[From Bill Powers (980428.0134 MDT)]
Bruce Nevin (980426.1630 EDT)--
This has been a puzzlement for me for some time. Reference signals come
from memory. Yet they come from error output of higher levels. How is this
resolved? Just a pointer to the relevant part of B:CP would help.
The answer I proposed in B:CP (Fig. 15.2, p. 218) is that the output signal
from the higher system's output function, rather than directly entering the
lower system's comparator, enters the local memory where it addresses a
recorded value of a past perceptual signal. The replayed recording then
becomes the lower system's reference signal. This is in the chapter on
memory. This arrangement might not be present at all levels. I don't think,
for example, that it exists in the spinal-reflex systems.
A plausible answer is that an error-driven reference value from on high
overrides a signal recalled from memory; and, conversely, in the absence of
a reference signal from above the reference value is recalled from memory.
The term "overrides" is hard to tie to any mechanism. How does one signal
override another? The intuitive picture is that somehow the overriding
signal has the ability to replace the other signal, so the other signal's
normal effects don't occur. But if you try to model that sort of effect,
you find that all sorts of _functions_ are necessary. The overriding signal
has to enter a function that interrupts the other signal's effects, while
turning on a gate that lets the first signal appear at the place where the
other signal used to appear. While such circuitry is possible, when you
recognize what's required the "simple" idea of an overriding signal no
longer appears so attractively simple.
Part of the trick of modeling -- an important part -- is to recognize when
your proposed solution contains a hidden mechanism. Above, you say "in the
absence of a reference signal from above the reference value is recalled
from memory." But HOW IS THAT DONE? Some mechanism must be sitting there in
the background, monitoring the reference signal. When there is a reference
signal of any nonzero magnitude, it is routed to the lower comparator, but
when the reference signal is missing, the output of memory is routed to the
comparator instead. And _which_ memory is used? What determines that? Our
background mechanism now has to have still more complex properties, the
ability to select one recording rather than another residing in memory. And
what do we mean by "recall from memory?" How is a particular memory
recalled? There must be some addressing mechanism, possibly content
addressing as I proposed in B:CP. Does this mechanism have the ability to
know what the content of memory _means_?
As you see, it's easy to propose an idea in words that, on closer
inspection, expands into great complexities in terms of actual mechanisms.
This doesn't mean we have to back off from complexity. It means that we
should not judge proposals just on the basis of how simple they sound when
expressed in ordinary language. They may actually be more complex than the
complex idea they're meant to replace, when turned into some sort of
realizable model.
In the imagination loop, a copy of the reference is taken as input and
controlled perfectly.
Yes. This also happens under my proposal.
Perceptual inputs are passed up to higher-level perceptual input functions
(PIFs).
If a copy of the remembered reference value is passed up to a higher-level
PIF, and at the same time all other inputs to that PIF also originate from
lower-level memory rather than from controlled perceptual input, then the
higher-level perceptual signal (from that PIF) is also being recalled from
memory. Or rather, it is being constructed from signals recalled from
memory.
Look at Figs. 15.1, 15.2, and 15.3 which form a progression showing how the
idea of the imagination connection developed. Imagine 15.3 with the
switches in each of the four possible configurations. The discussion that
starts at the top of p. 220 should help. There are some missing details, of
course, like what throws the switches and how the memory addressing works.
If this is the case, it follows that memory at the lowest levels is the
basis for constructing memory (and imagination) at higher levels. This
gives specific, possibly testable, meaning to the suggestion I have read
that memory is stored throughout the body and recalled through body
sensations. It explains some characteristic experiences that meditators
report, some otherwise puzzling statements in buddhist psychology, and some
reports as well of body-based therapists (e.g. Gendlen's notion of
grounding). It may even be that memory is "housed" at the level of
intensity perceptions, and that memories of sensations, configurations, and
transitions are constructed.
What is a "body sensation" that is different from just a sensation? I think
what you're overlooking is that when you attend to various parts of your
body, you're attending to _perceptual signals in your brain_. The actual
physical place to which you're attending is not your physical nostrils (in
the example you've used), but the place inside your brain where the signals
are that represent your nostrils. It's all perception.
I think that memories exist at every level as recordings of perceptual
signals at that level. Since perceptual signals at one level are functions
of perceptual signals at lower levels, it follows that memories of one
level are functions of memories at lower levels, but not directly.
Look at Fig. 15.3 and imagine a duplicate of it at the next higher level.
Note that the higher level input function receives a perceptual signal
which can equally well be a remembered perceptual signal or a present-time
perceptual signal from still lower systems. The higher system has no way to
know whether its inputs are coming from memory or from the environment. Its
memories might all come from real-time perceptual signals, or they can
consist mostly of lower-level imagination, of a mix. Remember, too, that
there would be many systems at the lower level contributing to the
perception at the next level up.
It's not quite so obvious that a memory signal and a perceptual signal take
their meanings from input functions. It's the input function that
determines what aspect of a set of input signals is being represented by
the perceptual signal. The perceptual signal itself is one-dimensional, and
so, therefore, must the memory signal be one-dimensional. In order to get
what we normally call "a memory," many systems must be in the imagination
mode, so a complex external situation can be imagined in the same way it is
perceived: as a collection of many signals, not just one signal. The
question of what flips the switches is of more than passing interest; the
combinations in which they are flipped will determine what we remember and
how we fill in information missing from normal perceptions. I think that
the switch-flipping may have a lot to do with "attention."
Remembering at a high level seems like remembering _that_ something
occurred, without actually remembering the details. Remembering at the
lowest possible level is equivalent to hallucinating.
This would account for the frequently diluted character of memory as
compared with the immediacy and fullness of present experience: only some
inputs to a higher-level PIF are recalled, so input to PIFs is sparse.
This needs some fleshing out (you will pardon the pun), but I think there's
something here.
I highly recommend that you re-read Chapter 15. Most of what you're talking
about is there, including the idea of associative or content-addressed memory.
Best,
Bill P.