mutuality

[Martin Taylor 931007 19:30]

In introducing PCT to a control engineer in our Human Factors division
yesterday, I came across a relationship that I am not aware of having
been discussed (but I haven't read all of LCS 1 and 2, so, like emotion,
it may be there). I call it "mutuality."

In "The Selfish Gene" Dawkins uses the image of the chromosome as a team
of agents, each of which does something that benefits the reproductive
success of the others, though not directly of itself. Genes that have
this kind of complementarity tend to be found nearby on a chromosome
because chromosomes can break and reconnect differently, reducing the
chances that distant genes will, over long time scales, be passed together.
Likewise (not from Dawkins) if there are two cells, each of which produces
as a side effect of its own operation a substance that enhances the operation
of the other, the bi-cell will survive better than either alone, if they find
a way of sticking together physically, and of reproducing in tandem.

I propose that it is very likely that evolutionarily early control systems
may have joined in such mutual arrangements, and that evolved control systems
continue to do so. Here's the argument.

A single ECS (A) controls its perception in the presence of disturbances to
its CEV (CEV-A). It can neither predict nor directly observe these
disturbances, except as fluctuations in the value of CEV-A. Any ECS has
limits on the speed with which it can compensate for the disturbances, and
the range of disturbance force that it can counter.

Consider a second ECS (B), which has as its CEV-B some part of the variable
that disturbs CEV-A. ECS-B has no perception of ECS-A or CEV-A. It simply
controls CEV-B. But as a consequence of that control, the disturbance
to CEV-A is reduced, easing the control by ECS-A of its own perception.
Now suppose that CEV-A is part of the variable that disturbs CEV-B. Now
ECS-B has an easier job of control, which directly eases control exercised
by A. Each ECS in this figure-8 loop has a multiplicative effect on the
apparent gain of the other.

Here's a picture:

                    > >
                    > ref-A | ref-B
          ----------O--------- ---------O----------
         > > > >
       PIF-A OUT-A PIF-B OUT-B
         > > > >
  =======^====================V======^===================V=======
         > > > >
          <-----CEV-A<-------- <------CEV-B-------
              > > > >
              V DIST-A<-------------------V DIST-B
              > ^
               ----------------------------------| This loop in the
                                                    environment is unperceived
                                                    by either ECS

In the diagram, DIST-A and DIST-B are supposed to be influenced by, but not
identical to CEV-B and CEV-A respectively. The effect of each control
system is to smooth the disturbance on the other, purely as a side-effect
of its own perceptual control.

If the necessary (and I think not uncommon) environmental conditions hold,
ECS-A and -B become symbiotes. It is advantageous for them to stay together.
Organisms that evolve so that they do will be more stable than those in which
the symbiotes are separated.

The mutuality connection need not, of course, be restricted to pairs of
ECSs. Any situation in the environment in which there is a cycle of
influences among variables that disturb controlled perceptions is enough
to enhance the stability of all affected control loops. But bilateral
mutuality (it's not cooperation) seems most likely to occur in a random
environment.

A consequence of this thought is that mutuality is likely to be quite
common within any evolved control hierarchy. There will be many ECSs
for which a biologist might say "the purpose of this structure is to
assist that structure," where a PCT analyst might initially say that
the second structure is not perceptible to the first, and that the control
exercised by the first is for its own selfish purposes. Not being able
to perceive the second structure or its CEV, the first could not control
for helping it. But evolutionary reorganization will have "controlled"
for the mutuality.

Do I see a model here for social structures, perhaps?

Here today, gone tomorrow.

Martin

[From Rick Marken (931009.1500)]

Martin Taylor (931007 19:30)

:A single ECS (A) controls its perception in the presence of disturbances to
its CEV (CEV-A). It can neither predict nor directly observe these
disturbances, except as fluctuations in the value of CEV-A.

A single ECS can neither predict nor directly observe these disturbances
AT ALL -- certainly not as fluctuations in the value of CEV. The fluctuations
in the value of the CEV depend, at every instant, on the combined effects of
disturbance and output; in the simplest case the value of the CEV at any
instant is proportional to the SUM of the instantaneous values of
disturbance and output -- CEV = d+o. So disturbances cannot be observed
in the fluctuations of the CEV because the disturbance caused component
of these fluctuations is completely "masked" by the the output caused
component. When you adjust the temperature of the water going into the
bath you cannot tell how much of the current temperature of the water is
caused by the prevailing temperature of the hot and cold water in the pipes
(the disturbance) and how much is caused by your outputs (the amount by
which you turn the hot and cold water faucets). All you perceive are the
fluctuations in the CEV (water temperature).

Fluctuations in the CEV (the input to the control system) are no more
important to the process of control than fluctuations in the output. The
fluctuations in the CEV are just as responsible for the fluctuations in
output as the fluctuations in output are responsible for the fluctuations
of input. Those (like me) who have been trained to look at behavior in
input-output terms tend to focus on inputs as the "start" of the control
process; so it is natural to think of fluctuations in the CEV (the input to
the control process) as the start of the control process, followed by comp-
arison to a reference and, finally, generation of output. But, in fact, the
control process has no start and no end -- it is a wheel of causation. What
results from the operation of this causal wheel is control of one variable;
perception of the CEV is kept NEARLY equal to another variable, the reference
variable. The CEV is more like a dependent than an independent variable (where
the term "independent variable" is typically used in the behavioral sciences to
refer to the variable that "starts" a particular causal process). The fluctua-
tions in the CEV, then, are not potential observations regarding the effects
of disturbances (because these fluctuations are the result of BOTH disturbances
and system outputs); rather they represent uncontrolled variations in the CEV.
Unless the disturbances are overwhelming to the control system (as when you
run out of hot water, so there is no amount of faucet turning that will produce
warm water) the fluctuations in the CEV (even when they are very small) are
the result of BOTH disturbance and output.

All of the above is said, not just because I think it bears frequent repetition
(since it is the basis of what makes PCT revolutionary for psychology) but
also because I suspect that it is the basis for Martin's example of
"mutuality":

Consider a second ECS (B), which has as its CEV-B some part of the variable
that disturbs CEV-A. ECS-B has no perception of ECS-A or CEV-A. It simply
controls CEV-B. But as a consequence of that control, the disturbance
to CEV-A is reduced, easing the control by ECS-A of its own perception.
Now suppose that CEV-A is part of the variable that disturbs CEV-B. Now
ECS-B has an easier job of control, which directly eases control exercised
by A. Each ECS in this figure-8 loop has a multiplicative effect on the
apparent gain of the other.

In fact, this situation (as described and shown in Martin's diagram) is
virtually identical to the interactive control situations analyzed by
Tom Bourbon and the coordinated control situations analyzed by yours truly
(and described with icy clarity in "Mind Readings"). Also of relevance
to Martin's example of mutualism is the talk given at the last CSG meeting
on "Helpful control". In fact, the mutualiam shown byMartin's diagram
is does not lead to a "helpful" interaction; the control actions of one
system are resisted (as usual) as disturbances to the controlled variable.
The control exhibited by each system in this interactive arrangement is
actually slightly POORER then the control exhibited either system alone
(I tested this quickly with a spreadsheet model).

There is a situation where I can imagine control systems evolving mutual
dependence -- so that the controlling done by one system depends on the
controlling done by the other system. Tom Bourbon showed a simple
example of interactive (two person control) where the ability of each person
to control a variable depended on the fact that the other person was also
controlling a particular variable. Tom's was a simple situation but, I'm
sure, one that has analogs in real symbiotic relationships between animals.
But the symbiosis (cooperation) does not result from the helpful control
proposed in Martin's diagram ("helpful control" being the situation where the
outputs of one control system are thoutgh to reduce the effects of the
disturbance to the controlled variable of another). There is no such thing
as "helpful control" because it is impossible for any control system to
reliably compensate for the disturbances to the CEVs of another; the only
reliable way to compensate for disturbances (whatever their source) to
CEVs is to monitor the perception of the CEV, continuously compare that
perception to its reference specification and act continuously (in ways that
are not perceptible to the controller itself) to "push" or"pull" that
perception toward the reference.

Best

Rick

[Martin Taylor 931013 12:00]
(Rick Marken 931009.1500)

Rick, I really think you should try to read more carefully before you post.

:A single ECS (A) controls its perception in the presence of disturbances to
its CEV (CEV-A). It can neither predict nor directly observe these
disturbances, except as fluctuations in the value of CEV-A.

A single ECS can neither predict nor directly observe these disturbances
AT ALL

Amazing how you take off over a point I was trying to emphasize as if your
discussion contradicted what I was trying to say. The next two paragraphs
simply reinforce my point.

Consider a second ECS (B), which has as its CEV-B some part of the variable
that disturbs CEV-A. ECS-B has no perception of ECS-A or CEV-A. It simply
controls CEV-B. But as a consequence of that control, the disturbance
to CEV-A is reduced, easing the control by ECS-A of its own perception.
Now suppose that CEV-A is part of the variable that disturbs CEV-B. Now
ECS-B has an easier job of control, which directly eases control exercised
by A. Each ECS in this figure-8 loop has a multiplicative effect on the
apparent gain of the other.

In fact, this situation (as described and shown in Martin's diagram) is
virtually identical to the interactive control situations analyzed by
Tom Bourbon and the coordinated control situations analyzed by yours truly
(and described with icy clarity in "Mind Readings").

Well, you may be right, but it looked (and looks) different to me. I'm
looking at why multicellular organisms exist, and why complementary
cooperative structures have evolved to be so prevalent. There is no
scintilla of "helping" in my diagram, since neither ECS can perceive any
aspect of the CEV that corresponds to the other's controlled perception.
BOTH act quite independently on controlling perceptions according to their
independent reference signals, and it is only a strange vagary of the
environment that lets an outside analyst see that there is some interaction
between the two CEVs and therefore between the two control loops.

In fact, the mutualiam shown byMartin's diagram
is does not lead to a "helpful" interaction; the control actions of one
system are resisted (as usual) as disturbances to the controlled variable.
The control exhibited by each system in this interactive arrangement is
actually slightly POORER then the control exhibited either system alone
(I tested this quickly with a spreadsheet model).

It seems to me that last year you presented a situation in which the opposite
was true.

The result you get will be likely to occur when there is tight coupling
between CEV-A and DIST-B and vice-versa (though not guaranteed). In the
situation I described (though not in the diagram, because of the resulting
confusion of alphanumerically drawn lines), CEV-A influences DIST-B. There
are many other influences on DIST-B, but when CEV-A is well controlled,
then DIST-B will have slower and/or smaller excursions. In the limit of
tight coupling and high-gain control by A, DIST-B will stay almost constant,
and control by B is almost unnecessary, other than to apply a constant
output that opposes the force applied to CEV-B by DIST-B. But more often
than not, the tightly coupled interaction of the two control systems
will lead to some kind of oscillation, particularly if the two control
systems have similar dynamical characteristics.

There is a situation where I can imagine control systems evolving mutual
dependence -- so that the controlling done by one system depends on the
controlling done by the other system.

Oh, yes. There's a lot of that. But I was interested in a situation that
seemed to me not to have been discussed. If it has been, there's no harm
in seeing it in a different light.

There is no such thing
as "helpful control" because it is impossible for any control system to
reliably compensate for the disturbances to the CEVs of another;

This sentence, more than any other in your posting, convinces me that you
missed my point entirely.

Martin