research is important because
it tells us about the relationship between what we think are the
basic dimensions of physical (e) variation (intensity, frequency,
etc) and their perceptual correlates (p); that is, psychopysics is
aimed at discovering the function (f()) relating p to e (p = f(e)).
Martin Taylor (980129 22:50) --
I'm mildly interested in why you say this. Is it because you think
it true in spite of all I have written to the contrary? I hate to
put words to the other possibility that lurks unbidden in my head.
No sinister motives at all. Perhaps I didn't state it well but
I meant to descibe what psychophysics thinks it's about. What I said
above is how I would have descibed the aim of psychophysics in an
Intro Psych class. I guess I'll read on and see what you think
psychophysics is really about.
For the record, I am not sure that for any given physical variable
there exists a function that relates it to a perceptual signal
I agree. But that's not inconsistent with my definition of
psychophysics, is it?
Also for the record, the canonical version of PCT says that such
functions do exist. They are defined by the perceptual input
functions--in fact they _are_ the perceptual input functions.
Yes. So?
The reason I say I am not convinced they are time-invariant
(as would be required if we were to believe Rick's constant
assertion that the only goal of any PCT-valid research is to
find "the controlled perception") is what I said a month ago
in a discussion with Bill P and amplified in later messages
I don't see why my assertion about the _main_ (not the _only_)
goal of PCT research requires that perceptual functions be time-
invariant. Actually, I don't know what your problem is. The point
I was trying to make in the post to which you are responding was
simply that psychophysical experiments (whatever their goals and
whatever the merits of those goals) are as invalid as any other
conventional behavioral experiment on control systems where a
controlled variable is not identified; the observed relationship
between inputs and outputs in such experiments represents
characteristics of the environment, not of the organism. If
the researcher thinks the results of psychophysical experiments
reveal something about the organism under study (such as one of
its perceptual functions) then she has fallen pray to the
behavioral illusion.
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[Martin Taylor 980201 10:40]
Rick Marken (980130.0750)]
Me:
Martin thinks psychophysical research is important because
it tells us about the relationship between what we think are the
basic dimensions of physical (e) variation (intensity, frequency,
etc) and their perceptual correlates (p); that is, psychopysics is
aimed at discovering the function (f()) relating p to e (p = f(e)).Martin Taylor (980129 22:50) --
I'm mildly interested in why you say this. Is it because you think
it true in spite of all I have written to the contrary?
I don't see why my assertion about the _main_ (not the _only_)
goal of PCT research requires that perceptual functions be time-
invariant. Actually, I don't know what your problem is.
My problem is not what you think psychophysics may be about. That's
your affair, right or wrong. That you "don't know what [my] problem
is" is unsurprising, given your penchant for substituting your controlled
imagination of what people say in place of what they do say.
My problem is that you ascribe to me beliefs that are contrary to what I
have written, beliefs that I have gone to some trouble to say are wrong.
Then you use my supposed holding to those beliefs as reasons to criticise
my general approach to PCT and to make predictions (not insults) such as
"I'm sure you never will [understand]". This happens time and time again. I
thought that this time I ought to call you on it.
Bill Powers answered my message by stating a lot of stuff I have previously
stated, perhaps better than I could or did do. Of course I agree with
most of his message, and would have done before I ever heard of PCT.
(I don't agree with all of it, but that's a separate issue).
----------------------------
The point
I was trying to make in the post to which you are responding was
simply that psychophysical experiments (whatever their goals and
whatever the merits of those goals) are as invalid as any other
conventional behavioral experiment on control systems where a
controlled variable is not identified; the observed relationship
between inputs and outputs in such experiments represents
characteristics of the environment, not of the organism.
The "observed relationship between inputs and outputs" is the action
of the subject that indicates the subject saw or did not see (hear, taste,
smell...) something when a real physical "something" is known to have
been available to see (hear, tast, smell...). You seem to be saying that
whether the subject can taste salt at a concentration of x per million
is a characteristic of the environment, not of the subject. If that is
so, you might explain what it is about the environment that does the
tasting and allows the subject to act according to whether the evironment
tasted the salt or not.
If
the researcher thinks the results of psychophysical experiments
reveal something about the organism under study (such as one of
its perceptual functions) then she has fallen pray to the
behavioral illusion.
Sorry about that. I thought that seeing (hearing, tasting, smelling...)
was something an organism did. I am indeed subject to the behavioural
illusion. I must learn: " All perception is done by the environment."
----------------------
A comment on Standard HPCT and perceptual functions.
According to "Standard" HPCT, the hierarchy consists of a series of levels
of elementary control units (ECUs). Each ECU contains an input function
that accepts possibly many inputs from the sensors or from the outputs
of lower level ECUs. The output of an input function is a perceptual
variable. The sensors transform some physical variable in the environment
into a signal value suitable for input to an ECU input function.
Perceptual variables are therefore functions of physical variables.
According to "Standard" HPCT, the form of a perceptual input function is
determined by reorganization, which in turn depends on the degree to which
intrinsic variables are controlled to be near their inborn reference
levels. This implies that the perceptual input functions of (at least)
low-level ECUs change slowly, if at all, over time. Hence, again according
to "standard" HPCT, there should exist functions that relate the values
of perceptual signals to the values of physical variables.
I happen to believe that Standard HPCT is inadequate in this linkage
between perceptual and physical variables, and I said so. There are many
reasons why I think it inadequate, based on the results of "conventional
research", some of which I have mentioned in previous messages. Bill P
agrees with me that it is inadequate, for the same reasons (980130.0517
MST) "I can reproduce the same perception I had before, but there is
no guarantee that in doing so I am producing the same situation as a
physicist would deduce it."
Most psychologists know this, from the results of "conventional"
psychophysical studies, but so far as I can interpret Bill's message,
his agreement is based on his powerful intuition, since to him the
results of non-PCT experiments are inadmissible.
The ways in which a fixed function relating a perceptual signal to
physical variables is inadequate (according to "conventional" studies
can be summarized in the oversimplification that the output of the
sensors (and by extension the output of any perceptual function) is
dependent on the context extended over time. I have added that it is
possible that the output of the perceptual input function of an ECU
may depend on whether that ECU is actually controlling the perception
at the tested moment--but this addition is purely intuitive, and is made
largely to warn people further against taking seriously the results
of experiments that purport to "measure" the magnitudes of perceptual
variables.
Bill P has it right when he says (980130.0517 MST): "In psychophysical
experiments we are always comparing one perception with another perception,
not one perception against a physical measure." This is why I argued for
the admissibility of psychophysical experiments that measure the ability
of people or animals to do just that. Experiments that determine the
ability of people to distinguish whether the values of a perceptual
signal at two times differ are experiments that can be used to determine
how well an ECU could control the studied perception. As in the following
interchange that formed part of a dialogue about the usefulness of
psychophysical studies that showed trained subjects to be able to detect
sound changes typically within about 4 or 5 dB of what a mathematically
ideal observer could achieve under a wide variety of conditions:
--Martin Taylor 980104 11:50--
-->1. In standard HPCT, is it assumed that one can control the values of
-->a perception more finely than those values can be discriminated within
-->the perceptual signal.
--
--Bill Powers (980104.1041 MST)
--No. But it can control less finely;
-------------------------------------
Rick Marken asked (980109.0840):
+Perhaps you could describe, in detail, a psychophysical experiment
+that has already been done -- one that provides (according to you)
+results that are valid and useful from a PCT perspective. The
+presentation of such an experiment would be a strong argument
+against my claim (and isaac's and Bill's too, I think) that all
+conventional behavioral research results must be reevaluated
+using PCT methodology to see what was actually going on.
I didn't reply, because I had in previous messages done exactly what he
asked. I had thought of making a somewhat snide reply (of the kind that
my own messages sometimes receive) by quoting some citations in the paper
"The Hierarchical Behaviour of Perception" (Closed Loop 3/4, pp33-54)
by an author with the initials RSM, but I decided it might be unkind to
do so.
Anyway, the upshot is that I think that (as a minimum) experiments that
put a bound on the ability of people to perceive changes in a perceptual
variable are useful to PCT. And the citations Rick used in his paper
provide other examples of potentially useful conventional studies.
Martin
[From Rick Marken (980202.0845)]
Martin Taylor (980201 10:40) --
My problem is that you ascribe to me beliefs that are contrary
to what I have written, beliefs that I have gone to some trouble
to say are wrong.
Ok. I'll let you speak for yourself from now on. You do just
fine on your own;-)
For example, I said:
If the researcher thinks the results of psychophysical experiments
reveal something about the organism under study (such as one of
its perceptual functions) then she has fallen pray to the behavioral
illusion.
And you say (sarcastically but, I'm sure, tactfully;-))
Sorry about that. I thought that seeing (hearing, tasting,
smelling...) was something an organism did. I am indeed subject
to the behavioural illusion. I must learn: " All perception is
done by the environment."
Your sarcasm suggests that you disagree with my statement about the
behavioral illusion. That is, you disagree with my claim that, in
psychophysical experiments, the relationship between stimulus
and response -- the "psychophysical law" -- is a characteristic
of the environmental feedback function relating outputs (and
stimulus) to a controlled variable. You apparently believe that
this is not true; that somehow this characteristic of the behavior
of control systems doesn't pertain to behavior in psychophysical
experiments. Apparently, you believe that the behavioral illusion
doesn't apply to the results of psychophysical experiments; that
the relationships between stimulus (disturbance) and response
found in such experiments actually do tell us something about
the nature of the organism -- in particular, the organism's
perceptual function.
I realize that it is difficult to believe the implications of
the behavioral illusion for _all_ of psychology (including
psychophysics). It seems absolutely unthinkable that the nature
of the effects we find in conventional behavioral research --
even when we use the method of specimens (test one individual
at a time) and find highly reliable relationships, as we do in
psychophysics -- tell us mainly about the nature of the organism's
environment and little about the orgnism itself.
Believing that the behavioral illusion occurs is like believing
that we actually live on the surface of a sphere when so much of
our everyday experience says that we live on a flat surface. But
if we take seriously the possibility that organisms are input
control systems, and if we understand what the behavioral illusion
_is_ (I recommend that you carefully review my behavioral illusion
demo at
http://home.earthlink.net/~rmarken/ControlDemo/Illusion.html
if you are not clear about it) then we have to admit that the
results of virtually all conventional psychological research
(including psychophysical research) tells us very little about
the _organisms_ under study but quite a lot about the _environmental
connection_ between those organisms and their controlled perceptual
inputs.
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[Martin Taylor 980203 01:02]
Rick Marken (980202.0845)
For example, I said:
If the researcher thinks the results of psychophysical experiments
reveal something about the organism under study (such as one of
its perceptual functions) then she has fallen pray to the behavioral
illusion.And you say (sarcastically but, I'm sure, tactfully;-))
Tact was ever one of my strong point, my dear old fellow;-)
Sorry about that. I thought that seeing (hearing, tasting,
smelling...) was something an organism did. I am indeed subject
to the behavioural illusion. I must learn: " All perception is
done by the environment."Your sarcasm suggests that you disagree with my statement about the
behavioral illusion. That is, you disagree with my claim that, in
psychophysical experiments, the relationship between stimulus
and response -- the "psychophysical law" -- is a characteristic
of the environmental feedback function relating outputs (and
stimulus) to a controlled variable.
I think it might be very interesting to study what you choose to quote
and what not to quote in your messages.
What you did not quote was:
The "observed relationship between inputs and outputs" is the action
of the subject that indicates the subject saw or did not see (hear, taste,
smell...) something when a real physical "something" is known to have
been available to see (hear, taste, smell...). You seem to be saying that
whether the subject can taste salt at a concentration of x per million
is a characteristic of the environment, not of the subject. If that is
so, you might explain what it is about the environment that does the
tasting and allows the subject to act according to whether the evironment
tasted the salt or not.
I'd be very happy to learn how, as you once again assert, when the "response"
is "Yes it was there" or "No nothing was there,"
the relationship between stimulus
and response -- the "psychophysical law" -- is a characteristic
of the environmental feedback function relating outputs (and
stimulus) to a controlled variable.
and if you would explain the mechanism whereby the environmental feedback
function determines that the subject asserts (wrongly, according to you)
that she can taste the salt in a solution at a concentration of x parts in
a million that the experimental apparatus placed on her tongue. I asked
you to explain it once. I ask again.
You apparently believe that
this is not true;...
Yes, I believe it is a property of a person whether the person can see,
hear, taste or smell something. Therefore I believe it is not true that
it is a property of the environmental feedback function that determines
whether the person _says_ they can see, hear, taste, or smell it, when
the person is intending to be truthful.
If I come into a room and say "I smell something burning," what aspect
of what environmental feedback function did the smelling?
... that somehow this characteristic of the behavior
of control systems doesn't pertain to behavior in psychophysical
experiments.
What is "this characteristic"? We are discussing the ability to discriminate
between perceptions, such as the perception of zero and non-zero quantity of
some thing, or whether pattern "A" has some quality different from that of
pattern "B." Certainly that ability is important in the behaviour of control
systems. It sets a lower bound to their accuracy. So I believe that this
characteristic of control systems (not of their "behaviour") is important
not only in psychophysical experiments, but in everyday life. People who
can't smell burning might just die in a fire they would have survived if
they could smell. It _is_ important to control that one be able to create
a perceptual signal of variable magnitude when the sensory input changes.
Apparently, you believe that the behavioral illusion
doesn't apply to the results of psychophysical experiments; that
the relationships between stimulus (disturbance) and response
found in such experiments actually do tell us something about
the nature of the organism -- in particular, the organism's
perceptual function.
They don't tell anything about the perceptual function as such, because
they aren't designed to do anything more than determine whether the
subject can distinguish one perception form another--as Bill P pointed
out, if for some reason you don't like to believe my writing.
I realize that it is difficult to believe the implications of
the behavioral illusion for _all_ of psychology (including
psychophysics). It seems absolutely unthinkable that the nature
of the effects we find in conventional behavioral research --
even when we use the method of specimens (test one individual
at a time) and find highly reliable relationships, as we do in
psychophysics -- tell us mainly about the nature of the organism's
environment and little about the orgnism itself.
I'm intrigued, then, as to why you felt it reasonable to use the results
of "conventional behavioural research" in your "Hierarchy of Perception"
paper--or is that another one of your now disowned products of a mis-spent
youth?
Martin
[Martin Taylor 980203 10:50]
Bruce Nevin (970128.0952)
From Rick's Dancer paper (Conclusion, first para):
"... the relationship between IV and DV reveals something about the nature
of the connection between the organism and aspects of the environment that
it controls."Is this what Martin is saying about the value of psychophysical research
for PCT? That this part of the loop through the environment is important to
study so that we can model it correctly?
No. What I am saying about the value of psychophysical research is that
it can provide bounds on how well any particular perception _could be_
controlled. If control is worse than that, one might well look for reasons.
When the psychophysical method that indicates the greatest sensitivity
shows that trained people are capable of discriminating between
perceptions almost as well as a mathematically ideal observer could do,
that's a fairly good indication that the psychophysical method is really
measuring the person's capability. This is why psychophysicists over the
years have spent so much effort working on techniques that allow the subjects
to demonstrate the sensitivity they do have (e.g. forced choice versus
"Yes-No--it is/isn't there"). If the "measured" sensitivity is much worse
than the ideal, we have to worry as to whether the discrepancy is due to
the technique or is a real property of the person (as is the relative
inability of a person with noise-induced hearing loss to hear soft high-
pitched sounds). Some people just _are_ more capable of discrimination
than are others.
Psychophysical methods can never assure us that a measure at a particular
time represents the best a person can do. But they do assure us that the
person can discriminate _at least as well_ as the method indicates. And
that's what's important.
Rick seems to have an _ide'e fixe_ that psychophysics means determining
the shape of some function p = f(s), where p is the magnitude of a perceptual
signal and s is some complex of sensory input. I have to grant that such
a function f is often called "the psychophysical function," but I think
that's a mistake. And it isn't what most psychophysicists are interested
in.
Martin
[From Rick Marken (980203.0830)]
Me:
Apparently, you believe that the behavioral illusion doesn't
apply to the results of psychophysical experiments
Martin Taylor (980203 01:02)--
They don't tell anything about the perceptual function as such,
because they aren't designed to do anything more than determine
whether the subject can distinguish one perception form another
This doesn't solve the behavioral illusion problem. You still
measure a person's ability to distinguish one perception from
another by looking at the relationship between the person's
responses and the stimulus correlates of the perceptions being
distinguished. If the person is controlling for, say, avoiding
the draft (like most of my friends and myself in the 60s -- I
was close to being a Canadian like you) then your measure of
the person's ability to distinguish perceptions (like tones)
may differ substantially from the person's actual ability to
make such distinctions. This is the behavioral illusion: the
observed relationship between inputs and discriminative responses
does not reflect an actual characteristic of the behaving
system -- the relationship between its inputs and its perceptions.
Yes, it's true, Martin. Everything you think you discovered
in psychophysics is just an illusion. But, hey, the same is
true for me; that gorgeous study of auditory filter bandwidth
that I did as my PhD thesis: illusion. But don't worry, Martin.
Now you understand PCT so you can start studying perceptual
control systems the right way -- using the Test to determine the
perceptual variables that organisms control.
I'm intrigued, then, as to why you felt it reasonable to use the
results of "conventional behavioural research" in your "Hierarchy
of Perception" paper--or is that another one of your now disowned
products of a mis-spent youth?
First, I used it because this research was _suggestive_; it was
easy to _guess_ at the perceptual variables that the subjects
_might_ be controlling in these experiments. If the subjects
actually were controlling these variables, then that could
_possibly_ explain the results. The "Hierarchy" paper was
theoretical; it suggested the kind of (PCT type) research that
could be done to test these theoretical ideas. The "Hierarchy of
Perception and Control" demo at:
http://home.earthlink.net/~rmarken/demos.html
is an example of the kind of PCT type research that can be done to
test some of the ideas in the "Hierarchy..." paper (which is
available at http://www.ed.uiuc.edu/csg/documents/behave_p.erc.html)
Second, the only thing wrong with conventional research is taking
the results at "face value". I don't think we should ignore the
conventional research; it is a good sources of _guesses_ about
the variables that organisms _might_ be controlling. But before
concluding (on the basis of this research) that we have now learned
something about control, we should defintely _redo_ the research
using PCT methodology. This is what Bill is trying to have Bruce
Abbott do with the operant research. Taking that research at
face value suggested that reinforcement rate is a controlled
variable. Redoing this research has revealed that this is not
the case at all; it turns out that we need to start the study
of feeding all over again from a PCT perspective if we want to
know what organisms control in operant type situations.
Third, I always try to cite some conventional research when
I want to publish papers on PCT because reviewers and editors
want to see how what you are doing might relate to some existing
research thread -- especially one that has appeared in the
journal in which you want to publish. I think my inclusion
of the McBeath et al study in my "Dancer..." paper helped get it
published in Psych Methods. I am finding that there were some
major shortcomings in this study -- in particular, they didn't
measure all the possible controlled variables that I would like;
but at least the did get a "picture" of a hypothetical controlled
input (retinal trace -- as in my baseball demo) so it's still
closer to PCT research than most of what you find in the
conventional literature.
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken
[From Bruce Abbott (980203.1840 EST)]
Rick Marken (980203.0830) to Martin Taylor --
This doesn't solve the behavioral illusion problem. You still
measure a person's ability to distinguish one perception from
another by looking at the relationship between the person's
responses and the stimulus correlates of the perceptions being
distinguished.
Yes, it's true, Martin. Everything you think you discovered
in psychophysics is just an illusion.
Like Martin, I'm having a lot of trouble with this line of reasoning.
When serving as a participant in a psychophysical study, I am controlling
for producing statements that accurately reflect the state of the relevant
perception (such as whether I can hear a tone or not on each trial). This
is basically a tracking task: I am in effect verbally keeping my statements
"aligned" with my perception of the target, just as I might move a mouse to
keep a cursor aligned with a moving target. In the latter case, if I am
doing a good job, the movements of the cursor I produce will accurately
reflect the position of the target, as I perceive it, at any given moment in
time. Thus, if you want to know where I perceive the target to be, just
look at where the cursor is. Similarly, you can expect that, if I am doing
a good job of keeping my statements aligned with my perception of the tone,
my statements will accurately reflect the state of my perception (e.g.,
whether I can perceive the tone or not). If you want to know whether I
perceive the tone, just listen to what I report.
By varying the tone intensity, an investigator could determine the limits of
my ability to sense the presence of the tone (I will be unable to keep my
statements accurately aligned with the actual presence/absence of the tone,
although of course I will continue to keep my statements accurately aligned
with my perceptions of same. When I can accurately perceive (and thus
accurately report) tone presence/absence, then it is true that the
variations in my output (saying "present" or "absent" will merely reflect
the environmental function that relates saying "present" or "absent" to the
presence/absence of the tone. But this is not the relevant issue. What is
relevant is that whether I am able to correctly report the true state of the
signal (present or absent on a given trial) depends on the acuity of my
hearing, just as my ability to keep the cursor centered over a moving target
depends on my ability to correctly perceive the relative positions of the
cursor and target. This says something important about my limits as a
system attempting to control a relationship between the _actual_ state of
the tone and my reports of its state; it is not a manifestation of the
behavioral illusion.
Regards,
Bruce
[From Rick Marken (980203.1950)]
Me:
Yes, it's true, Martin. Everything you think you discovered
in psychophysics is just an illusion.
Bruce Abbott (980203.1840 EST) --
Like Martin, I'm having a lot of trouble with this line of reasoning.
I think you understand the reasoning; you just can't accept the
conclusion. For example, you know that:
When serving as a participant in a psychophysical study, I am
controlling for producing statements that accurately reflect
the state of the relevant perception
So you know that, in a psychophysical experiment, you are controlling
a perception (p) that depends on both environmental stimuli (s) and
you own outputs (o):
(1) p = f(g(o) + h(s))
where g() is the "feedback function" relating your own output
to the controlled input; h() is the function relating the stimulus
to the controlled input and f() is the perceptual function,
relating controlled input to controlled perception.
You even understand that:
This is basically a tracking task:
So you know that you are trying to keep the perception at
a target (reference) level. So you know that the controller
in this experiment can be modeled by the following equation:
(2) o = k (r-p)
Where k() is the "organism" function mapping error into output.
You probably even know that when you perform this experiment,
the observed relationship between o and s is proportional to
-h()/g() (both characteristics of the environment). So the
observed relationship between stimuli (s) and responses (o)
in a psychophysical experiment reflects characteristics of
the environment [h()/g()], not characteristics of the organism
[f() and k()]. Concluding that the results of such experiments
_do_ tell you about the organism [f() and k()] is the behavioral
illusion.
By varying the tone intensity, an investigator could determine
the limits of my ability to sense the presence of the tone
Surprisingly, no (it's a compelling illusion -- like the illusion
that the earth is flat). To determine the limits of your ability
to sense the presense of a tone the investigator would have to
be able to determine the function f(h()) relating tone (s) to
perception (p). But we have seen that all we get from this
experiment is h()/g(), not h() or f(h()). And the h() and
g() we get depend on what perception (p) you actually happen
to be controlling. You say you would be controlling "...statements
that accurately reflect the state of the relevant perception". But
someone else may be controlling something different (as my friends
were when they went to get the hearing test at their draft
physical).
So it may _seem_ like you are determining the "ability to sense
the presence of the tone" in these experiments, but you're not.
The appearance that you determinng this is an _illusion_ (if
your subject is a control system); you are measuring the relative
effects of the outputs and disturbances (stimuli) on _some_
(undetermined in most psychophysical experiments) controlled
variable.
Best
Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/
[From Bruce Nevin (980203.2049 PST)]
Rick Marken (980203.1950)--
Please help me out, I'm trying to follow along here. What's the distinction
between controlled input and controlled perception?
in a psychophysical experiment, you are controlling
a perception (p) that depends on both environmental stimuli (s) and
you own outputs (o):(1) p = f(g(o) + h(s))
g() is the "feedback function" relating o to the controlled input (in
environment)
h() is the function relating the s to the controlled input (in environment)
f() is the perceptual function relating controlled input to controlled
perception
[From Tim Carey (980204.1810)]
[From Rick Marken (980203.1950)]
So it may _seem_ like you are determining the "ability to sense
the presence of the tone" in these experiments, but you're not.
The appearance that you determinng this is an _illusion_ (if
your subject is a control system); you are measuring the relative
effects of the outputs and disturbances (stimuli) on _some_
(undetermined in most psychophysical experiments) controlled
variable.
Hi Rick, this is a fascinating thread ... I'm trying to follow along. Is
what you're saying this: when we see someone repsond to a stimulus in a
psychophysical experiment, all we are seeing are actions opposing
disturbances? Clearly, this opposition to disturbances has to do with
controlling a perception but these experiments don't touch this aspect at
all. Along the "inside/outside" boundaries of the system, these experiments
only focus on the outside.
Is this anything like what you're saying or am I way off base?
Cheers,
Tim
[From Bruce Abbott (980204.0505 EST)] <---way too early
Rick Marken (980203.1950 LALA TIME) --
Bruce Abbott (980203.1840 EST)
I think you understand the reasoning; you just can't accept the
conclusion. For example, you know that:
I understand the reasoning and therefore can't accept _your_ conclusion.
When serving as a participant in a psychophysical study, I am
controlling for producing statements that accurately reflect
the state of the relevant perceptionSo you know that, in a psychophysical experiment, you are controlling
a perception (p) that depends on both environmental stimuli (s) and
you own outputs (o):(1) p = f(g(o) + h(s))
where g() is the "feedback function" relating your own output
to the controlled input; h() is the function relating the stimulus
to the controlled input and f() is the perceptual function,
relating controlled input to controlled perception.
In the psychophysical example I used, we get into logic-level control and
the exposition becomes rather complicated. To keep things relatively
simple, imagine a psychophysical experiment in which the person is asked to
rotate a knob so as to keep a tone at a level that, for this person, is just
at the margin between perceptible and imperceptible, while the apparatus
applies a smoothly-varying disturbance to the tone's physical intensity. In
that case,
(1) i = g(o) + h(s)
where i is the physical intensity of the tone, o is the knob setting, g is
the function giving the influence of knob setting on tone intensity, s is an
intensity setting that is varying independently of the participant's actions
(disturbance) and h is the function giving the influence of this disturbance
on the tone's intensity.
(2) p = f(i)
This is the perceptual input function (PIF) relating the tone's physical
intensity to its perceived intensity. It is a property of the person
(system), not of the environment.
Putting (1) and (2) together gives Rick's equation: p = f(g(o) + h(s))
So you know that you are trying to keep the perception at
a target (reference) level. So you know that the controller
in this experiment can be modeled by the following equation:
(2) o = k(r-p)
Where k() is the "organism" function mapping error into output.
Yes.
You probably even know that when you perform this experiment,
the observed relationship between o and s is proportional to
-h()/g() (both characteristics of the environment). So the
observed relationship between stimuli (s) and responses (o)
in a psychophysical experiment reflects characteristics of
the environment [h()/g()], not characteristics of the organism
[f() and k()]. Concluding that the results of such experiments
_do_ tell you about the organism [f() and k()] is the behavioral
illusion.
Yes. So?
By varying the tone intensity, an investigator could determine
the limits of my ability to sense the presence of the tone
Surprisingly, no (it's a compelling illusion -- like the illusion
that the earth is flat). To determine the limits of your ability
to sense the presense of a tone the investigator would have to
be able to determine the function f(h()) relating tone (s) to
perception (p). But we have seen that all we get from this
experiment is h()/g(), not h() or f(h()). And the h() and
g() we get depend on what perception (p) you actually happen
to be controlling.
It is not the point of the experiment to find the function k coverting error
into output. It is the point of the experiment to find out something about
f(i).
In this example, I am controlling the perception of tone intensity, and,
following instructions, I am attempting to keep the tone at a level that is
just on the edge of perceptibility (r), against the disturbances to that
perception produced by the apparatus. You will find that my output pattern
will be the inverse of the disturbance pattern (allowing for scaling
differences). But you will also discover the intensity of tone that I can
just barely perceive (the delivered tone intensity will be varying closely
around this value), and that is a property of my perceptual input function,
not of the environment. In further work, you will discover that when you
ask me to maintain the tone at a particular intensity that is well below
this value, I will be unable to vary my output so as to track the changing
physical intensity of the tone, because within this range of values the
function f(i) is flat: all values of i yield p = 0. The intensity at which
the function f(i) ceases to be flat will be at the r inferred from the tone
intensity which the participant maintains against disturbances.
In yet other experiments, I could similarly determine the amount of _change_
in the tone's intensity that you could just barely detect, and could do so
at a variety of absolute tone intensities. Your ability to discriminate
such changes certainly must impact on your ability to maintain the tone's
intensity at a particular value against very small disturbances to that
intensity, and that ability is a characteristic of you, not the environment.
Making such determinations is what psychophysics is all about.
So you see, Rick, when I perform psychophysical studies I am learning a good
deal about the characteristics of the organism, contrary your assertion that
I am only learning about the characteristics of the environment. This is no
"behavioral illusion."
Regards,
Bruce
[Martin Taylor 980204 09:50]
Rick Marken (980203.0830)
If the person is controlling for, say, avoiding
the draft (like most of my friends and myself in the 60s -- I
was close to being a Canadian like you) then your measure of
the person's ability to distinguish perceptions (like tones)
may differ substantially from the person's actual ability to
make such distinctions. This is the behavioral illusion: the
observed relationship between inputs and discriminative responses
does not reflect an actual characteristic of the behaving
system -- the relationship between its inputs and its perceptions.
At least we can agree on something. This is the reason why I have
emphasized in each message the importance of treating the results of
discrimination experiments as limits--the person can do at least as well
as the experiment indicates. You're saying nothing I haven't said over
and over. The person can do worse than they are capable of, but can't
do better.
The exact same problem applies to using the Test in an approved PCT
experiment. If the subject decides that s/he isn't going to control
that perception as well as s/he could, there's not a thing the
experimenter can do about it. The controlled perception changes from
the one the experimenter is trying to test (the Test says "yes, there's
some control of something like this, but it's not good") to the subject's
perception of how well they are controlling the one apparently being
tested. How can the experimenter tell--and furthermore, how does this
square with the idea that the objective of PCT is, to paraphrase
(or quote?) you "to discover the controlled perception"?
There are all sorts of ways, in a properly designed psychophysical
discrimination study, to check whether the subject is doing their best.
Usually, an experimenter won't bother, because the assumption is that the
subject is controlling for pleasing the experimenter by trying hard. The
same, I would guess, applies in using the Test.
Yes, it's true, Martin. Everything you think you discovered
in psychophysics is just an illusion.
The world is but illusion. So what? The results still put limits on how
well people can discriminate the tested variables, and those limits
put bounds on how well they could control the same variables if they
wanted to.
I was re-reading yesterday a series of papers from the early 60's that
I think would be relevant to PCT, by Richard Held and his collaborators,
about the "adaptation" to prism spectables and similar transforming
devices for hearing. (Any guesses as to why I was re-reading those
particular papers?) The papers all were testing the proposition that
the "adaptation" can be essentially complete, but only if the subject
actively interacted with the environment. If the subject is simply
exposed to essentially the same "stimulus" changes, no adaptation
occurs. This seems to be equally true of kittens and humans (but
chickens, if I remember some other paper from that era, don't seem
to adapt to prism spectacles, even if they go hungry because they miss
pecking the seed they are trying to eat).
The effects seem to hold for each subject tested. They aren't PCT
studies. I think they are both explicable using straightforward PCT
reorganization, and valid in a world seen through PCT prism spectacles.
The PCT question would be (if my memory is correct) what it is about
chickens or the task they had to do (eat seeds off the ground) that
did not allow them to "adapt" to the spectacles.
I'd still like you to explain what it is in the environment that tastes
salt and lets the subject know that they should report _they_ taste it
when it is actually the environment that did the tasting:-)
Martin
[From Bruce Abbott (980204.1325 EST)]
Bruce Nevin (980203.2049 PST) --
Rick Marken (980203.1950)
Please help me out, I'm trying to follow along here. What's the distinction
between controlled input and controlled perception?
in a psychophysical experiment, you are controlling
a perception (p) that depends on both environmental stimuli (s) and
you own outputs (o):
(1) p = f(g(o) + h(s))
g() is the "feedback function" relating o to the controlled input (in
environment)
h() is the function relating the s to the controlled input (in environment)
f() is the perceptual function relating controlled input to controlled
perception
As Rick seems to be busy elsewhere, I'll respond. Strictly speaking, the
input is not controlled, only the perception is. However, if i and p are
perfectly related (as when f is a linear function with nonzero slope),
controlling p at some reference level will also appear to control i at some
reference level. Where the "connection" between i and p is less reliable
(e.g., the value of p produced by a given level of i "drifts" over time), p
will be maintained near its reference but i will appear to drift rather than
always staying near some particular value. (I am assuming that the
reference for p is constant.) Thus p will be controlled but i will not be
(or at least, it will not appear to be controlled very well).
Regards,
Bruce
[From Bruce Nevin (980204.1337 EST)]
Bruce Abbott (980204.1325 EST)--
if i and p are
perfectly related (as when f is a linear function with nonzero slope),
controlling p at some reference level will also appear to control i at some
reference level. Where the "connection" between i and p is less reliable
(e.g., the value of p produced by a given level of i "drifts" over time), p
will be maintained near its reference but i will appear to drift rather than
always staying near some particular value.
[...] Thus p will be controlled but i will not be
(or at least, it will not appear to be controlled very well).
Thanks Bruce! I assume that i is the same as s in Rick's formula. Making
the substitution:
p = f(g(o) + h(i))
p is the controlled perception
f() is the perceptual function relating controlled input to controlled
perception
g() is the envt "feedback function" relating o to the controlled input
h() is the envt function relating the s to the controlled input
o is the envt behavioral output, the organism's means of controlling p.
i is the envt stimulus, the observer's means of inferring control of p,
the "controlled input".
p and f() are properties of the organism.
If p is a first-order perception, then is f(() always a linear function
with nonzero slope? Or are intensity receptors more variable than this? Or
does this refer to fatigue, inurement, and so on?
If p is not a first-order perception, we're talking about the relation
between the participant's and the observer's nth-order perceptions of
i--setting aside the question of their observing the same aspect of the
complex variable in their shared environment (CEV). We could pretend that
the effect of the proximate input function f() could be aggregated together
with the effects of all lower-order control systems (with their functions
f(), g(), and h()) as though they were a single input function F(); i is
then a multiple term, all the inputs i to the first-order control systems
contributing to F().
Does it follow that psychophysical research ignoring control is useful only
for first-order perceptions?
Since we can only infer that f or F is linear or not (we cannot observe f
directly, much less F), what does this say about using i (or s) to infer p
or anything else internal to the organism?
Does this say that a model with near 100% fidelity is the only known basis
for inferring f, g, h, and p given i and o (S and R)?
Bruce Nevin
[From Bruce Abbott (980205.0945 EST)]
Rick Marken (980204.2210) --
Bruce Abbott (980204.2000 EST)
Think, Rick. How would the relationship between s and o change
if for every value of i, the PIF returned zero?
I have no idea what you are talking about or getting at. All we
observe in a psychophysical experiment is the relationship between
s (disturbance) and o (response). This relationship tells you
nothing about the perceptual function or any other function
in the organism under study.
I think, Sarsi, that you do know what I am getting at, but have chosen to
stonewall. You did not even attempt to answer my very simple and
straight-forward question, whose answer holds the key to the problem.
Instead, you reiterate your assertion, as if mere repetition would make that
true statement apply to the experimental result I described. We are not
talking, dear Sarsi, about the ordinary relation between s and o, or whether
it tells you anything about the perceptual function, despite your continued
insistance that we are. We are talking about what happens to the
relationship between s and o when the perceptual input mechanism can no
longer detect variation in the external quantity whose perceptual
representation is to be controlled. I have told you this, but you refuse
even to mention it in your reply, lest it destroy any apparent credibility
your argument (actually, assertion, as you have offered no argument) might
seem to have.
What you have said above is equivalent to "The behavioral illusion
does occur in psychophysical experiments; but my argument is that
the behavioral illusion does not occur in psychophysical experiments".
No, it is not, Sarsi, and repeating that falsehood will not make it true,
should you do so a thousand times.
This is the same argument you used in your review of my "Dancer..."
paper. What you said in that review was essentially "You have shown
that the behavioral illusion does occur in conventional psychology
experiments but you have failed to show that you cannot get useful
information about the organism from these experiments". Again,
you are saying that the behavioral illusion does occur in
conventional psychology experiments but that it doesn't.
I do not recall saying any such thing, dear Sarsi. It is, perhaps, what you
wish to have me saying, for then my statement would be easy to disprove.
[For everyone else, Rick sent me a draft of his Dancer manuscript for
comment prior to submission; this is the "review" he speaks of.]
So I give. I defer to your truly Talmudic reasoning skills.
And I, dear Sarsi, lay prostrate before your unmatched ability to avoid
having to conclude the obvious when the obvious conflicts with your prior
belief.
Regards,
Bruce
[From Bruce Abbott (980205.1355 EST)]
Bruce Nevin (980204.1337 EST) --
Bruce Abbott (980204.1325 EST)
if i and p are
perfectly related (as when f is a linear function with nonzero slope),
controlling p at some reference level will also appear to control i at some
reference level. Where the "connection" between i and p is less reliable
(e.g., the value of p produced by a given level of i "drifts" over time), p
will be maintained near its reference but i will appear to drift rather than
always staying near some particular value.
[...] Thus p will be controlled but i will not be
(or at least, it will not appear to be controlled very well).
Thanks Bruce! I assume that i is the same as s in Rick's formula. Making
the substitution:
p = f(g(o) + h(i))
No, I am sticking to Rick's symbols. I am using i (for "input") to
represent the proximal value (at the sensory interface) of the external
variable (such as "position of cursor") whose perceptual representation is
being controlled. This external variable is sometimes labeled the "CV"
(controlled variable), but this is something of a misnomer as it is only the
perceptual representation of the CV that is being controlled. This
representation is called the perceptual signal and labeled as p. For
simplicity, the value of the CV and i are normally taken to be the same (see
diagram below). So, to correct your substitution, we have
i = g(o) + h(s)
and
p = f(i)
Here is the complete diagram:
r
p | e
[Input Function]----->[comparator]------>[Output Function]
^ |
> i | o
> g(o) |
+-----------[CV]<-----[Feedback Func.]<----+
^
> h(s)
[Disturbance Func.]
>
s
Input function : p = f(i) p = perception
Output function: o = k(e) o = output or "action"
Feedback function: ? = g(o) e = error = r - p
Disturbance function: ? = h(s) g(o) = effect of output on CV
CV function: i = CV h(s) = effect of disturbance on CV
i = representation of CV at input
r = reference
I used ? in two places where Rick did not supply a letter to stand for the
variable in question.
If the questions you asked still apply after this correction, please ask
again and I'll provide answers if I can.
Regards,
Bruce