[Martin Taylor 2010.03.18.17.21]
For some obscure reason, Rick asked me specifically to comment on...
[From Rick Marken (2010.03.16.2200)]
Bruce Gregory (2010.03.16.1800 EDT)]
You appear to have a very restricted view of psychological experiments. In
many experiments the subject would seem to be controlling the perception
"following the instructions."
Yes, that's surely one perception that is controlled. Indeed, these
instructions ask the subject to control a particular perception, such
as "press the key when the tone comes on". Since the subject must
translate the words of the instructions into a perception that they
themselves will control --and a reference for that perception-- it is
not always clear what perception the subject actually ends up
controlling. But nothing much would happen in the experiment unless
the subjects does control a perception that is something like the one
he or she is instructed to control.
I suspect there would be little disagreement with this among readers of CSGnet. But Rick's request for me to comment was in the context of demonstrating to the wider world that tracking studies are a good model of all behaviour, including behaviour in standard psychological experiments, so I'm taking a Devil's Advocate position in what follows.
------------Devil's Advocacy starts here--------------
If one starts by asserting that the subjects in the experiment are controlling a perception such as that they are following the experimenter's instructions, one has missed a few steps. But since this paragraph was actually a response within a CSGnet discussion, it's not the same as an exposition to the unwashed.
It is true that most experimenters to do not
attempt to disturb this perception to determine if it is in fact being
controlled.
No, they disturb the perception that the subject is instructed to
control.
The "stimulus" in a "stimulus-response" experiment always influences a potentially large number of perceptions, some of which may be being controlled by the subject. One of the aspects of a "clean" experiment is that the experimenter tries to set it up so that only one controlled perception is disturbed by the stimulus. (Parenthetically, I think this is why so few lab studies provide results that can be generalized to the real world).
Usually, the experiment itself provides "The Test" for the controlled variable, because if the subject's output did not correlate with the disturbance supplied by the experimenter, the experimenter would probably conclude either that the subject did not perceive the stimulus or that the subject was not following instructions. If the subject's output correlates well with the stimulus, then at least the experimenter has reason to believe the subject was following instructions. If the subject's output correlates only moderately with the stimulus, the experimenter might conclude that the subject was erratically following instructions or, more probably, that the subject found the stimulus hard to perceive reliably (which would be the actual reason for doing the experiment if it were a psychophysical study).
But again, it's hard to see how this kind of analysis would lead a classical psychologist or anyone else to see that a simple tracking study provides a model for all behaviour, or for the behaviour within the experiment.
In the reaction time task, for example, the onset of the tone
disturbs the perception of responding when the tone comes on; the
subject controls this perception (compensates for the tone being on
without the press) by pressing the key after the tone comes on. If
the subject is not controlling for pressing the key when the tone
comes on (and not pressing when the tone is not on) then he subject
will not press the key when the tone comes on.
I don't think a S-R psychologist would disagree with this. Such a psychologist might use different language, for example "In a reaction time study, the subject knows that the button shouldn't be pushed until the tone comes on, so he primes the response to occur only after the stimulus, and as soon after as possible." The problem is to change the the mindset that leads to the language, but changing the mindset isn't done by changing the language.
So the existence of a
controlled variable is a sine qua non of behavioral experiments.
But an S-R psychologist might not follow the logic to get to this claim, even if she might accept the control language in the quoted paragraph. I suspect that the most probable response to this claim would be along the lines of "You have said that you can describe what happens in this experiment in terms of controlled perceptions, and I agree that you can. But there's an awfully long step between showing that you can do so in this case and showing that you can't analyze even this case without analyzing the controlled perceptions, let alone showing that you can't analyze any behavioural experiment without dealing with controlled variables".
Again, I come back to the context of your request for me to comment on your message, that it is part of how you would get the wider world to see tracking experiments as a paradigm for all experiments, and for all behaviour.
But,
of course, it is never "the point" of these experiments and, as you
say, there is never any effort made to identify exactly what
perception each subject is actually controlling. Doing that would be
one way that research aimed at understanding control would differ from
conventional research aimed at discovering causal relationships
between variables.
I think you make a false dichotomy. Even within a complete PCT framework, not all research is aimed at understanding control, or even at determining what variables are being controlled at any one moment (despite that you have often claimed the contrary). Within a completely S-R framework, not all research is aimed at discovering causal relationships. Within either framework, the structure is often assumed, and the research is aimed at elucidating the properties of structural elements. That's why we map models of control units onto the results of human tracking. The model's structure is assumed, and the parameters altered to optimize the fit to what the human does. Research "aimed at understanding control" can go hand in hand with research "aimed at understanding the components of particular control systems".
Referring back to what I said earlier, if the subject's output correlates poorly with the disturbance, and the experimenter has every reason to believe that the subject is attempting to follow instructions, then the experimenter has good grounds to argue that the subject has difficulty perceiving the "stimulus|disturbance", whether the experimenter is a thoroughgoing PCT researcher or a thoroughgoing S-R psychologist. The analysis of how well the subject perceives the stimulus is the same in either case. The PCT researcher isn't trying to understand control better, and the S-R researcher isn't trying to discover a causal relationship. Both are trying to discover the properties of a signal path -- in the one case it's a component path in a control loop, in the other a component path in a one-way signal flow.
It's not clear to me how this addresses the public relations question of showing how tracking studies offer a model for understanding all experiments.
That, however, would not seem to make the data gathered
completely useless. If I administer a Raven's Progressive Matrices test to a
group of subjects, are you saying that the scores tell me nothing if I do
not realize that the subjects are hierarchical control systems? That seems
extreme, to say the very least.
I wouldn't say it tells you nothing. But it doesn't tell you much
about what the subject is controlling for, which might be something
you would want to know if your job was to teach someone how to do this
task.
Yes, if you are convinced of the PCT framwork, that's a question you might ask. How would you convince someone who had heard of PCT but was a bit sceptical about its value outside the tracking world that they ought to want to know what the subject was controlling for? Why would they need to know anything more than "the subject is trying to do what is asked"? Yes, we might see that statement as synonymous with "the subject is controlling for a perception that the experimenter perceives that I am following instructions", but how do you persuade someone that some such circumlocution is preferable to "trying to do what I asked him to do"?
Also, I come again to the proposition that in most experiments, the experiment is set up to perform "The Test" for the controlled variable whether the experimenter realizes it or not, in that the experimental presentation offers a disturbance to the presumed controlled variable, and the experimenter sets it up so that the subject can perceive that variable (unless the experiment is to determine how well the subject can perceive it), and determines the degree to which the subject controls for maintaining that variable near its reference level in the presence of disturbances. The experimenter may not know that "The Test" exists, but in most cases it's pretty clear that if The Test would fail, the experimenter wouldn't be very happy with the experimental results.
Again, it's not clear how this helps to persuade anyone of the analogy value of tracking studies.
Perhaps not worthless but certainly misleading. If the experimenter
doesn't know what the subject is actually controlling for in an
experiment the interpretation of what the subject is doing can be
quite far off base. See, for example, consider my demo of the
"Economic Control" at
http://www.mindreadings.com/ControlDemo/Economics.html. If the
experimenter didn't know that the subject was _controlling_ for
caloric intake _and_ for staying within budget then he/she would
conclude that, in the "Poor Man" situation, increases in the cost of
bread cause an increase in demand for bread.
Let's consider what a non-control (or at least a not overtly control) analysis might say (remember, I'm trying to be a Devil's Advocate). If you have a limited budget, and you must eat, you will eat as much of what you prefer as you can afford (unless its unhealthy to eat so much "meat" and you don't want to get sick), and fill up on the less preferred food. If your less preferred food gets more expensive, you won't be able to afford as much of it, so you have to buy more of the other if you are to stave off starvation.
I recognize that if you already think in terms of controlled variables, it's the same analysis, but it doesn't sound as though it is. It sounds like a classical linear programming optimization problem, and the point of the demo is to show that you must use a control analysis to account for the effect. (Incidentally the demo on your web site really does make it look like a linear programming problem -- at least it does to me).
If the only experiments you are interested in involve tracking, I can see
why you dismiss most behavioral science as uninteresting. Of course, you may
be right. Maybe everything we need to know can be found in tracking
experiments. That certainly makes one's reading list much more manageable.
Tracking tasks are simply a nice way to see all the connections in
closed loop behavior.
Yes, they are.
I see the behavior in all experiments as being
closed loop, the same as that in the tracking task.
Yes, you have PCT thoroughly embedded in your thinking, and you do see that. In fact, understanding PCT, you see all behaviour, not just behaviour in experiments, as being closed loop, whether the behaving organism is a human or a bacterium. But the people you are trying to convince don't think that way, so they don't (yet) "see the behavior in all experiments as being closed loop, the same as that in the tracking task." The problem at hand is how you get them to see it that way.
But unlike in the
tracking task it is harder to see the components of the closed loop in
a conventional experiment, particularly the controlled variable.
And sometimes the problem shows its other face, where "it is harder to see the components of the closed loop in [any] experiment, [because of concentrating only on] the controlled variable." That's a problem that may bedevil someone who sees the controlled variable as the blindingly important component of the control loop, hiding all the other components in the glare of its importance.
For
example, I see the classical conditioning experiment as closed loop.
The controlled variable (CV) is the viscosity of the bolus in the
mouth; this variable is equivalent to the cursor in the compensatory
tracking task. The disturbance is the food powder (US) placed in the
mouth; this is equivalent to the computer generated disturbance to the
cursor in the tracking task. The output is salivation (UR) which is
equivalent to mouse movement in the tracking task.
I guess you could test this hypothesis by disturbing the viscosity of the bolus in some way unrelated to the US. If all you want to do in an experiment is to discover the controlled variable, you would have to do that. But it's not usually what interests someone in a conditioning experiment. They might, perhaps, be interested in how variations in the timing relations between the US and the CS affect the salivation when the CS is later presented alone. Whether the controlled variable is viscosity of the bolus would not be of any interest, whether the experiment was conducted within a PCT or an S-R framework. In the PCT framework, it would suffice to say that the controlled variable is something disturbed by the US. Later, when looking for a physiological mechanism for the conditioning effect, it might become important to identify some controlled variable that is disturbed by both the US and the CS, since if the CS actually disturbed the primary variable disturbed by the US (call it "viscosity of the bolus"), the CS would itself be a US, invalidating the conditioning study no matter what paradigm it was run under.
The conditioning part of classical conditioning is probably a result
of reorganization. I see from the Manchester Conference Program that
Bill Powers is scheduled to give a talk on reorganization and
classical conditioning so maybe he will have a model of the process by
that time. I'm working on such a model myself so it will be
interesting to see what we come up with independently.
All of what you say makes sense within the PCT world. None of it, to my mind, comes anywhere near close to suggesting why a person unfamiliar with PCT should be expected to see a tracking study as an analogue of going to vote for a particular candidate or of any classical psychological experiment such as the Milgram study in which people were induced to torture strangers (or at least to think that's what they were doing). And that's the problem at issue that you asked me to address in this response. There has to be some kind of smooth progression of examples that starts with straightforward pursuit tracking and winds up with, say, casting a vote at an election.
Don't ask me how that progression should go, but I think somehow in a series of examples the controlled variable has to shift from the very concrete pursuit tracking of the location of a target matched by the location of a cursor, through compensatory tracking, and then to less concrete targets with less obvious environmental feedback paths. Somewhere along the line, the rubber band, and Bill P's "square circle" demos probably get into the act, but really I have no clue as to how to make the analogy clear to anyone who hasn't been exposed previously to PCT.
Sorry if this sounds negative, but you did ask me to comment in the context of how to get across the correspondence between tracking studies and everyday behaviour or behaviour in "classical" psychological experiments.
Martin