[From Bruce Abbott (2017.10.20.0820 EDT)]
Ordinarily it is correct to say that a control system controls both the perceptual signal and the environmental variable (or combination of such) of which the perceptual signal is a function. This is true so long as the systemâs input function remains stable. There are situations, however, in which that function drifts or changes over time, and in those cases, it cannot be said that the system is successfully controlling both the perception and its environmental analog. Control of perception is primary and control of the environmental correlate of that perception is secondary. When the input function is stable, both are controlled, but this is not always the case.
Consider, for example, the process of sensory adaptation to darkness. The photoreceptors in the retinas of our eyes gradually increase their sensitivity to light in a darkened room. Mathematically this amounts to a changing input function where the intensity of light falling on the retina gets converted to a neural signal. There is no longer a constant relation between the lightâs physical intensity (the environmental variable) and its perceived intensity (the magnitude of the neural signal). A person controlling for constant light intensity would have to be reducing the physical intensity of the light during the period of darkness adaptation in order to experience the perception of constant light intensity. She would be successfully controlling the lightâs perceived intensity, keeping it at a constant reference level, while failing to do so with respect to the intensity of the light that is entering the eye.
An engineered system for controlling the temperature of a water bath might experience a similar problem if the heat sensorâs electrical connection became corroded. Imagine that this condition causes the sensed temperature to fluctuate rather wildly. The control system would respond by increasing or decreasing the output of the heater in an attempt to keep the sensed temperature constant (control of perception) even though the water bath was already at the desired temperature. The controller might be controlling its perceptual signal successfully, while failing to properly stabilize the temperature of the bath. Again, control of perception is primary, successful control of the environmental correlate of that perception depends on the function relating the perception to the environmental correlate remaining stable.
Bruce
Bruce Abbott (2017.10.18.2100 EDT) –
BA: Given what I wrote in Bruce Abbott (2017.10.18.1745 EDT) in response to Fred Nickolsâ query, it may come as a surprise that I agree with Martinâs analysis below. The position I described is based on what I would call âinformed realismâ? to distinguish it from ânaïve realism.â? The latter is the belief that what you perceive is exactly what is actually out there beyond the senses. I recognize that our perceptions are very selective and incomplete representations of what may actually be out there. The evolutionary process has given us perceptual systems that generally do a good enough job of representing those aspects of reality that usually are relevant to maintaining our well-being. When I see a baseball coming straight for my face, experience tells me that Iâd better duck or else prepare for a painful experience. My perception of that baseball is not a perfectly accurate representation (I donât perceive the infrared radiation it emits, for example), but itâs usually a good enough representation to avoid being struck by whatever the reality is beyond that perception of it.
BA: Although recognizing that all we actually have are our perceptions, I prefer to take the point of view of the engineer, who works with what she takes to be real objects with measurable properties and with real sources of energy whose properties also are measurable. When designing a carâs cruise control, she assumes a real car with a real speed along a real road and possessing a real set of mechanisms operating according to well-defined rules (laws of physics, etc.). From that point of view one can design a material sensor to pick up the carâs speed (many ways to do that), convert it into a perceptual signal within the carâs cruise control system, and so on, so that the system functions to keep the carâs actual speed over the road near the cruise controlâs set point.
BA: In other words, I am taking the designer or analystâs point of view when describing a control system and how it works From that point of view, I assume a real controlled variable in the environment, one whose values are being sensed by a real mechanism that transduces those values into perceptual signals, etc. I provide the same kind of description whether the system being analyzed is a carâs cruise control or a personâs control of the loudness of the music being listened to on an iphone. For my purposes I find this approach clearer than the alternative that places the material world beyond perception off limits.
Bruce
[Martin Taylor 2017.10.18.16.18] –
From Bruce Abbott (2017.10.18.1505 EDT)]
From: Boris Hartman [mailto:boris.hartman@masicom.net]
Sent: Wednesday, October 18, 2017 12:51 AM
To: csgnet@lists.illinois.edu
Subject: RE: Behavioural Illusion
Bruce
Bruce Abbott (2017.10.15.0820 EDT)]
From: Boris Hartman [mailto:boris.hartman@masicom.net]
Sent: Sunday, October 15, 2017 4:34 AM
To: csgnet@lists.illinois.edu
Subject: RE: Behavioural Illusion
Rick Marken (2017.10.11.1745)]
Bruce Abbott (2017.10.10.1015 EDT)–
BA: I appreciate the point. According to PCT, behavior is driven by error in a controlled perception, so the experiment must have created a disturbance to some controlled variable. But does the CV that is disturbed by the manipulation necessarily the same variable that is under investigation? The answer is âno.â?
HB : I hope Bruce that you are using »some CV« in the sense of perceptual variable. That’s the only controlled variable in the loop.
[MT] As long-term readers of CSGnet will know, I held a long-standing opinion that what Boris says here was true, and have argued with Rick about it. But for several reasons over the last year or two I have modified that opinion in favour of a more nuanced view. I don’t know if Rick will agree with my view, but at least I am less likely to complain if he says that an environmental variable is controlled. Let me review, in no particular order, some of the reasons.
[MT] (1) Collective control. When many people control related perceptions of related CEVs (“Complex Environmental Variable”, Bruce), they may all be controlling their perceptions, but the environmental that is most closely stabilized may not be the CEV that is perceived by any of them. In Kent’s original demonstration of collective control at CSG-93, two controllers controlled perceptions of the same CEV and wound up in conflict but to an outside observer of the CEV they were apparently controlling that CEV to a reference value actually held by neither.
[MT] When there are more than two, the more the merrier, the actual CEVs might be quite different in each individual, but it nevertheless looks as though a CEV perceived by none of them is strongly (high gain) controlled to a reference value held by none of them. The same may be true within a brain. Bill approximated the firings of many neurons, each of which would have its own individual collection of synaptic collections, and called the sum of the firings averaged over some time interval a “neural current”. The value of that neural current, in one particular part of a control loop, is a “perceptual variable” or a “perception” in PCT. But nowhere in the brain is that perception actually represented. It is distributed over tens, hundreds, or thousands of neurons, and is stochastically represented over time.
[MT] (1a) Related to (1) is the paper and demo by Bill for which Rick [From Rick Marken (2017.10.17.0840)] provided links. It demonstrates that to control a single environmental variables does not require a dedicated special perception of that environmental variable. The CEV represented is distributed over the brain even more widely than is suggested in (1).
https://www.dropbox.com/s/2u00ac87bix2sjv/MultiControlPrj.exe?dl=0
https://www.dropbox.com/s/rwoqfa8v96g62ob/multiple_control.pdf?dl=0
[MT] (2) Evolutionary necessity. A rock that smashes your skull will kill you. A perception of such a rock will not. A shield raised to ward off the rock will save you, but if you control the position of the shied to where you perceive it will intercept the rock and you are wrong, you are just as dead, even though you controlled your perception of the position of the shield extremely well. What matters is what happens in the environment. Controlling a perception is useful only if the environmental property to which it corresponds actually exists and is as stably related to a perceptual reference value as is the perception itself.
[MT] (3) Basic mathematics of the control loop. An environmental property is what it is, whether it is perceived or not. When it is perceived, that perception is delayed, if only slightly, and there is always a resolution limit to the precision with which it is perceived. That’s why we have microscopes and telescopes. The perception is what it is, but every value of the environmental property corresponds to a range of perceptual values that it might induce, and vice-versa. Nevertheless, on average any specific value of the environmental property is most likely to correspond with a specific value of the perception, and vice-versa, provided that the time-delay of perception is not long compared with the rater of change of the environmental property. Therefore, when the perception is controlled in a canonical control loop, so is the CEV that corresponds to that perception.
[MT] (4) “Real reality”. We can never know what is truly “out there”, whatever we perceive. But what is really “out there” that determines our life course. That we can control many perceptions means that we can have an illusion that we are influencing real reality – think of Bishop Berkeley kicking a rock to prove its existence (or was it the friend who did the kicking?). In that sense, it is only the control of perceptions that matters, if anything does. The correspondence between the CEV and real reality might be entirely illusory, but so long as it is perceived as acting as it would if it were real, the Evolutionary Necessity argument overrides the “it’s all perception” argument.
[MT] (5) On the Other Hand: When we control a perception, that perception, not the CEV, is what we are acting to vary so as to bring it near a reference value and keep it there. We do so by apparently acting on something in the environment that changes what our senses tell us No matter that if we waste resources by controlling perceptions of things that aren’t there we may put our survival in jeopardy, we still can control only our perception, however distributed across the brain its physical/physiological manifestation may be.
[MT] (6) Summing up. Perception is all we have that we can control. The environment is that we really need to control. We can control the environment only to the extend that it behaves as though our perceptions correspond to it reasonably well. So for most practical purposes other than careful theory or metaphysical philosophy, it doesn’t matter a whit whether we say that perception or the CEV is controlled.
[MT] On nomenclature: I think it unfortunate that Bill used “CV” (“controlled variable”) even casually to refer to an environmental variable. The theory is called Perceptual Control Theory for a good reason. In discussions with Kent and Eetu a similar complaint was raised about “CEV” (“Complex Environmental Variable”). Any variable in the environment could be one of those, so we agreed among ourselves that “CEV” (Corresponding Environmental Variable) would be better. Corresponding to what? To a perception, controlled or not. Above, I mentioned the old expansion of the acronym, because it might have been familiar to many long-time readers. But in future I will try to use “Corresponding Environmental Variable” as something in the environment that is perceived.
Martin