Albus, Appearances

[From Rick Marken (930823.0900 PDT)]

Michael Fehling to Bill P. (930821?) --

Sincere thanks for the extremely thoughtful and detailed discussion of your
views on Jim Albus' hierarchical control model and its (non)relation to PCT.

Michael. I would also be interested in your reaction to my comments
(Rick Marken (930820.2000 PDT)) about Albus' model that I made in a post
with the subject head "Good disturbance, control of perception". In that
post I claimed that any input-output model that exists in a closed
loop negative feedback relationship with its environment is a perceptual
control system. I just don't think Albus understands the implications
of that fact. At the end of that post I encouraged Bill Powers to
explain what the implications of that fact are for PCT modelling.
Bill did that, but apparently not in response to my plea. Bill made
two main points, I think: 1) PCT models behavior from the perspective
of the behaving system (the test for the controlled variable is a means
of determining what that perspective is -- ie. what perceptual variables
are under control) and 2) PCT models run "on their own" just like real
behaving systems; an effort is made to make sure that the modeller is
not part of the model. I think these two points capture the essential
difference between PCT and conventional approaches to understanding
behavior. They sound like very simple concepts but they seem to be rather
difficult for many conventional psychologists to understand or accept.

Let me suggest that one of the major reasons why conventional psychologists
(especially behavior modellers -- including AI types) have not taken PCT
seriously is a preoccupation with APPEARANCES. PCT is concerned with
how behavior works in the real world of invisible and unpredictably
changing disturbances. These disturbances are typically invisible to
both the behaving system AND the observer of that system. If a modeller
builds a simulation that displays some complex behavior (like a six-
legged gait) in a world without disturbances, the behavior of this
model will, nevertheless, impress an observer. A model that performs a
simpler behavior (like pointing at a target) in a realistic environment
(with disturbances) will be less impressive because the observer cannot
see the disturbances -- just the apparently simple appearance of behavior.

The preoccupation with appearances is understandable -- people have been
amused and astounded by the complex appearance of the behavior of automata
since people started building such things in the 17th century. The behavior
of living systems, however (as Bill Powers puts it) "is not a show put on
for the benefit of others". What we see as "behavior" in living systems
is a SIDE EFFECT of the process of perceptual control; living systems
continuously vary outputs AS NECESSARY in order to keep their perceptual
experiences matching internally selected (and continuously changing)
reference specifications for these perceptions. The goal of PCT modelling
is to find the perceptual variables that, when controlled by a model
in the same kind of environment as the one accupied by the living
system, produce the same kind of behavioral appearances (side effects
of control) for the observer. We have been able to do this with, what
I think is, rather impressive success.

Fourth, perhaps you can answer a question regarding the relevance of PCT to
my lab's current research focus on social/organizational systems. My group is
now developing a theory of organizational problem solving as a control system
that manages the "sensory effects of outputs" (to borrow your phrase).
So, can you tell me whether you or anyone else in
the PCT community has applied this theory to group/organizational/social
systems?

Some of the best research on the application of PCT to the behavior of
interacting control systems has been (and is being) done by Tom Bourbon,
who, I'm sure, will be responding to your request for information pretty
soon.

In the mean time, perhaps you could describe your social/organizational
research in a bit more detail. I think it would be very interesting to
see how PCT might deal with the kind of behavior you observe.

Best

Rick

From Tom Bourbon [930823.1230]

[From Rick Marken (930823.0900 PDT)]

In reply to

Michael Fehling to Bill P. (930821?) --

Michael, I agree wholeheartedly with Rick's assessment of the difficult job
we seem to have in selling PCT modeling to the AI and motor plan
people. Appearances seem to count as everything, when people are shopping
around for models, and the tracking-pointing-crowd-bacterium nodels from PCT
don't appear as impressive, on casual inspection, as the results of
traditional puppet-animation aproaches. But first impressions can be
deceiving, and the performance of PCT models in disturbance-prone
environments sets a standard for modelers using other approaches to achieve.
But it certainly would help our case were someone to build a six-legged PCT
model that totters around in a disturbance-filled setting -- then we would
catch people even after their first glance.

Michael:

Fourth, perhaps you can answer a question regarding the relevance of PCT to
my lab's current research focus on social/organizational systems. My group is
now developing a theory of organizational problem solving as a control system
that manages the "sensory effects of outputs" (to borrow your phrase).
So, can you tell me whether you or anyone else in
the PCT community has applied this theory to group/organizational/social
systems?

Rick:

Some of the best research on the application of PCT to the behavior of
interacting control systems has been (and is being) done by Tom Bourbon,
who, I'm sure, will be responding to your request for information pretty
soon.

In the mean time, perhaps you could describe your social/organizational
research in a bit more detail. I think it would be very interesting to
see how PCT might deal with the kind of behavior you observe.

Thanks for the nudge, Rick. I am posting this during a short break
and will not get back to the mail until the end of the day. I want to reply
to Michael's inquiry about social applications. Michael, if you post a
description of your work on social phenomena, as Rick suggested, it would
help me, and several others on the net, see whether or not our work
addresses some of questions that interest you.

Until later,
  Tom

In re Tom Bourbon 930823.1230 --

Tom,

You asked for some more information about our research. I've just sent a
rather long comment directed to Bill and Rick. In it I mention some of the
research we are doing (albeit rather abstractly). Let me add one other item.
  We are developing a model of organizational effectiveness as part of our
research on organizational design. This model depicts
        (a) coordination among the decision processes in an organization as
            they manage "physical" processes (e.g., workflow in a manufacturing
            setting (e.g., management activities such as forecasting, setting
            of production policies
        (b) resource flow among the physical processes and with units in the
            external environment (e.g., a manufacturing "supply chain")
        (c) information flow among the decision-making elements and the
            physical processes,
        (d) information flow between organizational units and the external
            environment (e.g., resources providers, consumers of output,
            etc.), and
        (e) activities of the elements in the organization's environment in
            regard to information and resoource production/comsuption.

We are developing this model to explore the interaction between an organzation
and its environment. In particular, the organization and its constituent
elements operate in an ongoing, closed-loop relationship with its environment.
We are studying how the organzation controls its own operations in order to
achieve its objectives. Similarly to PCT (I think), our model requires that
the organization and its decision making units accomplishes this control in
terms of its interpretations of environmental conditions. These perceptions
and their (lack of) correspondence with "objective" external condtions is
impacted by the changing beliefs and performance skills of the organization's
actors. Changes in these beliefs and skills result from
        (a) new information from external sources,
        (b) problem solving done by organizational units (e.g., market
            forecasting), and
        (c) "institutional" inputs (e.g. social norms, governmental policies,
            industry standards of practice, professional standards, etc.)
This inflow arises in part by communication between the organization and
external sources and in part by inflow of new organizatinal participants.
  In sum, this model approaches organiational performance as a process of
adaptive control. Since we emphasize the way in which the organzation's
perceptions do or do not correspond to actual external conditions, our model
seems compatible with the tenets of PCT.

[From: Bruce Nevin (Tue 930824 08:57:08 EDT)]

Mon, 23 Aug 1993 18:40:21 -0700
Michael Fehling <fehling@lis.stanford.edu>

In re Tom Bourbon 930823.1230 --

the organization and its constituent
elements operate in an ongoing, closed-loop relationship with its
environment.
We are studying how the organzation controls its own operations in order to
achieve its objectives.

If this is so, then organizations differ from organisms in an interesting
and perhaps revealing way. Organisms vary their "operations" in whatever
way suffices to make their real-time perceptions match their objectives
(their reference perceptions). Organisms do not control their operations
(behavioral outputs). Perhaps organizations could be run this way too.

  In sum, this model approaches organiational performance as a process of
adaptive control. Since we emphasize the way in which the organzation's
perceptions do or do not correspond to actual external conditions, our model
seems compatible with the tenets of PCT.

Similarly to PCT (I think), our model requires that
the organization and its decision making units accomplishes this control in
terms of its interpretations of environmental conditions.

Who specifies what the "actual" and "objective" external conditions are?
Is that agent in the organization or not?

These perceptions
and their (lack of) correspondence with "objective" external condtions is
impacted by the changing beliefs and performance skills of the organization's
actors. Changes in these beliefs and skills result from
        (a) new information from external sources,
        (b) problem solving done by organizational units (e.g., market
            forecasting), and
        (c) "institutional" inputs (e.g. social norms, governmental policies,
            industry standards of practice, professional standards, etc.)
This inflow arises in part by communication between the organization and
external sources and in part by inflow of new organizatinal participants.

Are you talking about what happens when the wrong perceptions are
controlled, or when reference values are inappropriately set for
controlled perceptions, or when coordinating agents are not controlling
the "same" perceptions (as they believed they were), such that agents in
the organization perceive that the organization is at risk, and that they
may lose control of their perceptions of their "livelihood" etc.?

    Bruce Nevin
    bn@bbn.com

[From Rick Marken (930823.0900)]

Michael Fehling (930823a)--

we view
agents as engaged in on-going (i.e., continuous and real-time), closed-loop
interaction with its physical and social environment.

We do too. But we also note that this closed loop interaction results in a
phenomenon called "control". Control is the process of keeping a perceptual
variable, p, matching a constant or varying reference specification, r.
Agents
maintain control (keep p = r) by varying their actions (o) as necessary
to counteract environmental disturbances (d), keeping the environmental
correlate (i) of the controlled perceptual variable at the value that produces

the appropriate value of p. The process of keeping p = r is also called
purposeful behavior. So the phenomenon of control is equivalent to what
has always been described as "purpose".

However, in any general model of control of action, including PCT, some
reference standard must be posited.

We've got some verbal problems here. PCT is not a model of control of
"action". We use the word action to refer to organism-produced
influences on the environmental correlates of controlled perceptual
variables. These actions are not controlled; they are caused -- ultimately,
by efferent neural signals. It is the perceived consequences of these
actions that are controlled -- ie. maintained at a constant or varying
reference level (specified inside the agent) against environmental
disturbance. The reference "standard" is not only positied by PCT --
it is one of the most important concepts in the model. The reference
standard, r, is an agent-specified neural signal -- the same KIND of
signal as the perceptual signal, p -- that "commands" the intended
value of p. The negative feedback closed loop that runs through the
environment is responsible for the fact that p TRACKS r -- p is
controlled relative to r.

A task, goal, or any
other representation of an objective is just a reference standard at a
relatively high level of abstraction. So, when an onlooker attributes task
objectives to an agent, s/he seems not to be violating the spirit of control
theory such as PCT.

This is true -- r could specify a p that represents a task like
"hammering a nail". Bill's point (I think) is that the "task" that the
observer sees may not correspond to the perception(s) that the
agent is controlling. There is a formal procedure, called "the test
for the controlled variable", that must be used by observers to test
hypotheses about the extent to which aspects of their own perceptual
experience (what most of us succumb to calling the "actual state of
affairs") corresponds to perceptual variables that the agent is
controlling. For example, it may look, to an observer, like an agent
is hammering a nail when, in fact, the agent is controlling a rhythmic
pattern that s/he wants to try at the club latter that night.

The validity of these ascriptions is, however, an empirical matter.

Yes, indeed. And "the test for controlled variables" is an empirical
approach to testing the validity of these "ascriptions" (hypotheses
about controlled perceptual variables). I have never seen any
formal or informal use of "the test for controlled variables" in any
of the literature on "task analysis" or cognitive modelling. I would
be very interested to know if the "validity of these ascriptions"
has ever been tested empirically in cognitive psychology.

So, Albus' model is not flawed just because it "speaks of
'tasks' that an organism carries out.

It is flawed because he never "tested the validity of his ascriptions".
Like most behavior modellers, he was "flying by the seat of his
pants", taking appearances (his own perceptions of behavior) com-
pletely for granted.

Instead [of "behaior"], I prefer the term "action" (as a
capability subjectively defened by, perceived by, and under the control of,
the agents being described).

Sounds like your "action" is what we call a "controlled perception", p.
as I said, we use the term "action" to refer to uncontrolled outputs (o).

I try to reserve the term "behavior" for
reference to observable properties of actions.

We usually use the term "behavior" to reer to the controlled consequences
of actions (as perceived by a observer). The problem is that "action"
and "behavior" are relative terms; a behavior in one context may be
an action in another. The movement of a hammer is a controlled result
of muscle forces; so hammer movement is a behavior -- a result of
the actions of the muscles. But movements of the hammer are used
to drive nails into wood. So now hammer movements are the actions
that produce another controlled result -- "nail into wood" behavior.
In PCT, "action" is synonymous with "means" and "behavior" is synonymous
with "ends". The relativity of action and behavior is captured by the
hierarchical structure of the PCT model.

Of course, the problem is that no descriptive theory can avoid commiting to
some things that in another theory are "mere appearances."

True. And it's one reason PCT is a GENERATIVE theory (it actually behaves
and is tested by the degree to which it mimics the behavior of living
systems) NOT a descriptive theory.

1. Many of the perceptions that agents control are just the kinds of
appearances of which you warn. I.e., they are attributions that the agent
uses to structure its self-perceptions and then use these attributions for
self-control.

Have you tested to determine that this is, indeed, what these agents are
controlling? Even if you have not, I agree that aspects of what people
control are based on "imagination" (self-generated perceptions that
are not based on external reality) -- so, in general, p =f(i,im); perceptual
variables are a function of both the sensory consequences of environmental
variables (i) and self-generated perceptual inputs (im). I think this is what
you mean in your "i.e.".

Do you know of any work in PCT that would be relevant?

I have a working hierarchical control model (3 levels, 6 systems at
each level) that is implemented in Lotus and Excel spreadsheets.
A different "type" of perecption is controlled by the systems at
each level of the hierarchy. It is possible to put some of the
systems into "imagination mode" so that the perceptual signals
in higher level systems are based, in part (or in full) on imagined
perceptions. It is possible to see the consequences of this kind
of imagination on the operations of the control hierarchy (these
consequences are surprisingly small).

Social psychologists and others have proposed another important type of
"appearance"--a concept or attribution that plays a role in the social
construcion of the self. Here, too, I am approaching this topic from a
systems perspective. Is there PCT-related work that tries to shed light on
relevant topics?

PCTers have done very little work on the control of high-level perceptual
variables (like "self") or on the workings of imagination (attribution).

Michael Fehling (930823b) to Tom Bourbon 930823.1230 --

We are studying how the organzation controls its own operations in order to
achieve its objectives.

An organization is a collection of agents (human control systems) is it not?
I don't see how this collection of agents can have objectives. The agents
can have objectives becuase they have nervous systems that can function
as control systems. Organizations can "appear" to have and achieve
objectives -- but this is truly an appearance. It is not what is going on.
I think you should take a look at Bill Powers' "Gatherings" program to
see how the appearance of organizational "control" (achievement of or-
ganizational objectives) "emerges" out of the operation of many independent
control system controlling variables that sometimes involve one another.

Similarly to PCT (I think), our model requires that
the organization and its decision making units accomplishes this control in
terms of its interpretations of environmental conditions.

In PCT, the "interpretation" of environmental conditions determines
WHAT is controlled. The controlled variable, p, is a function of environ-
mental conditions, i; p = f(i). f() is the "interpretation" of the
environmental
conditions. The nature of f() determines WHAT aspect of the environment
is controlled. If this is what you mean by "accomplishes control in terms
of its interpretations of environmental conditions" then your model is like
PCT in this sense.

In sum, this model approaches organiational performance as a process of
adaptive control. Since we emphasize the way in which the organzation's
perceptions do or do not correspond to actual external conditions, our model
seems compatible with the tenets of PCT.

PCT does not make a big thing out of correspondance; you control what you
perceive: that's it. I don't see how organizations can have perceptions --
agents in the organization can have them, organizations can't (as far as
I know). PCT certainly is relevant to the behavior of organizations -- but
it would model this behavior in terms of agents, not fictional entities
(like organizational perceptions and objectives).

Best

Rick

From Tom Bourbon [930824.1028]

Michael,

Thank you for providing additional information about your work on
organizations and about some of the ways you think PCT might be useful in
that work. Your post allows me to see more clearly how your work differs

23 Aug 1993 18:40:21 -0700,
Michael Fehling <fehling@LIS.STANFORD.EDU> writes:

In re Tom Bourbon 930823.1230 --

You asked for some more information about our research. I've just sent a
rather long comment directed to Bill and Rick. In it I mention some of the
research we are doing (albeit rather abstractly). Let me add one other item.
We are developing a model of organizational effectiveness as part of our
research on organizational design. ...

You listed several topics: "(a) coordination among the decision processes
    in an organization as they manage "physical" processes;" "(b) resource
    flow among the physical processes and with units in the external
    environment;" "(c) information flow among the decision-making elements
    and the physical processes;" "(d) information flow between
    organizational units and the external environment;" "(e) activities of
    the elements in the organization's environment in regard to information
    and resoource production/comsuption."

We are developing this model to explore the interaction between an organzation
and its environment. In particular, the organization and its constituent
elements operate in an ongoing, closed-loop relationship with its environment.

Several differences stand out between your work, or at least between your
description of the model you are developing, and most of the work within the
PCT community on social interactions. Perhaps these points of difference
will suggest to you some ways that PCT might help solve problems you
encounter in developing your model of organizations.

One apparent difference is in our use of ideas about information flow and
transfer. To date, none of the published material on PCT modeling,
including modeling of social phenomena, has included measures of
information. They have not been needed. Martin Taylor and his associates
are at work on projects in which that circumstance might change.

Another apparent difference is in our characterizations of organizations.
You speak of the interaction between an organization and its environment,
and of how the organization *and* its constituent elements operate in a
closed-loop relationship with *the organization's* environment. Up to now,
when social phenomema have been addressed within PCT, all of the modeling
and most of the speculating have represented social a organization, not as an
entity standing alongside or apart from its constituent elements, with its
own goals and interactions with the environment, but as an assemblage of
individuals, each acting to satisfy her or his own reference perceptions.
On that construal, the organization as such exists "only" as an idea,
probably as one among many reference perceptions held by each individual in
the assemblage. By that interpretation, each individual in the assemblage
is said to "operate in an ongoing, closed-loop relationship with its
environment;" but "the organization" is not. (I recognize that some
participants on this net will disagree with my interpretation of the
organization-individual relationship. If you have not been looking in for
long, Michael, prepare for a round of debate.)

These two apparent differences in our research programs concern topics that
are less fundamental than some others and are themselves the subjects of
frequent discussion and debate on this net. However, a few of the other
differences go to the heart of the behavioral model in PCT. Perhaps these
points of seemingly greatest difference will be those on which PCT can be
most useful to you.

These points occur in the following sections of your post. (Bruce Nevin
(Tue 930824 08:57:08 EDT) also replied to you concerning these points.)

We are studying how the organzation controls its own operations in order to
achieve its objectives.

As Bruce said in his reply, this sentence suggests a profound difference
between the behavior of organizations, as they are represented in your
model, and of individual organisms, as they are represented in PCT. In PCT,
organisms are understood and modeled as systems that allow their actions or
behaviors to vary any way necessary to achieve their intended perceptions:
they control their perceptions, not their actions. We will need to explore
this difference in some detail.

Similarly to PCT (I think), our model requires that
the organization and its decision making units accomplishes this control in
terms of its interpretations of environmental conditions.

This sentence is an example of how the apparent differences I discussed
earlier, and the genuine difference in our identification of that which a
control system controls, combine to produce serious consequences. In PCT,
organisms are modeled as controlling their perceptions, not their actions,
and organizations are construed as assemblages of individuals each of whom
controls his or her own perceptions. By that interpretation, an organism
accomplishes control *of* " its interpretations (TB: perceptions) of
environmental conditions," not "in terms of" that interpretation. This is
another significant topic for us to explore.

These perceptions
and their (lack of) correspondence with "objective" external condtions is
impacted by the changing beliefs and performance skills of the organization's
actors. ...

In his reply, Bruce also addressed this topic. When we say the PCT model
adopts the position of the controlling agent, we mean that "all the way."
We model an organism as a controlling its perceptions by way of actions that
go into a world the organism knows *only* as perceptions. It is as
though the organism affects and controls what it sees on the viewfinder of
a camcorder, but has no way of to look around from behind the camcorder to
peek at what is "really" (objectively) out in the world. What the organism
sees on the viewscreen is all there is. This construal of organism-
perception-"reality" relationships differs from the one you described. The
difference is one that will require some exploration by all of us. And I
believe this difference makes clear the importance of one of the apprarent
differences I mentioned earlier -- the two different ways we characterize
organization-individual relations. However they are related, the
individuals in an organization are the only means by which the organization
can "know its environment," and those individuals do not know it
"objectively." They know only what they perceive and, if the perception is
one for which an individual has a reference perception, whether the
perception matches the reference. Much for us to explore.

In sum, this model approaches organiational performance as a process of
adaptive control. Since we emphasize the way in which the organzation's
perceptions do or do not correspond to actual external conditions, our model
seems compatible with the tenets of PCT.

See what I mean?

It is good to have you on the net.

Until later,
Tom Bourbon
Department of Neurosurgry
University of Texas Medical School-Houston Phone: 713-792-5760
6431 Fannin, Suite 7.138 Fax: 713-794-5084
Houston, TX 77030 USA tbourbon@heart.med.uth.tmc.edu

···

from my modest applications of PCT to simple social interactions.

[From Oded Maler (930824)]

I have some Deja Vu's form the recent thread going on
with Michael Fehling, and I glad to remeark that certain
PCTers have progressed with the years and understood not
everybody else is a good-old-mainstream-paper-rejecting
psychologists, and that the commonalities and differences
between PCT and new-age-situated-robotics-AI are better
understood and chrystalized. In particular:

1) One can control only what it perceives - this is today
a tautology for modern AI robotnics, and at first instant
they don't understand what's the big deal.

2) The dichotomy starts when coming to purpose (of the reseaerch).

Appearances is what you sell when you design machines. The customer
of the robotics/manufacturing systems doesn't give a damn on
the internal "mental world" of the machine (except for prestige,
e.g., this machine was implemented using neural-net imlementation
of Fuzzy PCT) but on observable output (as perceived by the client).

At this stage a PCTers might say (as some said in the past) that
it is not interesting. But today, I think, it became clear that
among all other disciplines, practical AI-robotics is more open to
good ideas that work, as researchers are less comitted to behavioral
theories contradicted by PCT. If a PCT-based design will solve
a hard problem in robotics, no one could ignore it. And to sell
such a solution to a roboticist, you need not show him what behavior
"really" *is* (unless he is particualriliy interested) because it
is irrelevant for the machine he builds.

3) Concerning the specific research issues that Michael raised,
it seems that it might lead to areas of controversy concerning
higher-order entities such as organizations. What is a factory
controlling for? Should every perceptual variables of the factory
be encoeded using one transmission line ("one neuron for one percept")
or should they be distributed in time and space?

--Oded

···

--

Oded Maler, VERIMAG, Miniparc ZIRST, 38330 Montbonnot, France
Phone: 76909635 Fax: 76413620 e-mail: Oded.Maler@imag.fr