[From Matti Kolu (2013.08.02.2130 CET)]
Mrosovsky's "Rheostasis" was reviewed fairly favorably by Bill Powers
and Tom Bourbon during the early days of CSGnet. The original book has
been out of print for a while. There are of course used (and a few
purportedly new) books available to buy at Amazon and Abebooks, but
there's also an online edition available at:
http://labs.csb.utoronto.ca/mrosovsky/
Those of you who collect control related literature might want to
download the complete book (the last link on the page). According to a
colleague of his, Dr. Mrosovsky has retired, and the URL structures of
web sites such as these have a tendency to change every now and then.
I have pasted the posts of Powers and Bourbon below:
···
----------------------------------------------------------------
Subject: Rheostasis
Date: Sun, 9 Jun 91 12:11:51 -0600
From: POWERS DENISON C <powersd@TRAMP.COLORADO.EDU>
[From Bill Powers (910609.1100)]
Gary Cziko et. al.
I obtained Nicholas Mrosovsky's book *Rheostasis* (New York: Oxford
University Press, 1990) and have been reading it with pleasure. Mrosovsky
has carried his analysis of biological control systems to about the point
where Clark and I were with overt behavioral systems in the mid-1950s,
just as our breakthroughs in understanding were beginning to occur. I
don't say that from a sense of superiority, but out of recognition of a
fellow-explorer working a different terrain and beginning to make sense
of it in the same sequence of steps we went through -- and arriving at
the brink of the same understandings for the same reasons.
If this book accomplishes nothing else, it will be the ideal source for
persuading biologists to give up their abhorrence of the idea of
reference levels. The chapters on Programmed Rheostasis and Reactive
Rheostasis contain innumerable examples of set-points that change. The
examples in this book, in general, will be a rich lode for control
theorists.
What Mrosovsy must learn next is the concept of hierarchical control. At
one point in his book he rejects the notion of "strict hierarchy," but he
construes "hierarchy" in the sense of priorities among controlled
variables at the same level. "Suppose, for example, that water balance
were top in the hierarchy, then a starving animal that had only a modest
water deficit would look for water rather than food. On telological
grounds, strict hierarchy will not be considered further." (p. 21).
I hope this conclusion, which is correct, does not leave Mrosovsky
convinced that the very idea of hierarchy has been dealt with once and
for all so that a more productive concept of it will be rejected without
further consideration. He trembles on the verge of seeing how hierarchy
works in control system assemblies, but isn't quite there yet.
One way he approaches it is through "feed-forward." He correctly defines
feed-forward as reacting to a disturbance rather than correcting error.
But all of his examples involve disturbances of a kind that would need a
higher-order control system in order to be resisted. In many examples,
the response to the external disturbance results in resetting the
reference signal for a lower-level system (as we would see it). Instead
of asking how the organism must be organized in order for this
appropriate "feed-forward" to take place, he treats it simply as a
reaction.
The same half-blind analysis (how well I remember doing exactly the same
thing) is used in the concepts of "programmed" rheostasis and "reactive"
rheostasis. It hasn't occurred to him to see the programming as evidence
of a higher level of control -- even though he recognizes that the
outcome of this programming is a variation in the reference-level of a
lower-level control system, and not simply a behavior (an example he uses
is hibernation, which involves a well-documented change in temperature
set-point and not, as claimed by many others, a failure or loss of gain
in the temperature regulation system). In the idea of "reactive"
rheostasis he comes even closer, for he has the organism reacting to
changes in external conditions by altering set-points.
The war between the old SR viewpoint and the new CT viewpoint is obvious,
or would be to anyone concerned with current behavioral control theory.
On the one hand, Mrosovsky's grasp of the principles of control is very
good, avoiding most of the misconceptions that are rife in the literature
(the one exception being feed-forward, and even there he does not see it
in an unreasonable way). On the other hand, the idea of "causal
variables" is still important in his arguments. Now, however, these
causal variables operate one level up from the level at which the control
systems work: the causal variables cause changes in set-points! This is
the basis of "reactive" rheostasis.
Chapter 6 is about "second-order rheostasis. He says "But then these
superimposed controlling systems themselves become a target for still
further controls, and the possibility arises for second-order rheostasis,
that is for modulation of the way or the rate at which rheostasis is
altering the set-point of the basic regulatory systems." He does not see
the potential of this idea, because the pattern of regulatory output
setting reference levels, and of higher-order controlled variables being
build out of lower-order variables, is simply missing. The examples in
this chapter are ponderous and complicated, and involve too great a gap
between the lowest level (i.e., biochemical regulation) and the highest
(timing of developmental changes in at puberty). There is no conception
of an experimental approach to testing definitions of controlled
variables, a lack that leads Mrosovsky to some improbable conceptions of
relationships among control systems. Nevertheless, Mrosovsky is clearly
on the brink of seeing how hierarchical control works, and his path to
the brink is uncannily like the one I remember taking.
In Chapter 7, Mrosovsky demonstrates his understanding of basic control-
system operation in a boxed section (pp.127-128) where he derives the
relationship between loop gain and the diminution of the effects of
disturbances on a controlled variable. But he fails to arrive at the
basic relationships of control, which Rick Marken just recently expressed
as "p = r and o = -d/k." I think that taking the step from "reactive
rheostasis" to these two relationships is equivalent to taking the final
step out of the world of stimulus and response and into the world of
control theory. If Mrosovsky follows the same timetable that Clark and I
did, he should see the light in about two or three years.
----------------------------------------------------------------
Subject: Re: Rheostasis
Date: Sun, 9 Jun 91 23:31:57 CDT
From: RLPSYU08 <TBOURBON@SFAUSTIN.BITNET>
from Tom Bourbon
Bill Powers [090691] described his reactions to *Rheostasis: The
physiology of change* bu Nicholas Mrosovsky. I have also just
finished reading the book (first mentioned by Gary Cziko, who
learned of it from Elizabeth Satinoff). I share most of the
impressions described by Bill, excluding of course his personal
recollections of having been at a similar point in his thinking
in the 1950s!
Mrosovsky describes many examples of what can be taken as
the resetting of set points in biological control systems. He
also presents alternative explanations that might obviate the
need to invoke set points for some of the phenomena.
Like Bill, I was impressed by the boxed material on pages
127-129, in which Mrosovsky correctly derived the loop gain of
a control system and showed correctly the proportion by which
a given loop gain would reduce the effects of a given dirturbance
acting on a controlled variable. On that count, there is no problem.
But the derivation is never cited or discussed in the text. It
is as though it were an afterthought, or something that seemed
important to include, but that didn't really fit in the
flow of ideas in the book.
The only other place where a control system is disgrammed
or discussed is very early in the book, Again, the details of
a control system and how it functions are not discussed that
often in the text. As Bill remarked, Mrosovsky seems "just
on the brink" of putting it together.
In the early pages of the book, Mrosovsky discusses the
classical idea of homeostasis and describes the popular
invocation of negative feedback as the means by which
homeostasis is achieved. I am not certain how Mrosovsky conceives
of negative feedback, because it is several pages later that he
first introduces the idea of a set point, as a way of overcoming
problems associated with the appeal to negative feedback. It
could be that some of the authors he is addressing used the
idea of negative feedback as a verbalism, like many of the
uses reviewed by George Richardson in his book, *Feedback
thought in social science and systems theory*. Many people
write of negative feedback in a very general way, with no
firm idea of a mechanism that can achieve control via negative
feedback.
I think that is also why Mrosovsky so quickly calls in
the ideas of "positive feedback" and of "feedforward." For
a person who hasn't come across the idea of control by a
hierarchical negative feedback system, those ideas seem
almost essential, but the phenomena they are intended to explain
can be handled effectively by hierarchical perceptual control
systems.
An interesting book, filled with good examples -- some of
them downright fascinating (in the chapter on "Resolving
conflicts" between set points there is a description of a
study in which rats were housed in a warm "home," and could
eat all they wanted in a "restaurant" that was "open" only
two hours a day. The home was warm, the restaurant VERY cold.
Read the book (pages 16, 17) to learn how cleverly the rats
worked their way around those conditions!
Tom Bourbon <TBourbon@SFAustin.BitNet>
Dept. of Psychology
Stephen F. Austin State Univ.
-----------------------------------
Matti