i am reading powers 1983 method for deriving transfer functions (p145, living systems x?) which is an iterative one rather than analytic: a transfer function is assumed and progressively corrected as one takes input, runs it thru the ansatz and compares real output with predicted output.
anyone to comment on why/whether such method may or may not fall in the trap of the behavioral illusion as defined in powers 1978 paper, namelly, feedback function g taken as organism function f?
AGM: i am reading powers 1983 method for deriving transfer functions (p145, Â living systems x?) which is an iterative one rather than analytic: a transfer function is assumed and progressively corrected as one takes input, runs it thru the ansatz and compares real output with predicted output.
AGM: anyone to comment on why/whether such method may or may not fall in the trap of the behavioral illusion as defined in powers 1978 paper, namelly, feedback function g taken as organism function f?Â
RM: Excellent question. The answer is that Bill’s algorithm for deriving the input-output transfer function is implemented in a negative feedback control loop. Note that the weights that define the transfer function are adjusted based on the change in the size of the error signal in the control loop of which the transfer function is a component. There is no behavioral illusion because the transfer function is being derived for what is known to be a closed-loop control system with a constant feedback function (and, a constant reference signal as well).Â
RM: The behavioral illusion described in the 1978 paper refers to the mistaken interpretation of the disturbance-output relationships seen in the behavior of a control (N) system as the transfer function (behavioral law) that relates input to output in an open loop (Z) system. Since scientific (experimental) psychology is based on the assumption that observed input - output relationships reveal something about the psychology of organisms, the entire edifice of scientific psychology collapses if the organisms under study are, indeed, input control (N) rather than input-output (Z) systems. So it’s little wonder that nearly all scientific psychologists continue to ignore, misinterpret or disdain Powers’ brilliant “spadework” described in Powers (1978).
BestÂ
Rick
–
Richard S. MarkenÂ
"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.�
                --Antoine de Saint-Exupery
It can't, in principle, fall into the behavioural illusion trap,
because the f(t) that it computes is a whole-loop transfer function.
The behavioural illusion is the mis-partitioning of f(t) into
g(t)*x(t), where g(t) is the internal function and x(t) the external
environmental path function. Powers does not address that
partitioning in the paper you reference, so far as I can see.
The 1983 paper (it’s in LCS II) is interesting in its own right,
because it describes the technique used in Powers’s 1994 “Artificial
Cerebellum” (AC), which is a function that “whitens” the loop
transfer function when it is used as the output function for a
control loop. The AC can be seen as w(t) = (g(t)*x(t)).
In the 1983 paper, Powers implicitly worries about the degrees of
freedom involved in the computation of f(t), in that he allows 256
sample delays (“t” in this message, “tau” in the paper) but is
concerned about the system being overdetermined because there are
2304 data points used in the adaptation. He should not have worried.
He says that his disturbance had a bandwidth of around 0.2 HZ, which
means it has only one degree of freedom every 2.5 seconds (1/2W). If
his sampling rate for f(t) was the same as his presentation sampling
rate (30 per second), his f(t) had just over 8 degrees of freedom in
total. That’s an awful lot of overdetermination, so there are a lot of
degrees of freedom in the result that are useful mainly for
estimating the measurement noise level per degree of freedom. Since
his adaptation procedure for each of the 256 samples is simply the
convergence of a simple control loop, he needed enough data to allow
that to happen. The degree of success of that convergence is the
measurement error in question.
Powers used a polynomial fit, so if there were only about 8 degrees
of freedom he should have found no more than 8 of the coefficients
appreciably greater than zero. It would be interesting to repeat
Powers’s experiment and do a principal components analysis of the
256 sample values for f(t). He didn’t have the computational power
to do such an analysis, but we do now. I would assume that if we did
that, the significant components would yield smooth and orthogonal
functions of time. The shapes of the functions would form a basis
space for representing a hierarchy of control loops mathematically
analogous to a Fourier analysis for representing single waveforms.
Many years ago Louis Pols and I analyzed speech waveforms in much
the same way to see if there were differences between language, and
if a fluent bilingual changed pattern between languages. We
discovered that the first few (I think four) principal components of
the log spectra were the first four cosine and sine components of a
Fourier series, and the next component differed between Dutch and
English. If I remember correctly (this was never published) our
speaker who sounded native to speakers of both languages but was
born in the Netherlands had the Dutch pattern for that component.
I mention the speech analysis because it is likely that a speech
perceiving front-end would take advantage of that kind of
regularity. So would a hierarchy that had reorganized to perform
tracking tasks every day, for example, picking up and placing
things. If a PCA basis space for control could provide hints about
important levels of control in a simple tracking study, it would
open a new possible line of research in PCT, especially in the area
of interpersonal differences and the effects of brain damage on
mobility and dexterity. I think the data Max reported in the recent
paper would be ideal for such an exploratory analysis.
Martin
···
On 2017/09/5 5:02 AM, Alex Gomez-Marin
wrote:
i am reading powers 1983 method for deriving
transfer functions (p145, living systems x?) which is an
iterative one rather than analytic: a transfer function is
assumed and progressively corrected as one takes input, runs it
thru the ansatz and compares real output with predicted output.
anyone to comment on why/whether such method may
or may not fall in the trap of the behavioral illusion as
defined in powers 1978 paper, namelly, feedback function g taken
as organism function f?
AGM: anyone to comment on why/whether such method may
or may not fall in the trap of the behavioral illusion as
defined in powers 1978 paper, namelly, feedback function g taken
as organism function f?Â
Â
MT: The behavioural illusion is the mis-partitioning of f(t) into
g(t)*x(t), where g(t) is the internal function and x(t) the external
environmental path function.
RM: This is not correct. The behavioral illusion is seeing an observed relationship between a disturbance (stimulus input) to a controlled variable and the action (response output) that compensates for that disturbance as a reflection of the “organism transfer function” – the psychological processes that convert stimulus input into response output. So the behavioral illusion is the appearance that an organism is an input-output (Z) system when it is in fact an input control (N) system. The term “behavioral illusion” also refers to the fact that the observed function relating input to output – the function that is presumed to reflect a psychological characteristic of the organism  – is the inverse of the feedback function connecting the organism’s output to the input variable that the organism is controlling – the controlled variable. This is all illustrated in my “Behavioral Illusion” demo at http://www.mindreadings.com/ControlDemo/Illusion.html. Â
RM: The behavioral illusion is important because it is what makes PCT revolutionary. It does this by showing that scientific psychology has been based on a fundamentally wrong idea about how organisms work – the idea that sensory input is the cause of behavioral output – and showing why this idea has persisted (because it looks like output is caused by input when the variable under control – the controlled variable – is ignored or unnoticed). This wrong idea about how organisms work has led to an approach to behavioral research that is aimed at finding causal variables that don’t actually exist while ignoring (or denying the existence of) controlled variables – the variables around which all purposeful behavior is organized.Â
RM: The behavioral illusion shows that a whole new approach to behavioral research is required – one aimed at determining the variables organisms control and how they control them. So the PCT revolution is truly revolutionary because, as Powers said in his Foreword to my MIND READINGS collection “…if the phenomenon [purposeful behavior – control] you see here really works as this model shows it to work, then a whole segment of the scientific literature needs to be deposited in the wastebasket” (p. vii).
RM: I believe you have never really cared for the idea that PCT is revolutionary in this way (in a way that requires starting the project of understanding behavior largely from scratch) and your interpretation of the “behavioral illusion” is consistent with this belief. I find this unfortunate, not only because it deprives me of a compatible colleague, but also because of what Bill wrote two months before he passed away, in a paper entitled “The World According to PCT”  – a paper that apparently was (or is) to be a prologue to the papers in the still forthcoming LCS IV: Â
BP: We are now facing reality. This is going to be a revolution
whether we like it or not. There are going to be arguments, screaming and
yelling or cool and polite. It’s time to sink or swim.
RM: I know that it is unlikely, after nearly 30 years of seeing it as you do, that you will change your mind about the revolutionary nature of PCT with respect to scientific psychology but I hope that, in respect of Bill’s wishes, whatever you do publish in LCS IV is not going to be as “counter-revolutionary” as your interpretation of the behavioral illusion.Â
BestÂ
Rick
–
Richard S. MarkenÂ
"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.�
                --Antoine de Saint-Exupery
Just as I said in what you quoted, though I used more technical
language that is actually relevant to the context of Alex’s
question. In your answer to Alex, you totally missed that the paper
by Bill about which Alex asked actually had nothing to do with the
behavioural illusion at all.
I will let the rest of your calumnies lie uncorrected. Others can
judge for themselves how very little truth there is in it. My words
and actions over the last quarter century can speak for themselves.
Once upon a time, I used to ask you from time to time “Why do you
say that” when you wrote things that directly contradicted the
evidence at hand or that had been written in a message to which you
were responding. You never answered that question, so I am no longer
bothering. But it still might be of some mild interest to know.
Martin
···
Martin Taylor (2017.09.05.13.39)
AGM: anyone to comment on
why/whether such method may or may not fall in
the trap of the behavioral illusion as defined
in powers 1978 paper, namelly, feedback function
g taken as organism function f?
MT: The behavioural illusion is the
mis-partitioning of f(t) into g(t)*x(t), where g(t) is
the internal function and x(t) the external
environmental path function.
RM: This is not correct. The behavioral illusion is
seeing an observed relationship between a disturbance
(stimulus input) to a controlled variable and the action
(response output) that compensates for that disturbance as
a reflection of the “organism transfer function” – the
psychological processes that convert stimulus input into
response output.