[From Bill Powers (2009.07.29.0851 MDT)]
Martin Taylor 2009.07.28.17.43 --
The inputs to a physical control system enter its sensors or input functions, and the outputs come from its effectors or output functions. Why make it more complicated than that?
What is "a physical control system" to a naive user? Consider a thermostat. To an everyday user, the input to the thermostat is the dial setting (to a PCT scientist, the reference value), while the output is the desired temperature (the controlled variable). Very simple.
Yes, for uninformed users. Typically of most oversimplifications, it misinforms more than it informs. It is misleading enough to have given a lot of people (like Bertalanffy, for example) the impression that a control system is just a stimulus-response system with a feedback loop added, because it encourages confusing the reference input with a sensory input. I tried to lay out the model and define its parts to eliminate the confusions that I saw. You may have noticed that the diagrams like those Wiener used don't even show a sensor. Most psychologists who have borrowed that diagram think of the "input" label as meaning "sensory input." The actual sensory input isn't even labeled in that diagram.
The user does not need to see anything about connections to the air conditioner or furnace, need know nothing about the (PCT) effectors, and just knows that if he inputs 21 degrees C, the room goes to 21 degrees C. Simple input-output.
I have pointed out that very interpretation as the cause of many misunderstandings about control systems. It gets pretty hard to systain that rather careless interpretation when speaking about control systems like the iris reflex, that act to control the illumination of the retina. It's rather strained to insist that the light falling on the retina is an output of the control system. It's a little hard to keep a straight face while explaining that the output of the body's thermoregulation system is the sensation of skin temperature or internal body temperature.
When one is trying to introduce PCT to a newbie, it can help if the terms can be mapped onto something the newbie already knows.
What that usually accomplishes is to mislead the newbie into thinking he already understands negative feedback control. My career has been littered with such people. I much prefer to find out what the newbie already understands, and contrast it with the PCT understanding so the difference is made perfectly clear. Then it's up to the newbie to grasp the difference if he or she wants to. If the newbie doesn't want to grasp the difference, the newbie can go study something else.
If control engineers use the common language, in which the "output" is the controlled variable, then they can perhaps come to terms with PCT more easily if they are told not that they are wrong, but that in PCT discussions, the control loop is split into halves, in which the "input" and the "output" are defined not by the bounds of the control loop but by the seperatrix of the halves of the loop. Yes, it's a simple idea, but not simple to one used to thinking of inputs and outputs of a loop as being from and to the environment external to the loop.
That sounds plausible but I think the confusions are much worse than that. At a Gordon Research conference on cybernetics some years back, I showed the Little Man and explained my way of modeling control systems. Afterward, a man from the University of Texas came up to me and said he had been teaching control theory to undergraduates for some years, and for the first time he thought he could actually get the way control systems work across to them. He had been very disappointed with the results and now knew why.
Unfortunately, you would have to explain to the engineers that the "input" they talk about is not a sensory input, and that the apparent link to stimulus-response psychology is extremely misleading because of that. You would also have to point out that in their diagrams they commonly leave out the actual sensory input, which increases the confusion, and they also leave out the environmental functions between the actual outputs of the controller and the sensory input. The actual outputs of the controller are also omitted. It's a sloppy mess because the engineers think they're communicating with idiots and leave out all the details that might make their explanations correct.
The output comes out of the control system; the control system puts it out into the environment. The input goes into the control system; the environment puts it into the control system. ...
All of this is true, but remember that the "environment" of a higher-level control unit consists of all the control units to which it sends reference values, and its input consists of all the control units whose perceptual signals it senses. That's useful to a PCT analyst, but it does not provide for any observable inputs and outputs.
Of course it does. You observe them in the same way you observe your own higher-level inputs and outputs. You observe the degree to which someone is acting honestly, and you observe the effects of his actions on your perception of honesty. You can do this because you're observing from inside your own higher levels, and can experience the world from various points of view.
The "environment" for the observer consists of the multiplicity of outputs from the lowest level actuators, and the input of the multiplicity of inputs to the peripheral sensors. These are not the inputs and outputs in the PCT sense to and from the higher-level control system.
No; the inputs are the environmental inputs to the higher-level system's input function; out of that input function comes the perceptual signal that is the system's perception of a controlled variable.
The problem is that you're jumping around among points of view. It's by far the simplest to say that a control system lives in an environment consisting of lower-level systems. It receives inputs from those systems (copies of their perceptual signals) and it generates outputs that go into those systems (the reference signals for the lower systems). This view can be maintained consistently for any level of system, with the slight modification that the lowest-level control system gets its input from sensors and sends its output signals into effectors instead of comparators. This leaves the environment outside the hierarchy where it belongs, and lets us deal with it explicitly as a function that makes the controller's inputs partly dependent on the controller's outputs and by no means identical to them (as Wiener's "feedback take-off" implies).
However, if you naively think of the controlled variable as the output of a control system, then, to take an example, the action output is "knocking on the door" and the controlled variable (output) is "the door opening".
I'm not interested in thinking of the controlled variable naively; I want to think about it correctly and consistently. If you said the above, I would ask you how you know the door is opening. You would have to say you see, feel, or hear it opening. I would ask if seeing, feeling and hearing should be called outputs or inputs. I hope you would be embarrassed to find yourself saying "outputs." If you're not embarrassed I can't explain control theory to you -- you wouldn't understand it.
Which way of looking at the world of control is simpler depends on where you are coming from, and what you want to analyze.
Absolutely. That's why we have to find a way of looking that is consistent with itself and everything else we observe, and make sure that our method of analysis doesn't lead to contradictions.
All I'm saying is that nomenclature can be a barrier to understanding, if the same words are used to mean something one does not expect them to mean. And most people think of the output of control as being the controlled variable.
Which is a barrier to understanding. They should change their understanding because it is wrong and impossible to use consistently. Wasn't it Will Rogers who said that what's wrong isn't because of what we know, but what we know that ain't so? Well, controlled variables are not outputs from the control system, and I can prove it by measuring the actual effector outputs and showing that they do not always, or even very often, covary with the state of the controlled variable. If you want to object to that, you have the onus of explaining how something going into the control system can be an output, while what is actually coming out of the control system doesn't even correlate with it.
Or are you going to start maintaining that what the effectors do to the environment is the same thing sensed by the control system? We already know that that ain't so.
Einstein said we should make our explanations of nature as simple as possible, but no simpler than that.
Best,
Bill P.
