Concepts & models; Busses

CSGnet Archive CSG1992
1

[From Bill Powers (920904.1130)]

Martin Taylor (920903.1900)]

RE: concepts and models

I don't have or expect to have a specific model of alerting systems.

I >can think of many. But the situation is like that of PCT itself.
The >basic proposition of PCT, that "all (purposeful) behaviour is the

control of perception" is incontrovertible. There are many models

that >might instantiate it. So with the alerting systems. Given the
basic >proposition of PCT, their existence in an organism supplied
with more >sensory than motor degrees of freedom is incontrovertible.
There are >many models for how they might work.

It's not self-evident that left-over perceptions or control systems
must have the function of "alerting." I don't dispute that there will
be perceptions and systems not in use, because of the df problem.

The "concept" of control that I was talking about was the concept that
behavior controls perception. This concept is seen as fitting
experience: that is, any behavior you happen to notice turns out to be
a perception, on closer consideration. Cut off the relevant perception
and you can't do the behavior, particularly in the presence of
disturbances.

The model is the CT model. It shows the kind of organization that's
required to create the effect observed. With due regard to alternative
but equivalent forms of the same model, it's the only known kind of
system that can produce this sort of phenomenon. There's no
alternative that I know of.

In the CT diagram, every block and every arrow is assigned a specific
function. By instantiating these functions it's possible to construct
a working model of a control system. Even in cases where we can't yet
instantiate the functions, we can see that if we could, or if we could
fake it, we'd have a runnable model.

This is what I'm after regarding your "alerting systems." You show
them as control systems not connected to lower systems, but sending
arrows over to the active control systems. What is the meaning of
these connections? What KIND of effect are they supposed to be having
on the target system, and what is there about their origins and
destinations that makes this effect plausible? Are you including
enough machinery in this model, even in block form, to enable it to do
the things you say it should do?

···

----------------------------------------------------------------------
--

As for "controlling for variability," I was more musing than

proposing, >but if you want a proposal, you mentioned a while back
that if you had >two one-way control systems back-to-back to make a
two-way one, and if >they had the right (square-law?) control
function, then the pair could >be controlled both for gain and for
reference level.

I can see that the variability of the perceptual signal relative to
the reference signal would be changed by changing the loop gain,
loosening or tightening control. But that is AFFECTING variability,
not CONTROLLING FOR variability. In order to control for variability,
the system that is varying the common-mode reference signal must
perceive something about the lower system that constitutes its
variability. What would that be?
----------------------------------------------------------------------
-
Avery Andrews (920904.0950) --

Re busses: fine. But as far as I can see, Chapman's architecture is
consistent with this - as long as you have a finite system of
categories sent along the bus, you can have a single wire coming off
for each one.

But then for each category wire, you need a category-perceiver at the
lower level. This just moves the problem down a level. Each category-
perceiver has to receive all the elements that could be categorized,
and respond to the one category it is designed to detect. There could
still be an infinity of categories, with only those actually perceived
having any effect in the system.

PS did Harry Erwin really send something to CSGNet? I posted

something >of his yesterday, but I also lost a whole swag of unread
messages 2 >days ago, so might well have missed a posting by him.

I thought the post was from him. Sorry.
----------------------------------------------------------------------
-
Best,

Bill P.

2

[Martin Taylor 920904 15:40]
(Bill Powers 920904.1130)

Very quick, or it will be 3 weeks till I answer...

It's not self-evident that left-over perceptions or control systems
must have the function of "alerting." I don't dispute that there will
be perceptions and systems not in use, because of the df problem.

Self-evident is a funny term. It's self-evident to me, in that I can find
very little that is more "atomic" than the simple statement of the df
situation, from which the existence of alerting systems follows directly,
given the basic axiom of PCT (not Greg's 6 axioms, but the axiom that
behaviour is the control operception).

Perhaps the problem is that we have a different sense of "alerting." The way
you use it, it sounds as if you make some connection with consciousness.
All I want to say is that the presently uncontrolled percepts are monitored,
and the consequences of this monitoring are that when something relevant
happens, control moves to percepts that are marked as possibly having become
important to control. That means either that there were already "spare"
degrees of freedom that were not being used actively for control (I suspect
this is the normal case) or that some degrees of freedom that were being
controlled are dropped in favour of the ones that were "alerted." I take
"alerting" to describe the marking of df as being probably useful to control,
and don't put any more on it than that.

In the diagram to which you refer. the arrows were meant to indicate this
kind of non-specific pointer function, not to specify a mechanism. On other
occasions I have proposed gain functions with dead zones as being a way that
ECSs can perform the alerting function on themselves. In that case, there
is no overt mechanism, but the df shift occurs as a consequence of conflict
between the "alerted" ECS and previously active ones with correlated perceptual
functions. There are probably dozens of other possibilities that come within
my concept of the necessity of "alerting."

As a related topic, it seems to me probable that there are usually far fewer
output degreed of freedom used in active control than are available, because
the use of all available ones seems rather like a very high workload condition.
If this is so, then the activation of an output degree of freedom as a
consequence of an alerting situation need not take a degree of freedom
away from what was already being controlled. It also implies that output
degrees of freedom may, in a way, "drift" in and out of use by the actively
controlling ECSs. There's a difference between a cat standing guard by a
mousehole and the same cat in a similar attitude, resting. The guarding cat
is actively controlling a lot, without moving, whereas the resting cat is not.

···

================

In the CT diagram, every block and every arrow is assigned a specific
function. By instantiating these functions it's possible to construct
a working model of a control system. Even in cases where we can't yet
instantiate the functions, we can see that if we could, or if we could
fake it, we'd have a runnable model.

This is the construct I called "symbolic" PCT, and that I see as the reason
you had problems with your reorganization simulation in flipping the signs
of output-to-reference links in an active controller. Symbolic PCT is
surely easier to program than is distributed PCT, but I doubt it is as robust
or as true to life.

I can see that the variability of the perceptual signal relative to
the reference signal would be changed by changing the loop gain,
loosening or tightening control. But that is AFFECTING variability,
not CONTROLLING FOR variability. In order to control for variability,
the system that is varying the common-mode reference signal must
perceive something about the lower system that constitutes its
variability. What would that be?

Again musing rather than proposing...Imagine a perceptual function consiting
of a leaky integration of a rectified time-difference signal. If I remember
my rusty electronics--

--->|--C1--------------
           > >
           R C2 where C2 >> C1 (I think I put the rectifier in the
           > > wrong place)
-----------------------

One can imagine other perceptual input functions that might do the job. They
don't have to be complicated, and they can combine the inputs from many
lower-level ECSs, either before this circuit or after it. depending on whether
they perceive something like the variation of the average or the average
variation of a lower-level ECS.

Martin