[From Bill Powers (2010.04.30.0835 MDT)]
Bruce Gregory (2010.04.29.1952 EDT) --
BG: Acknowledging my lack of understanding of PCT, I know of no mechanism by which a reference can be "retrieved from memory."
BP: See B:CP, Chapter 15, Figure 15.2 and 15.3. A higher system sends a reference signal to the control system portrayed, but according to this proposed modification, the reference signal coming from above is actually an address signal which selects and replays a recording of a previous perception at this level, which assures that the actual reference signal will be of a kind that this system can perceive. I had worried, perhaps needlessly, about how a higher system could know what lower-order perception it would require at the input to the higher input function, since it perceives only its own perceptual signal which is a function of many lower perceptual signals, not just a function of the one shown here.
When both parts of the switch in fig. 15.3 are thrown, the perceptual signal going upward comes from the lower system's memory instead of the lower system's perceptual input function. The address signal from above would thus be selecting the memory to be replayed. This would be the imagination mode. If the higher system were controlling as I assume it would be, it could vary the input signal it was receiving by varying the address, and thus affect its own perceptual signal just as if the lower system had been used and had succeeded in producing the required amount of perceptual signal.
Notice that as far as the upper system, above the system in fig. 15-3, is concerned, it controls its own perception just as it usually does. It doesn't know that part of its own perception is coming from a recording of a past value of this lower system's perceptual signal, instead of from lower systems or the environment. Many people have remarked that perceptions include a lot of imagined information; I was trying to guess how that might be done.
I'm sure you realize as I do that this proposal leaves a lot of questions unanswered (but perhaps a few less than existed before). This is the best I could do to incorporate a lot of memory phenomena into the model. I think it does pretty well at that, but it creates other questions that remain unanswered. If you read Chapter 15 again, all the way to the end, I think you'll find all the proper disclaimers, which are real, not conventional boilerplate.