[From Bruce Abbott (2017.10.17.2020 EDT)]
Rick Marken (2017.10.17.0950) –
Bruce Abbott (2017.10.16.1505 EDT)
RM: So that’s it. Like Powers I don’t reject causality but I do reject the cause-effect model that is the basis of experimental psychology. That is, I agree heartily with Bill’s statement above: "Not only the old cause-effect model breaks down when one is dealing with an N system, the very basis of experimental psychology breaks down also".
RM: I’d like to hear what you have to say about this too, Bruce A. Do you also agree with Bill, Martin and me that the cause-effect model of behavior breaks down when one is dealing with a negative feedback control system?
BA: …Your contention [that the disturbance is not a cause of output] is apparently was based on the idea that, if the output counteracts the disturbance, then the disturbance has had no causal influence on the CV
RM: No, my contention that the disturbance is not the cause of output in a control loop is based simply on the fact that the output of the system is caused by the CV, the value of which is determined by the combination of effect disturbance and output. .
BA: The interested reader (if any such exist, which I doubt) will note that in order to appear to be disagreeing with me, Rick has changed my “a cause� to “the cause.� Given what I actually did write, he is actually agreeing with me.
BA: I pointed out that for most folks, both causes would be considered to be acting on the CV, even when the net effect was zero.
RM: As do I. The value of the CV is caused by both the disturbance and output.
BA: O.K., let’s proceed on that basis.
BA: I also noted that in proportional control systems the output’s effect on the CV does not completely offset the effect of a disturbance, because some error must remain in order to produce the output necessary to oppose the disturbance, and error depends on the difference between reference and perceptual signal.
RM: Right, it’s the difference between reference and perceptual signal, the value of the latter representing the combined effects of disturbance and output. So the discrepancy between perception and reference is not caused by just the disturbance so that the output caused by the discrepancy is always a result of an unknown combination of the effects of disturbance and output on the perception.
BA: I deal with that issue in my post as I develop the argument for my position. Apparently, at this point in writing your reply, you have not yet read my argument, let alone consider whether its reasoning is sound. If would save the reader a lot of useless reading if you would first read and try to understand the entire argument before you attempt a reply.
BA: You have also asserted that the CV is the only cause of the output, because in a circular loop of causality, a change in the CV is the only reason for a change in the output, which then feedback to affect the CV itself. However, given steady parameters and a fixed reference value, the only reason why the CV changes is the effect of the disturbance on the CV.
RM: No, the system is continuously varying its output. So the reason the CV changes is due to the combined effect of disturbance and output. That is, it’s always true the q.i = g(o) + h(d). There is never a time when the only reason q.i (the CV) changes is due to the effect of the disturbance alone.
BA: Yes, it is always true that q.i = g(o) + h(d). But there is indeed a time when a change in q.i. is due entirely to a change in the disturbance. This is when the system is initially in a steady state. When the disturbance changes, this change propagates around the loop to the output, which then also changes. Although q.i is the sum of both g(o) and h(d), I think you will agree that if o is initially constant, then the change in o must be due (via the causal chain) to the change in d.
BA: This effect propagates around the loop to change the output, and to the extent that this change of output fails to completely oppose the effect of the disturbance on a given iteration of the loop, the residual effect of the disturbance continues to propagate around the loop to produce yet more change of output, until the effect of the disturbance is almost (but not quite) entirely opposed. Thus, the only independent effect on the output is the effect of the disturbance.
RM: Again, this analysis assumes that the disturbance is the only cause of the changes in CV that cause the outputs that compensate for the disturbance. In fact, this is never true; the output itself is always causing variations in the CV along with the disturbance.
BA: How can this be “never true� when I just described a case in which it is true? Pay attention!
BA: It is true that q.i = g(o) + h(d). But it is also true that change in q.i = change in g(o) + change in h(d). If g(o) is constant, then change in q.i = change in h(d). So your argument against mine must be that g(o) is never constant, ever: “the output itself is always causing variations in the CV along with the disturbance.� This assertion simply ignores my analysis. In case you missed it, I reproduce it below . Try to reason through it this time.
BA: This effect propagates around the loop to change the output, and to the extent that this change of output fails to completely oppose the effect of the disturbance on a given iteration of the loop, the residual effect of the disturbance continues to propagate around the loop to produce yet more change of output, until the effect of the disturbance is almost (but not quite) entirely opposed. Thus, the only independent effect on the output is the effect of the disturbance.
BA: But perhaps you agree that this happens in the special case in which a constant disturbance is suddenly imposed. In that case your claim is that it does not hold true generally. That this may be your view is suggested by the following:
RM: There is no way to disentangle these two simultaneous causes of the state of the CV.
BA: But I just did, at least for the case of a suddenly applied constant disturbance. For the case in which the disturbance continues to vary over time, the effects of disturbance and output do indeed become entangled, with as-yet uncompensated effects of the disturbance producing variations in output that get added to in-progress changes in the disturbance. However, the ability of the analyst to disentangle these effects has no bearing on what is the ultimate source of those changes, which are the ongoing changes in the disturbance.
BA: Thus the only independent cause of a change in output is a change in the disturbance. (I am assuming a constant reference and constant system parameters.) If we add in a time-varying reference signal, then we have only two such independent causes.
RM: The output of a control system is simply aimed at bringing the CV to the reference; in the process it is opposing any effects of disturbances to that CV. The control system is not producing its output as a reaction to the disturbance; this is an illusion. It is producing output as a reaction to variations in the CV relative the reference signal.
BA: Yes, of course! But this truth has nothing to do with the analysis I have presented as to the source(s) of variation in the output. One can trace a causal chain from changes in the disturbance to changes in output. Because of negative feedback, the observed relationship between the two is approximately the inverse of the environmental feedback function. The behavioral illusion is not the idea that disturbances (“stimuli�) have no causal relationship to output (“responses�); my analysis shows that a causal relationship does in fact exist. The behavioral illusion is that the observed relationship is mistakenly viewed as a characteristic of the organism, rather than as is actually the case, mainly a characteristic of the environment.
BA: I think that you were at pains to “prove� that a causal relation between disturbance and output does not exist, because of a mistaken belief that this relationship would constitute the forward equation in a control system.
RM: It was no pain at all; Bill proved it in the 1978 Psych review paper. The output of a negative feedback control system is not caused by the disturbance to the controlled variable; it just looks that way.
BA: Bill proved no such thing. It is a sign of religious dogmatism when one ignores a sound analysis in order to continue to believe in an idea that has been demonstrated to be false. As I have shown, the output is caused by the disturbance to the controlled variable.  The illusion is that the causal relationship is a characteristic of the organism when in fact it is mainly a characteristic of the environment (the inverse of the feedback function).
BA: In fact it reflects mainly the inverse of the feedback function. Mistaking this inverse function (which is mainly a property of the environment) for the forward system equation is what Powers described as the “behavioral illusion.�
RM: Exactly!
BA: So we agree on that, at least.
BA: What Powers called the “cause-effect model of behavior� refers to the open-loop model of behavior, in which a cause (e.g., a stimulus) produces an effect (e.g. a response) in a direct causal chain not affected by feedback. In this model, the relation between cause and effect is a characteristic of the inner workings of the organism.
RM: Right. And that is the old cause-effect model that is the basis of experimental psychology that breaks down when one is dealing with an "N system ".
BA: If the stimulus functions as a disturbance to a controlled variable, and the response functions as the control system’s change of output due to the disturbance, then the open-loop, cause-effect model is incorrect;
RM: Right, if the system under study is an “N-system” – a negative feedback control system – then the open-loop, cause-effect model is incorrect.
BA: In such a case I agree with Powers and reject this misleadingly labeled cause-effect model of behavior. (Calling this the “open loop model of behavior� rather than the “causal� or “cause-effect� model would avoid the misleading implications that one is rejecting all causal models.)
RM: OK, so it’s the open-loop cause-effect model that Bill is talking about. So then do you agree with Powers that when we are dealing with an N-systems – closed-loop control systems, which all organisms are – not only the old open-loop cause effect model of behavior breaks down but the very basis of experimental psychology breaks down as well?
BA: No. The old open-loop, cause-effect model of behavior only breaks down in the context of a closed-loop system for which the variable in question serves as a disturbance. PCT holds that behavior is the control of perception; this implies that behavior observed in a psychological experiment must be the consequence of a disturbance to a controlled variable, either directly or indirectly. Thus, every experiment must get the participant to cooperate in controlling some variable as the means to elicit participant behavior. But although this behavior serves to control the CV, this typically is incidental to the purpose of the experiment.Â
BA: For example, an experimenter may present a consonant trigram (e.g., CGJ) and ask the participant to repeat it after a delay period during which she has been kept engaged In doing mental arithmetic. (The purpose of this is to minimize the participant’s opportunity to silently rehearse the trigram.) This is a classic test of short-term memory; the results demonstrate that the ability to correctly recall the trigram falls off steeply with time (Peterson & Peterson, 1959).
BA: The experimental instructions ask the participant to recall the trigram after a specific delay, having continuously performed mental arithmetic in the interim. If the participant cooperates, then she will control for doing the mental arithmetic during the delay and for repeating the trigram (if possible) at the end. Hearing the trigram is the cue to start doing the arithmetic and a cue given at the end of the interval is the cue to stop doing the arithmetic and try to repeat the trigram. One can describe this in terms of program-level control. But the ability to recall the trigram is not the controlled variable either during the interval or after it is over. It is simply a characteristic of the participant’s memory system, and the relationship between delay and accuracy of recall is not an example of the behavioral illusion. There are probably innumerable examples like this one that do not involve the behavioral illusion; consequently I cannot agree that the behavioral illusion “breaks down the entire basis of experimental psychology.� This is far too sweeping a claim.
Bruce
