[From Bill Powers (2005.08.20.1002 MDT)]
Rick Marken (2005.08.20.0840)–
On page 217 of the first edition
of B:CP, in the paragraph beginning �We do not need,� (and similarly on
page 219 of the second edition), we find: �all behavior consists of
reproducing past perceptions� and �all reference signals are retrieved
recordings of past perceptual signals.�Yes. In this discussion in B:CP I believe that the “past
perceptions” function as reference signals. What Bill is saying, I
believe, is that all reference signals are memories of past perceptions.
I think this is a little extreme, though. I bet we come into the world
with some reference signals built in – corresponding to perceptions we
have never experienced and, thus, cannot have stored as
memories.
There are two different subjects being mixed together in this discussion.
One is the source of reference signals; the other is the source of
perceptual signals. In Chapter 15 of BCP, I was constructing an
hypothesis, not saying how things really are. The point of this
hypothesis was to account for two indisputable phenomena: remembering
past experiences, and acting to make present perceptions match past
perceptions. Both are taken care of by the proposal expressed as
“All behavior consists of reproducing past
perceptions.”
That postulate is more general than it needs to be, and it’s not specific
enough because it doesn’t take levels of perception into account. I have
commented several times to the effect that this postulate probably
shouldn’t be applied at the lower orders of control, where lower-order
reference signals probably come directly from higher-order output
functions.
As to perceptions being functions of lower-order inputs and memories at
the same time, I think I ruled that out by postulating those switches,
which make sure that a perceptual input to a given level comes
either from lower-order perceptual signals, or from
short-circuited lower-order reference signals, but not both at once. If
you allow a given perception to consist of partly real and partly
imaginary components, then the controlled perception is not being
affected by behavior in the right way. It would be affected both by real
actions in the external world, plus disturbances, and by simply changing
an output signal from the higher system. I don’t think that would produce
control of either the real or the imagined perception.
Please examine Fig. 15.3. You will see that the inputs to the input
function simply come from lower-order systems – never from the
imagination connection at the level shown. The downcoming address signal
from above enters memory, which then produces a reference signal made of
past perceptions recorded from the input function of the depicted system
(Note that you can remember remembering a perception). That reference
signal then either enters the local comparator where it produces real
behavior, or it is short-circuited and sent back up to the higher system
in the same channel where normal perceptual signals would go. In either
case, the upgoing perceptual signal matches the reference signal set by
the higher system. As far as higher systems are concerned, the
upgoing signal is a normal input signal generated as a result of sending
a reference signal to a lower-order system. And note also that neither a
real nor a remembered perceptual signal from this level enters the input
function of the control system at this level.
What this means is that for each input to the higher-order input
function, the signal is coming either from actual perceptions generated
by a lower-order system’s input function, or from that lower system’s
reference signal – but never from both. A given perceptual signal is
either real or imagined, but is never a mixture of both, like 60% real
and 40% imagined. It’s either all real, or all imagined. And the
imagination connection at the level shown in the diagram contributes
imagined information to the higher system, not to the system shown.
This means that a higher-level perception may be a function of some real
and some imagined perceptions, but that higher system can’t tell the
difference. If perception p is a function of lower-level perceptions p1,
p2, and p3, that continues to be the case whether p1 or p1 or p3 happens
to be real or imaginary. Each lower-level perception received is either
totally imagined, or totally real: it makes no difference to the higher
system, except that if a given input signal is imagined, control is
better than it could be if actions were required. A higher-level
perception becomes more easily controllable if some or all of the inputs
to the input function at that level are imaginary.
One way to look at this is that the imagination connection allows a
control system at a given level to lie to the higher-level system that is
setting its reference signal. The lower-level system is telling the
higher one that the specified input has been produced through behavior
when in fact it has not.
The result is that we DO NOT have a case where the inputs to a given
input function consist of a set of real perceptions, and another set of
imagined perceptions entering in parallel to the first set (as you drew
it, Rick). There is only one set of incoming perceptual signals. Each
signal in that set comes either from a real lower-order perceptual
signal, or from the reference signal being sent to the system that
normally produces that perceptual signal, but never from both.
Model it, Rick (or Bruce A.) You’ll see.
Best,
Bill P.