Bob Clark's proposals

[From Bill Powers (921209.0900)]

Bob Clark (921205) --

It is my impression that Powers did not, and does not now, consider
that book to be complete and final.

Right you are.

However there seems to have been little discussion of possible
changes in the original Hierarchy. Rather, there seems to have
been discussion of various interesting and important applications
and related ideas.

I'm glad to see you opening up the discussion. "The" hierarchy is a
figment of my imagination, building on OUR imagination. For most of the
levels I've proposed, the only backing for the specific definitions is
anecdotal and subjective. As far as I'm concerned, these or any other
levels won't be "facts" until we have put them to experimental test.

I've always felt that defining the levels scientifically is a large
project, which should begin by experimentally verifying that people can
control variables of many different kinds -- anything anyone can think
of, without regard to levels. Even the most obvious variables should be
put formally to the Test, just so we can write down the parameters of
control and say that we have in fact observed such-and-such a variable
to be under control by a human being. This would be a beautiful thesis,
or series of them. On the other hand, maybe it should be the kind of
project to which all control theorists contribute, the way astronomers
put in some duty time measuring double star angles and separations (the
three well-spaced observations required to determine the orbital
elements needing, in many cases, 1000 years to complete).

Once we have a base of hundreds of certified controlled variables, we
can begin to try to put them into order. If there really is a hierarchy,
the variables will fall into classes and the classes will be related in
a hierarchical way. That is, in order to control a variable of one
level, it will be necessary to vary a controlled variable of a lower
level. And of course the only way to vary a controlled variable
arbitrarily is to alter the reference signal for the system that's
controlling it. Showing that this is the case leads to a new series of

When this project is done, HPCT will become a science. We will have
advanced from Galileo to Newton.

In the meantime, of course, we can argue. But without experimentation,
arguments are just a pastime.


I agree with Rick Marken about decisions: they represent conflicts.
Unless there were at least two competing goals to satisfy, there would
be no need to make a decision. You would just do whatever is required to
achieve the single goal. More commonly, there are multiple goals
involved in behavior, but we have learned to organize our actions (as a
result, largely, of resolving conflicts in the past) so that all thegoals can be
satisfied at once. When that is the case, again no
decisions are needed.

At the level I call "programs," symbol-handling processes occur which I
characterize as a network of choice points. These are like the TOTE
unit, in that there are tests for conditions, with the choice of a
branch being determined by a rule applied to the results of the tests.
The term "choice" seems to imply a decision, but in fact there are no
decisions at this level either. The conditions encountered at each
choice-point, plus the rules, completely determine the path to be
followed next. Only when there is ambiguity or when the rule is self-
contradictory (calling for more than one mutually-exclusive path to be
followed) is anything like a decision required. If you have an algorithm
for making decisions, you don't have to decide anything!

Note that operations occurring between choice points are sequences,
lists of reference levels to be brought about in order. Sequences are
the next level below programs. Programs are concerned ONLY with applying
rules to select branches, as I use the term here. They involve "flow
control" as they say in programming manuals. The parts of computer
programs that consist only of one instruction following another belong
at the sequence level here, not the program level.

If we eliminate programs -- the execution of algorithms for choosing
paths -- from decision-making, what is left? As far as I can see, only
the cases in which for some reason we wish to do two contradictory
things at once. At that point we must reorganize or simply suffer the
paralysis of conflict.

I think that we agree on this, at least to an important extent. You say

The Decision Making Entity, as here understood, can act
without being bound by past decisions. It frequently uses
them because they are readily available and alternatives
may be overlooked. It has the ability to be arbitrary.

This arbitrariness has the flavor of reorganization. But so far, at
least, I have not considered systematic reorganization. Anything that
could be called systematic, it seems to me, belongs in the already-
organized hierarchy. At the level of logic, systematic consideration of
previous choices and possible alternatives is an algorithm. As such, it
can be reduced to rules governing selection of paths connecting
sequences or lists of behaviors, where by behaviors I mean controlled
perceptions of the consequences of acting. When we remove all algorithms
by putting them into the program level of the hierarchy, all that is
left of decision-making is the arbitrary part: making a change for no

DEFINITION of the DME. The Decision Making Entity is defined in
terms of its Connections (Inputs and Outputs) and its

INPUT A) information about the Current Condition of
Physiological Systems.
INPUT B) information about past events -- memories, recordings,

This can be summed up by saying that the DME receives perceptions from
lower systems that are either in the Normal mode, receiving information
that comes ultimately from introceptive or extroceptive sensors, or from
lower systems that are in the imagination mode, deriving their
perceptual signals from memory.

REFERENCE LEVELS C) information specifying the acceptable
operating condition of the organism, A)a). These are the
"Intrinsic Levels" of other discussions.

This makes the DME look even more like the reorganizing system, with
reference signals specified genetically.

OUTPUTS D) information that acts throughout the Hierarchy.

This is typical of the reorganizing system as I perceive it. However, I
think that your DME includes both learned hierarchical systems and the
unlearned system I call the reorganizing system. When a "decision" is
reached, it must entail some sort of action, and to create any
systematic action a higher-level system must adjust reference signals
for lower-level systems. Furthermore, since nature never trusts an
organism's output to do what it is supposed to do, the consequences of
the action must be perceived by the level issuing the reference signals,
so the reference signals can be varied until the perceived result is the
intended one. If this control process takes place in an organized way,
it must be due to a learned system.

In my concept of the reorganizing system, I have extracted the arbitrary
non-systematic kind of action from the hierarchy as a whole and given it
a separate existence of its own as a built-in aspect of the organism
that functions from the beginning of life. We used to call this the
Negentropy System. I gave up the word because it implies things I don't
believe. I now just call it the reorganizing system.

A) Directing ATTENTION.
a) Selecting the information, INPUT A), to be controlled.
b) Selecting information from past events, INPUT B), for
comparison with the current situation.
c) Comparison of the current and projected ("anticipated")
situations with acceptable magnitudes of the variables
selected for control -- especially Intrinsic Variables, C),
when they are relevant. Everyday situations usually do not
directly involve Intrinsic Variables.

I will accept as part of the reorganizing system the direction of
attention. The rest I have incorporated into the hierarchy itself. I am
not sure what attention is for. We need to do some experiments to find

The comparison of current and projected magnitudes of variables with
acceptable magnitudes is simply the operation of any control system atany level
("projected" magnitudes require the imagination connection).
That kind of operation is adequately handled by the "canonical" control
system diagram, and when Intrinsic Variables are not involved is simply
the operation of the learned hierarchy of control systems. I allocate
Intrinsic Variables and Intrinsic Reference Signals strictly to the
reorganizing system, whose actions are arbitrary and random and serve to
alter connections and weights in the learned hierarchy. That is how the
learned hierarchy becomes organized.

B) Decision Making.
a) Selecting OUTPUTS, D), to be used by the DME as Reference
Levels for the Hierarchy.
b) Activating the Selected OUTPUTS for controlling the
Selected Systems.

I handle all this in the higher levels of the hierarchy, but would leave
decision making (as an arbitrary process) out of it.

In order to direct its attention and make its decisions, the
organism must be Conscious. Unconscious means that the DME is
unable to receive information from its inputs. However the
remainder of the systems may be functional, operating on the
basis of the most recent settings of their reference levels.
There are several interesting situations that can occur: sleep,
coma, paralysis, trauma, etc. These, and others, are worth
separate discussion.

I have formed a similar idea of consciousness (beginning with our
discussions of 35 years ago). However, I begin with awareness (which I
think you include). Awareness is the capacity of the reorganizing system
to receive information, regardless of its kind. I have proposed that
when awareness is receiving information selectively from a portion of
the hierarchy, the result is what we call consciousness. This allows us
to distinguish between one phenomenon which remains the same no matter
where it is applied -- awareness -- and another that changes its form
depending on the source of perceptual signals received in awareness --
consciousness. Consciousness always takes on the character of the
control systems to which awareness is connected.

Thus an apparent rule that seems to fit experience: you cannot be
conscious of systems that are in the conscious mode. Instead, you are
conscious of the lower-order world of perceptions received by those
systems, and experience those perceptions with the conscious
interpretation typical of the level (or levels) at which awareness is
connected. This interpretation appears to be an objective property of
the world.

Any system in the hierarchy can operate in the conscious or unconscious
mode. In the conscious mode only, it is subject to reorganization.

Implied by this model is the possibility that awareness can be
selectively connected to particular levels in the hierarchy. When that
is the case, you experience the world consciously as that level
perceives it, but you are unaware of applying any interpretation to
theperceptions. Instead, you see those perceptions simply as part of the
world. If you are operating in the logic or program level, you see the
world as full of choice-points and alternatives, with natural rules that
define a path through the choice-points. On the other hand, if you are
operating in the relationship level, you see a world in which everything
is related in some way; you see the constraints that make independent
objects and events maintain a certain constancy of interaction.

And while you are attending from the viewpoint of relationships, you are
not aware of any higher levels of perception and control. They are still
operating, and if you ask yourself why you are paying attention to
relationships, you will come up with higher-level reference signals --
what you hope to accomplish by attending to relationships. That is, you
can often "go up a level" and realize that higher-level control
processes were active all the time, even when not in consciousness. But
as soon as you do that, you are no longer seeing a world of
relationships. The nature of the conscious world changes as you move
awareness from level to level.

I think that this proposal is related to your concept of "modes."
However, I do not see these modes of consciousness as being modes of
just one level, your DME. I see them as resulting from awareness moving
from one place in the hierarchy to another. When one is attending to a
lower level of perception, higher processes are still operating but they
are not operating consciously. By your postulate, all modes would entail
consciousness of the highest-level processes.
Maybe you're right. But I think experience argues against this view.
At any rate, I think your picture is worth trying on for a fit.

This PROBLEM arose in our early discussions as we sought to
define higher levels. How can the changing behavior of an
individual be described when he is blocked? Analysis working
upward through the lower orders assumed (implicitly) a set of
fixed reference levels, especially at the higher levels.

This is no longer a problem in the hierarchy as I currently conceive it
(since 1973). Higher reference levels are no longer fixed, except at the
highest level. At intermediate levels, lower-order reference signals are
varied as needed to provide a higher-level system with the perceptions
it needs to match its own reference signals -- which in turn are being
varied as required by still higher systems. You might ask Rick Marken
for his spreadsheet demonstration of this arrangement -- it will run on
Lotus or Excel. It shows how a three-level hierarchy with six systems at
each level can simultaneously control three levels of perceptual
variables despite random disturbances from the environment, and despite
considerable interaction among the controlled variables at each level.

How could these be changed? How could the system be

They (reference levels) no longer require reorganization to be changed.
Reorganization is now needed only when the learned systems are not
capable of maintaining intrinsic variables at their reference levels (as
a byproduct of their actions). Since the model now includes many
"intellectual" functions such as classifying, ordering, reasoning,application of
principles, and control of system concepts -- all of
which are learned -- the reorganizing system does not have to carry out
any rational processes.

... what seems to be needed is the DME as suggested here.
I find combining the DME approach with the Hierarchical view
provides some additional answers, and leads to some revisions of

I think you will find that the levels I have added (categories,
sequences, programs, principles, and system concepts) contain much of
what you want to put into the DME. I agree that such functions are
required. I have simply broken them out into specific levels of
functions, while reserving the arbitrary reorganizing part to a separate
non-hierarchical system. I don't say that's right. It's just what I have
I have several arguments with your proposed levels, but will pass them
up for just one clarification concerning your Fourth Order, Temporal
Variables. For quite a long time after we parted, I considered just the
sequence level in the position where you put it, above configurations.
Then I realized that there are really two kinds of sequence variables,
one exemplified by the second-hand of a clock, and the other by the
notes of a melody.

The second-hand of a clock gives rise to a perception of continuing
angular motion, d/dt(angle). With angle as a configuration perception,
the new perception is simply its time derivative. As such it has a value
at all times, in present time, which can change in magnitude as the
angular (or other) velocity increases or decreases.

This is quite different from the temporal progression in the successive
notes of, say, "Taps," which can be played slowly or more quickly. In
the case of the melody, there is no simple motion signal, but the sense
of a specifically ordered progression of different sensations, one
following the other. What matters is not so much the speed, but the
ordering in time -- which note follows which.

On realizing this difference, I introduced the "transition" level, which
is basically derivatives (and perhaps derivatives of one variable
relative to another). This level went just above configurations, and is
where stroboscopic as well as continuous motion or change is perceived.
That left the sequence level to cover just the temporal ordering of
lower-level variables, including transitions. I called this the "event"
level, where an event was supposed to be a short familiar temporal
pattern of perceptions of transitions, configurations, sensations, and
intensities (you omit intensities).

Only a couple of years ago, Gary Cziko brought up some more examples of
temporal variables in which ONLY the ordering was important -- in
language, for example, the ordering of words. Here the temporal pattern
was not evident, for an ordering is quite independent of how long it
takes elements to occur or of the spacing between elements. This struck
me as different from an event which which there is a stereotyped unitary
pattern that forms a single package in perception. So the sequence levelended up
being split once again, the event level now meaning only brief
"packaged" temporal patterns recognized as a single thing like the
bounce of a ball, with pure sequential ordering -- lists -- being moved
to a higher level.

We can discuss the rest of your proposals for levels later -- I expect
that others will have questions and comments, too. I am glad to see the
subject opened up again, because I don't like the sensation of having my
hypotheses converted into Gospel. I think that by trying to boil down
all propositions to the basic underlying operations, and connecting them
with experience, we can arrive at an agreeable set of levels for
experimental test. Maybe the reason that there has been so little
questioning of my definitions is that nobody saw any conflicting
alternatives and thus didn't feel compelled to make a decision!

Bill P.