At the very least, there is the
possibility of there being in social interactions the same kind of
homeostatic loops as are hypothesized to exist at the top level of the
hierarchy within an individual – negative feedback loops that maintain
the level of their variables at some point around the loops, without
necessarily having anything equivalent to a reference input to a
comparator.
[From Bill Powers (2008.02.04.0820 MST)]
Martin Taylor 2008.02.03.11.30 –
I defined “reference condition” in B:CP as “that state of
a controlled quantity at which a control system’s output ceases to tend
to alter the controlled quantity.” A reference level was then
defined as the numerical measure of a reference condition. This means
that all control systems (in fact, all feedback systems) exhibit
reference conditions or reference levels. When the reference level proves
to be variable, as it does in both behavioral and homeostatic systems
(see Mrosovski’s “Rheostasis”), there is something playing the
part of a variable reference signal that can be used to adjust the state
of the controlled quantity. The definition says (deliberately) nothing
about whether the reference condition is a stable or unstable equilibrium
point.
The existence of
such homeostatic loops is almost inevitable in a large system of
interacting elements, whether those elements be simple chemical reactions
or complex control systems.
This is true if you interpret “homeostatic” loosely. But
control requires more than just a negative feedback loop: it requires a
significant amount of loop gain. Without a loop gain considerably more
negative than -1, the ability to resist disturbances and keep a
controlled variable near its reference condition is reduced to the point
where the environment can be said to control the control system as much
as the control system controls the environment. In systems without power
gain, the control system merges into the environment, becoming
indistinguishable from it. One could then draw the system-environment
boundary anywhere with equal justification, or simply treat the aggregate
as a single system.
While negative feedback exists in many physical systems, therefore, it
does not automatically imply that there is a control system interacting
with an environment. A marble in a bowl will, after a disturbance, return
eventually to the bottom of the bowl; if there’s water in the bowl the
released marble may not even overshoot the equilibrium position. But the
loop power gain in such a system is 1 or less; the marble controls its
relationship to the bowl and so does the bowl. We really have a single
system that has no natural boundary between controller and controlled.
Weiner, Rosenbleuth, and Bigelow didn’t think of this analysis when
debating (about pendulums) with the philosopher Taylor about purposive
systems.
In a collection of interacting human control systems, we have a somewhat
different case but with a similar result, I think. Here each system has
considerable loop gain in relation to its own controlled variables, but
if we consider only the social variables, the environment of each system
has about the same input-output gain as the average system we focus on.
So again there is symmetry and it is difficult to separate the controller
from the controlled. We have interactions, but not, automatically,
control of one system by another.
When you have
stable negative feedback loops, and they pass through complex control
systems such as humans or other pack animals, we could give them the name
of “social feedback loops”. It would be very strange if nowhere
in any of these loops were any of the signal paths to include reference
signals inside any of the individuals.
By what path would these social interactions be able to influence the
top-level reference signals?
When you think of
it this way, “group control” deson’t necessarily mean anything
to do with teams; it means the effect of the combined behaviour (control
of perception) of more than one individual, behaviour that stabilizes
some value of a signal somewhere in the social feeback
loop.
We can arbitrarily set the cutoff for using the term control at a loop
gain of -1. At that point there is no control because the controlled
controls the controller just as much as the other way around. In passive
physical interactions input-output power gains are always 1 or less, so
no system made only of interacting passive components can exhibit
control. I’m no chemist, but it seems to me that this narrows the field
of control pretty much to systems containing catalysts or sensors as well
as large amounts of stored energy, so amplification can take
place.
“Social feedback loops,” in the light of the preceding, are not
necessarily the same things as “Social control systems.” In
fact there is one class of social feedback loop that can be eliminated
immediately: the loop that is generated when the actions of two negative
feedback control systems disturb each other’s controlled variables with
both disturbances being in the same sense (both aiding or both opposing).
It can be shown that this results in a positive feedback loop, with
instability resulting when the two-system loop gain exceeds +1. You get
either a flip-flop relationship or oscillations, and certainly no
control.
In looking for possible social control systems, therefore, we have to
look for cases in which all loop gains are negative and there is an
assymetry of effects such that a controlling system can be distinguished
from a controlled system. This does not include the case of “ganging
up”, in which one set of control systems combines forces to
overwhelm a smaller set or an individual. That simply removes the
unsuccessful system from the collection of control systems since it can
no longer control (its “marginal loop gain,” to borrow a term,
is zero). In fact, for a higher control system to exist, it must act
strictly through adjusting the highest reference conditions or the input
functions in the proposed “lower” systems, because acting
arbitrarily to alter any other signal or function would simply arouse
opposition from the highest level being controlled. I think there is
probably a necessary principle to the effect that a lower system can’t
resist control by a higher system, because it doesn’t even know it’s
being used for higher-level control. It can’t perceive anything at a
higher level.
Looking at it that
way makes it easy to think of this kind of homeostatic social loop as
equivalent to the top-level control in an individual, and almost demands
that we look to see whether there might not also be social signal paths
that are equivalent to the perceptual and output signals within an
individual hierarchy.
I really think we need at least one concrete example of this – and if
such an example could be found, that alone would greatly weaken if not
negate this proposal simply because we can comprehend it. That would say
it is not at a higher level than exists in any individual.
Just as a neuron
can’t see its role in the larger control system of a person, maybe it’s
hard for us to see how our actions form parts of multiple interlocking
social feedback loops that stabilize the social structure (or detabilize
it if the gain goes positive and greater than
1.0).
When the proposal gets to this point, I think it reaches a dead end. It
amounts to saying that we are all controlled by social systems of which
we can know nothing. If we could know of them, they would be individual
perceptions, not social systems. And if we can’t know of them, they can’t
be shown to exist in any way that a human being could comprehend. What we
are left with is The Boogie Man, the powerful invisible entity that
governs our lives leaving us with only the (convincing) illusion of
autonomy.
It is this that
makes the statistical exploration of the social environment valuable, not
the possible reinterpretation of the social statistic as applicable to
individuals (valuable though this latter may sometimes be). The
statistical environment of a person has a huge differential effect on
what kinds of feedback loops the person’s actions may participate in.
That matters.
Again, this comes down to a self-defeating concrete demonstration. If you
can statistically demonstrate a social control system that has a gain
more negative than -1 and that operates without our knowlege, from then
on it no longer can operate without our knowledge and we can decide for
ourselves whether or not to be controlled by it (with the usual
tradeoffs). That converts it merely to a social interaction, not a
control system. If it doesn’t have a loop gain more negative than -1,
it’s just an interaction anyway.
There are, I’m sure, many presently-unknown facets of social interactions
that we can explore using PCT, but we will never know anything about
social control systems of which we can know nothing. What’s the point of
even discussing them?
This whole discussion is pertinent to previous discussions about
influences over us which are inherently incomprehensible.
Best,
Bill P.