[From Rick Marken (2015.11.06.1030)]
RM: Now that all of us (well, almost all of us) are on board with the idea that behavior is control, the next step is to see what perception has to do with it.
RM: Once you realize, as Bill did, that what organisms do – their behavior – is control, then you know that the only way to explain their behavior is using control theory; and that the main thing to be explained about behavior as control is the existence of reference states for controlled variables (see p. 176, LCS I, para 2 and 3 in particular). That is, control theory has to explain the facts associated with each Behavior in Table 1, p. 172 of LCS I (and in the “Behavior is Control”. spreadsheet): the fact that what we see as Behavior involves bringing Controlled Variables to and maintaining them in Reference states using Means that precisely counter normal Disturbances to those Variables.
RM: The situation is similar to trying to understand the behavior of a thermostat. Say you have observed (correctly) that the Behavior of the thermostat is control (which it is, of course): there is a Controlled Variable (room temperature) that is being kept in a constant Reference state (say, 68 degrees F) using Means (turning a heater on and off) that precisely counter normal Disturbances (such as variations in outdoor temperature). Control theory tells you that this behavior can be explained by assuming that the thermostat is controlling a perception of room temperature relative to an internal reference for that perception, the difference between reference and perception driving the heater output. And it turns out that this is exactly how the thermostat works; it controls the temperature as perceived in terms of the size of a bimetallic strip (the perceptual signal in the thermostat) which is compared to a reference, that is a contact inside the thermostat, and the difference between reference and perception (the error, measured by whether or not the bimetallic strip touches the contact) turns the heater on or off.
RM:Of course, the big difference between the control organization that explains the controlling done by the thermostat and that which explains the controlling done by living organisms is in who sets the reference for the controlled variable. In the thermostat the reference is set from outside the system by the user of the system; in living organisms the reference is set by the system itself. This obvious difference between artificial and living control systems was not understood when control theory was first applied to the behavior of living systems because behavior was thought of as an output produced in response to input stimulation. Powers was able to apply control theory correctly, by having the reference signal set inside the behaving system itself, precisely because he understood that behavior is control.
RM: So perception is part of the theory (control theory) that accounts for the fact of control. Perception doesn’t play a big role when control theory is used to understand the behavior of artifactual control systems, like the thermostat, because engineers know what variables they want these systems to control so they construct perceptual systems, like a bimetallic strip, that produce perceptions that vary in proportion to variations in those variables. But perception plays a central role when control theory is used to understand the behavior of living control systems because we don’t know what variables the system has been “built” to control and we can only understand what a system is doing (that is, we can only understand its controlling) when we know what variable it is controlling. And we know from control theory that the variables a control system controls are defined by its perceptual functions.
RM: The importance of the fact that it is a perception rather than an objective state of the world that is controlled was brought home to me recently as I was analyzing some data for a study that was done by Warren Mansell and his students. I was asked to build a PCT model of the behavior in a video of the rubber band demo. The data, which were derived from screen captures of the video, were temporal variations in the position of the knot and S’s and E’s ends of the rubber bands. The goal was to build a PCT model that mimicked S’s behavior: controlling the position of the knot, keeping it over the dot while compensating for the disturbances produced by the movements of E’s end of the rubber bands.
RM: I was able to build a simple control model that fit the data quite well (correlation between model and actual movements of S’s end of the rubber band was .98, RMS deviation was 5.1 pixels out of a possible 70). But in order to get this fit I had to set the model’s reference for the distance between knot and dot (the presumed controlled variable) to a value that was much greater than 0. This implied that S was not following instructions, which was to keep the knot over the dot. These instructions suggest that S’s reference for the distance between knot and dot should have been set to 0.
RM: And then it hit me. S was indeed keeping the distance between knot and dot at 0, but S was keeping this distance at 0 from S’s perspective. S was looking at the knot from the side so there was parallax in S’s view of the knot – a displacement of the image of the knot/dot distance relative to what this distance would be when viewed from directly above (as it was in the video). The model was “picking up” this displacement by requiring a non-zero reference specification in order to get a good fit. But it was actually S’s perception of the knot/dot distance that was displaced, not S’s reference.
RM: So the modeling led me to realize that the variable controlled by S was not the “objective” distance between knot and dot, as measured from the video, but, rather, the distance between the dot and a parallax displaced image of the knot as seen from the S’s perspective. When I changed the model so it was controlling this “parallax” perception of the knot/dot relationship, the reference could be set to 0 and the model fit the data exactly as well as it had when it was controlling the “objective” knot/dot distance relative to a “displaced” reference value.
RM: The lesson here is that behavior can be objectively determined to be a control process without saying anything about it theoretically being perception that is controlled. You can objectively determine that S is controlling the knot/dot distance in the rubber band demo, for example, by observing that this distance varies far less than it would be expected to as a result of E’s disturbances (movements of E’s end of the rubber bands) and that this is due to S’s compensating actions (movements of S’s end of the rubber bands). But you can’t tell precisely what variable S is controlling unless you know that S is controlling a perception, rather than the objective state of affairs as seen by you, the observer. This is where PCT comes in. Using the PCT model (as I did in modeling the behavior of S and E in the rubber band demo) you can get beyond knowing that a person is controlling to knowing precisely what they are controlling.
RM: Of course, the perceptual variable a person is controlling is going to be related to the “objective” variable that you can see being controlled – as was the case with the rubber band demo model, where the perceptual variable controlled was just the parallax displaced objective variable – but in order to have a science of purposeful behavior – control – you have to have precise definitions of the variables the system control in order to develop correct models of how the system works. PCT gives you this precision.
Best
Rick
···
–
Richard S. Marken
www.mindreadings.com
Author of Doing Research on Purpose.
Now available from Amazon or Barnes & Noble