Third Try

[From Rick Marken (930827.1200)]

Here is my third attempt to put this post on the Net.

My first attempt to post it was 930826.0900.

···

------

Avery Andrews (930828.1508) --

Although organizations are indubitably made up of lots of

individuals

controlling for things, it still might (or might not) be appropriate
to think of them as unified control systems. We know, for example,

that

mammalian nervous systems involve big populations of neurons

interacting

in extremely obscure ways, but this doesn't stop us from thinking

that

`lumped variable' models can be useful

I think this is a mistaken analogy. "Lumping variables" is not the
problem;
it's the way the individuals interact that differentiates an
organization
of people from an organization of neurons (and other variables).
Take
two elements, A and B, that interact in an "organization". If this
organization is a nervous system that the elements are neurons
(say a sensory and an interneuron) and a and b can represent the
variable firing rate in each neuron. The normal mode of operation of
these elements is that the firing rate in A causes a proportional firing
in B. So A and B are causally related (with a little stochasticity
thrown
in) so that, to a good approximation b = f(a). B has no choice about
this;
its firing rate depends on the firing rate in A and that's that. The
operation of a control system made out of neuron REQUIRES this
causal
dependence between individual components. "Lumping" does not
change
this causal dependence in the nervous system. If A and B are nerves
rather than neurons (and a and b are the firing rates of the nerve --
the lumped set of neurons) the same causal dependence exists (we
know this from studies of the nervous system). So in the lumped
system
it is still true that b = f(a) and a control system based on the "lumped
variables" works AND corresponds to the reality of how its
components
work.

Now suppose that the organization is a factory and the elements are
people. A is one person, B is another. Let's say that A is a manager
who
is supposed to get B to inspect a certain number of widgits per
hour. A does this by doing something to B -- say talking to B. So
person
A talking is like neuron A firing -- so the level of "talk" is variable a.
The
rate of widget inspecting by B is like neuron B firing so variable b is
widget
inspecting rate. But it is NOT true (as it is with the neurons) that
b is causally dependent on a. It might seem like it much of the time
--
B might generally do what A says ; so it LOOKS like b = f(a). But
since A and B are control systems (because thay are people) we
know
that it is VERY LIKELY that the relationship between a and b exists
because of (among other things) the current setting of B's reference
signal for "cooperating with the manager". Lumping As and Bs
together
does not change this situation -- there is NO determinstic relationship
between a and b (when these variables are the inputs to and outputs
of
hierarchical control systems) as there is between a and b (when
these
variables are the inputs and outputs of a mechical system --the
neuron).

Oded Maler (930826 15:00 ET) --

Concerning organizations, one can supress these meta-physical
questions, and assume that an organization has some higher-level
perceivable goals ($$ as written in some data-base) which are
related to some lower-level preception-action hierarchies in
the organization's sub-systems (production, marketting, etc.)

But these higher level goals only exist as perceptions (and
references)
in the individual members of the organization. "The organization"
has no goals -- just the individuals. Even if you write the goals on
parchment (or stone tablets) they can only be implemented as they
are perceived, interpreted and turned into references by each
individual seperately. Any model of an organization MUST take this
physical FACT into account. It would not be correct to model a group
of
individuals with the "higher level goals" OUTSIDE all the individuals.
This is an important point becuase it would be easy to make this
mistake in a computer simulation. For example, I could build a
simulation
of a group of control systems and have memory locations containing
the reference levels for ALL individuals -- functionally acting like
"tablets" in the external environment providing reference signals
directly into the control systems of the individuals. This is do-able
in a computer simulation but it doesn't happen in the real world
becuase
control systems (which are in brains) cannot receive reference
inputs directly -- unless surgery is performed. I could build a
computer model which would be equivalent to having each of the
ten commandments connected to an electrode that goes directly into
the brain of each individual in the tribe. So "thou shalt no steal"
becomes
an r = 0 going into each individuals control system that controls the
perception of "stealing". "Thou shalt have no god before me"
becomes
an r = 1 going to each individual's "god sequence" perceptual control
system, specifying that what is to be perceived is "yahweh, baal,
isis.."
and not "baal, yahweh, isis.." . etc.

Although such models could be implemented on a computer they
would
not correspond to the physical reality of organizations. The
individuals
in organizations do not get the "organizational goals" piped directly
into their brains as reference signals. These goals (spoken, written,
or whatever) enter each individual as an input variable; they are
experienced as a perception and, perhaps, turned into references by
the individual. In modelling organizations it is VERY important to
keep this fact in mind. Saying that "the organization" controls is
not just a philosophical mistake; it can lead to real mistakes in
modelling,
because the equations or computer program will not prevent the
modeller from creating a physically impossible model -- a model that
actually DOES have the organization controlling.

Best

Rick