Comments on Bolles

CSGnet Archive CSG1995
1

[From Bruce Abbott (950915.2125 EST)]

My thanks to (in order of receipt) Avery Andrews, Rick Marken, and Bill
Powers for their comments on the Bolles quote on regulation I posted
yesterday (950914.1840 EST). One thing I've been trying to do lately is
find out what happened to the intense interest in regulatory (control)
models involved in feeding, drinking, and other activities which seemed
aimed at controlling some physiological quantity or quantities such as
cellular hydration, blood glucose level, or body weight, which was in
evidence in the late 70's. After a brief "heyday," this approach seems to
have withered under intense criticism levied against it, such as Bolles
expresses. I thought posting Bolles's critique would provide an opportunity
to deal publicly with the issues Bolles (and others) have raised. In so
doing I am neither espousing nor defending the position Bolles takes.
However, given that this kind of reasoning seems to have led to the
dismissal of control theory as an adequate model of physiologically-based
motivation (hunger, thirst, etc.), it would seem essential that one be able
to answer these criticisms when they are raised.

One of the difficulties Bolles (and others) apparently had with applying
control theory seems to be that they expected the elements of the model to
be represented somewhere in the body literally as diagrammed. If the block
diagram shows a comparator, then one must be able to find the comparator,
the mechanism in the body where the reference signal is physically
subtracted from the perceptual signal to produce the error signal. If the
diagram of the standard control system shows a proportional compensator able
to oppose errors from both above and below, then the organism had better
contain such a mechanism. If there is a set point in the system, there must
be a physical signal, being generated by something, which represents the
desired state of the controlled variable. When these physical entities were
not discovered, this evidence was taken to repudiate the control-system model.

Bolles is not the only writer to complain about the problem. After
describing temperature regulation by the gound squirrel (which forages in
the desert sun until it gets hot and then dives into its burrow and lays
flat on the cold floor until it looses the extra heat) and the bee (which
regulates the hive temperature through a set of fairly complex interactions
among individual bees), J. D. Davis in the same volume says ". . .
although one might want to posit the existence of a set-point somewhere in
the brain of the ground squirrel it is very difficult to imagine what a
set-point for bee hive temperature would be like, where it would be located,
or how it would work." D. A. Booth, after outlining his marvelously
detailed Mark 3 computer model of feeding, has the following to say under
the subtitle "Regulation - a Pernicious Concept":

    There is the seductive convenience of comparison between thermostats
    which everyone knows about and homeostats which we hope to learn about.
    In control theory, the value of the set-point is not identical to the
    value of the controlled variable when the system has steadied, but it
    can be made similar; thus it is easy to skip the distinction and read
    the value acheived by homeostasis into the setting of the homeostat.
    Most serious of all, the set-point formalism can be interpreted as a
    physical mechanism and then what may be equilibration between body,
    environment and brain is attributed to a mechanistic marvel within the
    brain. Some have claimed to resist this temptation and to intend their
    invocation of regulatory set-points merely as description of the system's
    performance characteristics. However, as equilibrium or unreferenced
    feedback control on the one hand and regulation on the other hand are
    intertranslatable formalisms for describing performance, the only point
    in invoking set-point regulation would appear to be the implication that
    a physical monitor-comparitor mechanism exists which explians the
    observed stabilising achievements. However, such inference is
    fallacious. The postulate is arbitrary unless and until independent
    evidence for a physical comparator mechanism is obtained by anatomical
    observation and physiological or biochemical measurement at the cellular
    level in the relevant location(s).

    Thus, the result of this definition of regulation (even though not its
    logical requirement) is that stabilities and their defense that can be
    generated by equilibrium processes which are abundantly evidenced are
    being "explained" by physical mechanisms of much greater complexity than
    need to be invoked and are of a type for which there remains no direct
    physiological evidence in the whole of biology (not even for
    thermoregulation, or for metabolic biochemistry or genetic expression).
    The genes, with or without environmental aid, would have to specify a
    cellular arrangement to produce an extremely precise and unvarying signal
    of a particular strength throughout an animal's life; this would be the
    set-point. A monitor has to produce a graded signal of comparable
    accuracy and a comparator has to subtract signals and emit another graded
    signal which is converted into physiological or behavioral activity.
    Mechanical and electronic engineers find such material systems easy to
    construct but living matter does not have the properties that they exploit.

Whether this argument is well framed or not, it appears to have been
excepted among the majority of researchers in this area. I am reminded of
our own little e. coli nutrient-control model in comparison to Koshland's
description of the physical arrangement actually found. In real e. coli,
apparently, the rate of change of nutrient concentration as the organism
swims about simply alters the rate at which tumbles are initiated via some
chemical mediator. All that is required is that a direct relationship be
established between rate of nutrient concentration increase and time to next
tumble. No reference signal, no comparator, no error signal. Just greater
rates of increase lead to longer intervals between tumbles. Our
control-system model of e. coli, however, contains an explicit reference
signal, comparator, and error signal. Anyone looking into e. coli's system
for the physical representations of these entities is not going to find them
there; logically, the system merely acts AS IF they are. Perhaps that is
the important thing.

Regards,

Bruce

2

[From Bill Powers (950915.2140 MDT)]

Bruce Abbott (950915.2125 EST) --

Bruce, this is great work you're doing. All these things you're citing
are exactly the reasons for which PCT has not been accepted.

     J. D. Davis in the same volume says ". . . although one might want
     to posit the existence of a set-point somewhere in the brain of the
     ground squirrel it is very difficult to imagine what a set-point
     for bee hive temperature would be like, where it would be located,
     or how it would work."

This manner of speaking uses words that say "I have difficulty
understanding how ..." but which really mean "It is ridiculous to
propose ...". Clearly, the idea of controlling a _perception_ of beehive
temperature would solve Davis's problem, because that would tell us
where the set point is located (inside the bee), and a little education
in control theory would explain how it would work. But many people have
taken Davis' attitude of "If I don't understand it, there can't be
anything to it."

Booth has a similar problem:

         There is the seductive convenience of comparison between
     thermostats which everyone knows about and homeostats which we hope
     to learn about. In control theory, the value of the set-point is
     not identical to the value of the controlled variable when the
     system has steadied, but it can be made similar; thus it is easy to
     skip the distinction and read the value acheived by homeostasis
     into the setting of the homeostat.

This word "similar" implies "mere" similarity, hiding the fact that the
actual temperature can be maintained within a fraction of a degree of
the set point. If he had said "almost identical" the whole meaning of
the sentence would have been changed. But that isn't Booth's only
problem. He says

     The genes, with or without environmental aid, would have to specify
     a cellular arrangement to produce an extremely precise and
     unvarying signal of a particular strength throughout an animal's
     life; this would be the set-point.

The term "extremely precise and unvarying signal of a given strength"
reveals a misconception, because there is no "set point" that is
invariant for the life of the organism or even for one whole day. A good
antidote for this misconception is Myrsovski's _Homeorhesis_, in which
he shows that (a) the set points of all known biological regulatory
systems are variable, and (b) these set points can be proven to change
by showing that the system defends different levels of the variable
against disturbances at different times.

Perhaps someone influential in biology once misinterpreted a "set" point
to mean a point that was fixed forever in some optimum state. It's that
sort of pompous ignorance that I have been fighting for 40 years.

I do get hot under the collar at statements like this:

       A monitor has to produce a graded signal of comparable
     accuracy and a comparator has to subtract signals and emit another
     graded signal which is converted into physiological or behavioral
     activity. Mechanical and electronic engineers find such material
     systems easy to construct but living matter does not have the
     properties that they exploit.

This is merely a pretense at expertise on a subject of which the author
knows nothing. One of the main reasons for which control systems are
used in engineering is that they do NOT require precise or stable
components to achieve precise and stable results. One of the main
reasons for which control theory is an ideal theory of real human
systems is that it describes an organization that can work with the
sorts of sloppy components found in living systems, yet achieve quite
accurate results. It is the theories that do NOT use control principles
that run into problems of precision and stability.

What elevates the temperature of my neck protector is these authors'
totally unwarranted claim to understanding. The anti-control-theory
literature in the behavioral sciences is full of this sort of thing:
announcements of how control systems work and what they can or can't do
that are based on a few minutes of reasoning and an abysmal lack of
actual knowledge. This is how all the myths about control theory have
got started; some amateur thinks that he's got it all figured out, but
never having studied the subject he makes all the elementary mistakes
that any sophomore undergraduate engineering student gets cured of in
the first six weeks of an introductory course. Unfortunately, the people
who make these mistakes have big names in their fields, so those who
read their words naturally assume they know what they're talking about.
So the myths just spread and spread.

     I am reminded of our own little e. coli nutrient-control model in
     comparison to Koshland's description of the physical arrangement
     actually found. In real e. coli, apparently, the rate of change of
     nutrient concentration as the organism swims about simply alters
     the rate at which tumbles are initiated via some chemical mediator.
     All that is required is that a direct relationship be established
     between rate of nutrient concentration increase and time to next
     tumble. No reference signal, no comparator, no error signal. Just
     greater rates of increase lead to longer intervals between tumbles.

Actually, if you read Koshland carefully, you'll see that there is
evidence for variable reference signals. There are "mutants" which
always tumble, and others that never tumble. A sufficiently rapid change
in the concentration, however, can make the always-tumblers stop
tumbling or (in the other direction) the never-tumblers start tumbling.
Among normal bacteria, there are different biases on the rate of change
of concentration that will begin to suppress tumbling, and of course any
such bias is evidence of a reference signal, and by implication some
comparison mechanism. The operational definition of a reference level,
which is quite independent of the block diagram by which you choose to
model the system, is "that level of input at which the output just
becomes zero." There is certainly such a level of rate of change of
concentration that can be determined for any individual bacterium. And
any way you choose to explain this effect is exactly equivalent,
mathematically, to the standard PCT diagram.

      Our control-system model of e. coli, however, contains an explicit
     reference signal, comparator, and error signal. Anyone looking
     into e. coli's system for the physical representations of these
     entities is not going to find them there; logically, the system
     merely acts AS IF they are. Perhaps that is the important thing.

The important thing is not the particular block diagram, but the
relationships observed among the variables. What we see is that

Output = k * (RefLevel - Input),

Where RefLevel is simply that magnitude of input at which Output = 0. If
the Output becomes zero as Input approaches 10 units, then Reflevel is
10 and the equation is

Output = k * (10 - Input).

Then by observing how the Output changes as Input departs from the value
of 10 units, you can determine the value of k. This is all purely
empirical, no theory involved.

There is absolutely nothing about this analysis that depends on the
particular way you propose that the nervous system creates this
relationship. We use the simple PCT diagram because it shows all the
basic operations needed to create such a relationship, and its parts
correspond to known components of some actual physiological systems. The
Input Function converts the physical variable outside the system into an
internal signal; the comparator implements the subtraction; the output
function implements the amplification and conversion of the difference
into a corresponding amount of output. Any two adjacent operations or
all three of them could be combined in a single physical element of the
system. A single neuron can function as a complete control system.

Koshland was actually quite interested in the 'signaling system', which
is the Input Function part of the standard model. Once you have split
out the input processes, you are left only with the comparison and
output processes to explain, and you can model them separately or as a
combined function -- until and unless further research enables you to
show that these two functions are carried out by different means in
different places.

This equation and the definition of RefLevel do not depend on the block
diagram. The PCT diagram is simply one straightforward way of
implementing the equation to produce the same input-output relationship.
All of the principles of control can be expressed entirely in terms of
observable variables, without ever trying to guess how the nervous
system is internally organized. But we -- or at least I -- want to try
to understand how the nervous system is organized, which is why I keep
looking for physical evidence about the actual wiring, and keep
proposing possible block diagrams.

Also, a block diagram communicates relationships which are hard for
people who are not comfortable with simultaneous equations to
understand. The other equation in the simultaneous pair is, of course,
the equation that describes how the Input depends on the Output via the
environment. The diagram of a control loop makes the solution of the
simultaneous equations obvious, or at least more obvious.

···

-----------------------------------------------------------------------
Best to all,

Bill P.