In my earlier message on defining feedback and feedforward, I tried to
say that people who take an endogenous point of view and think in
circular causal terms dont have much reason to think in "back" and
"forward" terms. System dynamicists happily talk about "feedback loops"
and use the term to stand for all manner of circular causal system
structures and processes. But we dont have a meaning for "feedback" by
itself, or "feedforward" by itself.
One might be able to identify certain links in particular circular causal
structures that people in various disciplines would recognize or label as
"feedback" or "feedforward" links. But it seems to me if the picture is
a loop of causal influences then what is "back" and what is "forward"
depends on where you start or stand in the loop.
So to Ed Id say that all three of his examples are examples of feedback
loops. He has identified links in the first two examples that people with
certain backgrounds would be comfortable labeling "feedback" and
"feedforward." But the system dynamicist needs a label for the loop in
all three, since all three involve vivid circular causal structure. So
the system dynamicist says they are all "feedback loops," and except for
making good interpersonal connections does not particularly care if
someone else wants to name certain links with "feedback" and
"feedforward." I dont think we have a "back" and "forward" -- we have
loops.
Does the terminology matter? Yes, if the people you are trying to talk
with have these terms but do not have our meanings. It would be bad if
people with a "feedback/feedforward" lexicon were to decide that the loop
in Newtons law of cooling is "neither feedback or feedforward," as Ed
offered, since that is perilously close to saying there is not "a
feedback loop" there. [If there isnt feedback and there isnt
feedforward, then how can we have a feedback loop, one might ask.]
And it would be similarly sad if we were not able to communicate well
with people who want to use "feedback" and "feedforward." We just must
understand that from the general (endogenous) point of view of circular
causal structures, our deep and persistent focus is on loops, which we
call "feedback" loops.
I understand that the "feedforward" terminology has an important use in
behavioral decision theory, where it signals (I think) the use of a model or
tool to project into the future as an aid to making a decision that will
influence that future. But that seems to be a very different use from
the engineers notion of feeding the error signal in a control process
ahead in a plant/controller sequence to improve the controllers
capabilities (thats Eds example of skin sensors and body temperature).
Estragon: Well, THAT passed the time...
Vladimir: It would have passed anyway.
...[Beckett, Waiting for Godot]
...Geo, waiting for more insight
-----------------------------------------------------------------------------
George P. Richardson G.P.Richardson@Albany.edu
Rockefeller College of Public Affairs and Policy Phone: 518-442-3859
University at Albany - SUNY, Albany, NY 12222 Fax: 518-442-3398
-----------------------------------------------------------------------------
Feedback -forward?
-
George Richardson
- Senior Member
- Posts: 68
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
I trust it will be clear to most who are reading these messages about
feedback/feedforward links and feedback loops (and I hope to Mr. Budiman
upon reflection) that I would NOT agree with his recent message, in which
he said Stermans exigesis of the coffee cooling negative feedback loop
incorrect, that the structure was "not feedback." His quote of what I
wrote looks to me to support Sterman rather than contradict, and I would
mean it to support Stermans view entirely. (See, e.g., Richardson & Pugh,
pp. 69ff where coffee cooling is used explicitly to illustrate the
simulation of a feedback system.)
This discussion of what deserves to be called "feedback" is really about
45 years old. Ashby and Von Bertalanffy argued it in almost exactly the
same way weve seen in this list in the last few messages on the subject.
(If you want to trace the details see pp. 113-123 of Feedback Thought.)
We can chose to keep the semantic conversation going or we can agree on
the usage we most like and then charge on to use it to solve real
problems.
We need a term for the circularities that are so crucial to an endogenous
view of a systems dynamics. Forrester wrote "Systems of information
feedback control are fundamental to all life and human endeavor, from the
slow pace of biological evolution to the launching of the latest satellit.
A feedback control systems exists whenever the environment causes a
decision which in turn affects the original environment" (Forrester 1958,
p. 4). Before I knew anything about differing usages, I wrote "A feedback
loop is a closed sequence of causes and effects, a closed path of action
and information" (Richardson & Pugh, p. 4). Ashby wrote in the same vein:
"When this circularity of action exists between the parts of a dynamic
system, feedback may be said to be present" (Ashby, Introduction to
Cybernetics, p. 53). And in Feedback Thought, where I first encountered
and uncovered the controversy in meanings, I wrote "We also have a
seriously argued point of view that would take offense at the broad use of
the word feedback in this investigation. Our interests, as I have said,
is the loop concept underlying feedback and mutual causality. I have
chosen to use the words "feedback" and "feedback loop" to stand for this
longer but undoubtedly more accurate phrase."
So lets agree, as our literature has for 40 years, that we will use the
term "feedback" to signal circular causality, loops of action and
information. We can all identify feedback loops, as any and all closed
causal loops of action and information.
Lets also agree that some may want to identify, for their own purposes,
a given link as a "feedback link" and to distinguish that kind of link
from some other kinds which they would say are not "feedback links." But
lets insist that if a causal loop is formed, we will all call it a
"feedback loop," not matter how its links are labeled, or by whom.
Similarly, lets agree that some may find a good use for the term
"feedforward link." For both these link terms, however, we would require
much stricter definitions than weve seen so far, and many of us would
probably want to see strong reasons for paying attention to the
distinctions in the dynamic policy systems we are addressing.
And let us again focus on what is most important here to ourselves and to
the larger world which for the most part does not share our perspective:
we are taking an endogenous view (and internal view, a "system as cause"
view) of system dynamics, and that forces us to see causality as
circular. It is the loop view that is crucial, not the name we attach to
any particular link.
In his 1948 control theory text, Gordon Brown (Jays mentor, and so in
some sense ours as well) wrote: "A closed-loop control system is thus an
error-sensitive system and, being such, it acquires certain pecularities
and idiosyncrasis which, in large measure, are the reasons for this book."
And for our entire field, Im sure he would be happy to add now.
...GPR
-----------------------------------------------------------------------------
George P. Richardson G.P.Richardson@Albany.edu
Rockefeller College of Public Affairs and Policy Phone: 518-442-3859
University at Albany - SUNY, Albany, NY 12222 Fax: 518-442-3398
-----------------------------------------------------------------------------
feedback/feedforward links and feedback loops (and I hope to Mr. Budiman
upon reflection) that I would NOT agree with his recent message, in which
he said Stermans exigesis of the coffee cooling negative feedback loop
incorrect, that the structure was "not feedback." His quote of what I
wrote looks to me to support Sterman rather than contradict, and I would
mean it to support Stermans view entirely. (See, e.g., Richardson & Pugh,
pp. 69ff where coffee cooling is used explicitly to illustrate the
simulation of a feedback system.)
This discussion of what deserves to be called "feedback" is really about
45 years old. Ashby and Von Bertalanffy argued it in almost exactly the
same way weve seen in this list in the last few messages on the subject.
(If you want to trace the details see pp. 113-123 of Feedback Thought.)
We can chose to keep the semantic conversation going or we can agree on
the usage we most like and then charge on to use it to solve real
problems.
We need a term for the circularities that are so crucial to an endogenous
view of a systems dynamics. Forrester wrote "Systems of information
feedback control are fundamental to all life and human endeavor, from the
slow pace of biological evolution to the launching of the latest satellit.
A feedback control systems exists whenever the environment causes a
decision which in turn affects the original environment" (Forrester 1958,
p. 4). Before I knew anything about differing usages, I wrote "A feedback
loop is a closed sequence of causes and effects, a closed path of action
and information" (Richardson & Pugh, p. 4). Ashby wrote in the same vein:
"When this circularity of action exists between the parts of a dynamic
system, feedback may be said to be present" (Ashby, Introduction to
Cybernetics, p. 53). And in Feedback Thought, where I first encountered
and uncovered the controversy in meanings, I wrote "We also have a
seriously argued point of view that would take offense at the broad use of
the word feedback in this investigation. Our interests, as I have said,
is the loop concept underlying feedback and mutual causality. I have
chosen to use the words "feedback" and "feedback loop" to stand for this
longer but undoubtedly more accurate phrase."
So lets agree, as our literature has for 40 years, that we will use the
term "feedback" to signal circular causality, loops of action and
information. We can all identify feedback loops, as any and all closed
causal loops of action and information.
Lets also agree that some may want to identify, for their own purposes,
a given link as a "feedback link" and to distinguish that kind of link
from some other kinds which they would say are not "feedback links." But
lets insist that if a causal loop is formed, we will all call it a
"feedback loop," not matter how its links are labeled, or by whom.
Similarly, lets agree that some may find a good use for the term
"feedforward link." For both these link terms, however, we would require
much stricter definitions than weve seen so far, and many of us would
probably want to see strong reasons for paying attention to the
distinctions in the dynamic policy systems we are addressing.
And let us again focus on what is most important here to ourselves and to
the larger world which for the most part does not share our perspective:
we are taking an endogenous view (and internal view, a "system as cause"
view) of system dynamics, and that forces us to see causality as
circular. It is the loop view that is crucial, not the name we attach to
any particular link.
In his 1948 control theory text, Gordon Brown (Jays mentor, and so in
some sense ours as well) wrote: "A closed-loop control system is thus an
error-sensitive system and, being such, it acquires certain pecularities
and idiosyncrasis which, in large measure, are the reasons for this book."
And for our entire field, Im sure he would be happy to add now.
...GPR
-----------------------------------------------------------------------------
George P. Richardson G.P.Richardson@Albany.edu
Rockefeller College of Public Affairs and Policy Phone: 518-442-3859
University at Albany - SUNY, Albany, NY 12222 Fax: 518-442-3398
-----------------------------------------------------------------------------
-
George Richardson
- Senior Member
- Posts: 68
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
Good stuff.
I avoid trying to label a given link as feedback or feedforward or any
specific kind of "feed" because I cant do it (with any reliability) in all
instances. I can identify some links that I know you (and von
Bertalanffy and others) would want to label feedback, such as the error
signal that a controller receives (as you said). But I would have
trouble with, say, the link from population density to births per person
per year (fecundity, family size, ...) because it has a "policy"
character in which it represents a response to an error signal (actual
pop density vs desired pop density) but it also has a kind of "naturally
occurring" structural character. Its often a hard distinction to make.
I also avoid the labeling of links because I dont think it helps anybody
focus on what is important: the loop that the links form. I want a name
for the loop ("feedback loop" works well) and I want to help people think
in loops. The distinction between "naturally occuring structure" and
"policy structure" is often very useful, but I dont think labeling one
as "feedback" helps people see what needs to be seen in the distinction.
So you see I see the labeling of links as "feedback" or "not feedback" is
(a) kind of hard and (b) not very useful, so I dont do it. And my
messages to the list were to encourage us to focus ourselves and our
various audiences on what is so crucial in what we do and say: the
endogenous point of view and the loops of circular causality that give
dynamic systems, as Gordon Brown said, their "certain peculiarities and
idiosyncrasies." Im convinced that any time spent drawing people away from
that great focus hurts more than it helps.
...Geo
-----------------------------------------------------------------------------
George P. Richardson G.P.Richardson@Albany.edu
Rockefeller College of Public Affairs and Policy Phone: 518-442-3859
University at Albany - SUNY, Albany, NY 12222 Fax: 518-442-3398
-----------------------------------------------------------------------------
I avoid trying to label a given link as feedback or feedforward or any
specific kind of "feed" because I cant do it (with any reliability) in all
instances. I can identify some links that I know you (and von
Bertalanffy and others) would want to label feedback, such as the error
signal that a controller receives (as you said). But I would have
trouble with, say, the link from population density to births per person
per year (fecundity, family size, ...) because it has a "policy"
character in which it represents a response to an error signal (actual
pop density vs desired pop density) but it also has a kind of "naturally
occurring" structural character. Its often a hard distinction to make.
I also avoid the labeling of links because I dont think it helps anybody
focus on what is important: the loop that the links form. I want a name
for the loop ("feedback loop" works well) and I want to help people think
in loops. The distinction between "naturally occuring structure" and
"policy structure" is often very useful, but I dont think labeling one
as "feedback" helps people see what needs to be seen in the distinction.
So you see I see the labeling of links as "feedback" or "not feedback" is
(a) kind of hard and (b) not very useful, so I dont do it. And my
messages to the list were to encourage us to focus ourselves and our
various audiences on what is so crucial in what we do and say: the
endogenous point of view and the loops of circular causality that give
dynamic systems, as Gordon Brown said, their "certain peculiarities and
idiosyncrasies." Im convinced that any time spent drawing people away from
that great focus hurts more than it helps.
...Geo
-----------------------------------------------------------------------------
George P. Richardson G.P.Richardson@Albany.edu
Rockefeller College of Public Affairs and Policy Phone: 518-442-3859
University at Albany - SUNY, Albany, NY 12222 Fax: 518-442-3398
-----------------------------------------------------------------------------
-
Primanata Tjandra
- Newbie
- Posts: 1
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
IRT SD0266
Hi,
I am not an expert in both System Dynamics and public policy, but here is
my attempt to respond to Benny. I might be just paraphrasing comments
that have been made, but I feel that the only way I truly learn is
by making explicit my assumptions and beliefs. Otherwise, I could be
clinging to a fallacy without me realizing it.
Benny,
You proposed that there should be a distinction of feedbacks as a result
of policy design and implementation from inherent causal links and loop of
actions, and you claimed that this distinction would be helpful in
identifying the lever points of the system.
First of all, I noticed that in having this mindset, you had actually
failed to see that the cup of coffee example is indeed an instance of a
negative feedback loop. It is not a requisite of a feedback loop to have
a conscious designer as part of the system to compare the output to the
desired level. Both positive and negative feedback loops are abound in
naturally or artificially designed/occuring systems.
Also, by the above differentiation, you are implicitly saying that there
should be a prioritization in policy design: change the level of the
variables first before attempting to change the structure of the system.
To me, this hard and fast prioritization would very well lead to tunnel
vision, and not seeing the forest. Justification for changing the value
of the variables against changing the structure should be done at a
contextual level.
I agree that we cannot change the embedded physics, but it would be
mistaken to assume that whatever occurs naturally is not changeable or is
fixed in nature. There is also a problem with the distincition of what is
intrinsic or extrinsic when we go beyond the more bounded nature of
engineering problems. For example, is Confucianist thinking and belief
that governs the Asian social system intrinsic or externally imposed? Is
it fixed until redesigned or is it dynamic? I would think that the rise
and fall of Confucianism would alter the structure of the social system,
whether it is a result of an inherent flow of trend from the West, or if
it is a result of an explicit policy imposed by the prevailing government.
In your example of the controller and the plant, the subsystems are
clearly defined. In other complex systems, the prescence of the observer
might alter the nature of the system itself. For example, in engineering
design studies, it is recently found that the requirement (by the
researcher) that participants of the protocol analysis to vocalize their
design intent has a more profound effect on the design process itself than
previously thought.
For the same reason, I dont think the "spaghetti" criticism of System
Dynamics is reasonable. In fact, I think that people are prone to be
simplistic and get locked into a single governing phenomena that is in
vogue, like the path dependence phenomena that has been discussed
previously.
Understanding the interplay of the various system archetypes embodied by
the system, and not the distinction of what is naturally or artificially
occuring information flow, is the key to enlightenment, one that would
result in better policy design, which is an interplay of finding pressure
points of the current system AND re-structuring, not one OR the other.
Prima
Primanata Tjandra
prima@u.washington.edu
Hi,
I am not an expert in both System Dynamics and public policy, but here is
my attempt to respond to Benny. I might be just paraphrasing comments
that have been made, but I feel that the only way I truly learn is
by making explicit my assumptions and beliefs. Otherwise, I could be
clinging to a fallacy without me realizing it.
Benny,
You proposed that there should be a distinction of feedbacks as a result
of policy design and implementation from inherent causal links and loop of
actions, and you claimed that this distinction would be helpful in
identifying the lever points of the system.
First of all, I noticed that in having this mindset, you had actually
failed to see that the cup of coffee example is indeed an instance of a
negative feedback loop. It is not a requisite of a feedback loop to have
a conscious designer as part of the system to compare the output to the
desired level. Both positive and negative feedback loops are abound in
naturally or artificially designed/occuring systems.
Also, by the above differentiation, you are implicitly saying that there
should be a prioritization in policy design: change the level of the
variables first before attempting to change the structure of the system.
To me, this hard and fast prioritization would very well lead to tunnel
vision, and not seeing the forest. Justification for changing the value
of the variables against changing the structure should be done at a
contextual level.
I agree that we cannot change the embedded physics, but it would be
mistaken to assume that whatever occurs naturally is not changeable or is
fixed in nature. There is also a problem with the distincition of what is
intrinsic or extrinsic when we go beyond the more bounded nature of
engineering problems. For example, is Confucianist thinking and belief
that governs the Asian social system intrinsic or externally imposed? Is
it fixed until redesigned or is it dynamic? I would think that the rise
and fall of Confucianism would alter the structure of the social system,
whether it is a result of an inherent flow of trend from the West, or if
it is a result of an explicit policy imposed by the prevailing government.
In your example of the controller and the plant, the subsystems are
clearly defined. In other complex systems, the prescence of the observer
might alter the nature of the system itself. For example, in engineering
design studies, it is recently found that the requirement (by the
researcher) that participants of the protocol analysis to vocalize their
design intent has a more profound effect on the design process itself than
previously thought.
For the same reason, I dont think the "spaghetti" criticism of System
Dynamics is reasonable. In fact, I think that people are prone to be
simplistic and get locked into a single governing phenomena that is in
vogue, like the path dependence phenomena that has been discussed
previously.
Understanding the interplay of the various system archetypes embodied by
the system, and not the distinction of what is naturally or artificially
occuring information flow, is the key to enlightenment, one that would
result in better policy design, which is an interplay of finding pressure
points of the current system AND re-structuring, not one OR the other.
Prima
Primanata Tjandra
prima@u.washington.edu
-
"William Steinhurst"
- Member
- Posts: 21
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
George Richardson <gr383@cnsvax.albany.edu> wrote:
> We need a term for the circularities that are so crucial to an
> endogenous view of a systems dynamics. Forrester wrote
"<snip> A feedback control systems exists
> whenever the environment causes a decision which in turn affects the
> original environment" (Forrester 1958, p. 4). Before I knew
> anything about differing usages, I wrote "A feedback loop is a
> closed sequence of causes and effects, a closed path of action and
> information" (Richardson & Pugh, p. 4). Ashby wrote in the same
> vein: "When this circularity of action exists between the parts of a
> dynamic system, feedback may be said to be present" (Ashby,
> Introduction to Cybernetics, p. 53).
Far be it from me to engender any further semantic debate on this
matter! However, I detect a possible clash of assumptions and
terminology. Could it be that the terms "information," "action,"
and "decision" are, to some people, so loaded with connotations of
self-awareness that they automatically assume coffee cups (and other
natural systems) can only have information and decision feedback if
we accord them conscious intent? In other words, some, while readily
conceding that a cup of hot coffee as a thermodynamic system is a
dynamical system with a cooling rate dependent on the relevant state
variables we all know and love from Georges book, may also hold that a
cup of hot coffee cannot have "feedback" because theres no
"steersman" (kybernetikos?) to make a "decision" and "act" by pushing
or pulling on a lever.
William Steinhurst
wsteinhu@psd.state.vt.us
> We need a term for the circularities that are so crucial to an
> endogenous view of a systems dynamics. Forrester wrote
"<snip> A feedback control systems exists
> whenever the environment causes a decision which in turn affects the
> original environment" (Forrester 1958, p. 4). Before I knew
> anything about differing usages, I wrote "A feedback loop is a
> closed sequence of causes and effects, a closed path of action and
> information" (Richardson & Pugh, p. 4). Ashby wrote in the same
> vein: "When this circularity of action exists between the parts of a
> dynamic system, feedback may be said to be present" (Ashby,
> Introduction to Cybernetics, p. 53).
Far be it from me to engender any further semantic debate on this
matter! However, I detect a possible clash of assumptions and
terminology. Could it be that the terms "information," "action,"
and "decision" are, to some people, so loaded with connotations of
self-awareness that they automatically assume coffee cups (and other
natural systems) can only have information and decision feedback if
we accord them conscious intent? In other words, some, while readily
conceding that a cup of hot coffee as a thermodynamic system is a
dynamical system with a cooling rate dependent on the relevant state
variables we all know and love from Georges book, may also hold that a
cup of hot coffee cannot have "feedback" because theres no
"steersman" (kybernetikos?) to make a "decision" and "act" by pushing
or pulling on a lever.
William Steinhurst
wsteinhu@psd.state.vt.us
-
"Paul Atkins"
- Junior Member
- Posts: 3
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
WRT SD0260 and SD0262 Feedback and coffee cooling....
Isnt there a question of levels here? Benny writes:
> Temperature is an indicator of energy content, so molecules of a warmer
> body have more (kinetic) energy than those of a cooler one. These
> molecules then "transfer" their energy to their counterparts through
> interactions (collisions) at the boundary. The process then continues
> as the collisions "diffuse" into the originally cooler body.
At the molecular level I agree with Benny. If you include things
bumping into one another as feedback then everything is feedback and the term
loses some of its punch IMO. Does Newtons third law of motion ("... equal and
opposite reaction") count as an example of feedback?
On the other hand, John discusses the system as if it is goal-oriented and his
argument makes perfect sense too. Clearly this is a system that can be
modelled using feedback at the level of the whole system. I just wonder
whether, in this case, anything is gained by looking at it at this level when
the molecular level explains the phenomenon perfectly.
Regards,
Paul
--------------------------------------------------------------------------------
Paul Atkins Email: p.atkins@unsw.edu.au
School of Psychology Phone: +61 2 385 1461
University of New South Wales Fax: +61 2 385 3641
Sydney NSW 2052 Australia
Isnt there a question of levels here? Benny writes:
> Temperature is an indicator of energy content, so molecules of a warmer
> body have more (kinetic) energy than those of a cooler one. These
> molecules then "transfer" their energy to their counterparts through
> interactions (collisions) at the boundary. The process then continues
> as the collisions "diffuse" into the originally cooler body.
At the molecular level I agree with Benny. If you include things
bumping into one another as feedback then everything is feedback and the term
loses some of its punch IMO. Does Newtons third law of motion ("... equal and
opposite reaction") count as an example of feedback?
On the other hand, John discusses the system as if it is goal-oriented and his
argument makes perfect sense too. Clearly this is a system that can be
modelled using feedback at the level of the whole system. I just wonder
whether, in this case, anything is gained by looking at it at this level when
the molecular level explains the phenomenon perfectly.
Regards,
Paul
--------------------------------------------------------------------------------
Paul Atkins Email: p.atkins@unsw.edu.au
School of Psychology Phone: +61 2 385 1461
University of New South Wales Fax: +61 2 385 3641
Sydney NSW 2052 Australia
-
scuba@usa.net (Bob Powell)
- Junior Member
- Posts: 4
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
George Richardson wrote in SD0261:
>But lets insist that if a causal loop is formed, we will all call it a
>"feedback loop," not matter how its links are labeled, or by whom.
>Similarly, lets agree that some may find a good use for the term
>"feedforward link."
In reading the exchange on this topic, it occurred to me ask whether it
would be useful to distinguish between
- the "feedback loop" in which an error signal is generated and used to
influence the future state of the system and
- the "algorithm" used to calculate the error signal?
For instance, in industrial settings PID controllers are used control
furnace temperature. For any who might be unfamiliar with the term, PID,
individual gains (feedback strengths) can be set for each PID signal component.
For An error signal component is generated proportional to the
P - difference between the (sampled) current temp & desired temp,
I - integral of the difference (over the "recent" past), and
D - first derivative of the difference (based on the "recent" past).
The discussion here seems to refer to
- feedback as based on a signal derived based simply on the past and/or
current state of the system and
- feedforward as based on a signal derived using some projection into the
future based on past and current states.
But, if were going to distinguish different types of feedback loops based
on the algorithm used, perhaps weve got *lots* of different types (and
flavors) to create.
Types based the above (many other possible, of course):
P - feedback current based on a current time sample
(a feedback ... could be "feedback_now"?)
I - feedback recent history based on the recent past
(a feedback ... could be "feedback_past"?)
D - feedback projected future base on the recent first derivative
(a "feedforward" ... could be "feedforward_linear"?)
As George suggests, perhaps simply using "feedback loop" will do?
Bob Powell
scuba@usa.net
>But lets insist that if a causal loop is formed, we will all call it a
>"feedback loop," not matter how its links are labeled, or by whom.
>Similarly, lets agree that some may find a good use for the term
>"feedforward link."
In reading the exchange on this topic, it occurred to me ask whether it
would be useful to distinguish between
- the "feedback loop" in which an error signal is generated and used to
influence the future state of the system and
- the "algorithm" used to calculate the error signal?
For instance, in industrial settings PID controllers are used control
furnace temperature. For any who might be unfamiliar with the term, PID,
individual gains (feedback strengths) can be set for each PID signal component.
For An error signal component is generated proportional to the
P - difference between the (sampled) current temp & desired temp,
I - integral of the difference (over the "recent" past), and
D - first derivative of the difference (based on the "recent" past).
The discussion here seems to refer to
- feedback as based on a signal derived based simply on the past and/or
current state of the system and
- feedforward as based on a signal derived using some projection into the
future based on past and current states.
But, if were going to distinguish different types of feedback loops based
on the algorithm used, perhaps weve got *lots* of different types (and
flavors) to create.
Types based the above (many other possible, of course):
P - feedback current based on a current time sample
(a feedback ... could be "feedback_now"?)
I - feedback recent history based on the recent past
(a feedback ... could be "feedback_past"?)
D - feedback projected future base on the recent first derivative
(a "feedforward" ... could be "feedforward_linear"?)
As George suggests, perhaps simply using "feedback loop" will do?
Bob Powell
scuba@usa.net
-
George Richardson
- Senior Member
- Posts: 68
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
Steinhursts observation is the Ashby/Von Bertalanffy debate.
What are we going to do about this? We could:
a) stop calling feedback loops "feedback loops" and agree to use
something like "circular causal loops";
b) continue calling all circular causal loops "feedback loops" and put
disclaimers or asterisks on them to make sure all know that some links
are deliberate "feedings back" while others are somehow more "natural";
c) continue calling all circular causal loops "feedback loops" and not
worry about it;
d) some other option -- Im sure there must be other ways out of this.
My understanding is that in the past the system dynamics literature and
some other related literatures have done (c). For what its worth, given
what I know now thats what Id vote for.
...Geo
-----------------------------------------------------------------------------
George P. Richardson G.P.Richardson@Albany.edu
Rockefeller College of Public Affairs and Policy Phone: 518-442-3859
University at Albany - SUNY, Albany, NY 12222 Fax: 518-442-3398
-----------------------------------------------------------------------------
What are we going to do about this? We could:
a) stop calling feedback loops "feedback loops" and agree to use
something like "circular causal loops";
b) continue calling all circular causal loops "feedback loops" and put
disclaimers or asterisks on them to make sure all know that some links
are deliberate "feedings back" while others are somehow more "natural";
c) continue calling all circular causal loops "feedback loops" and not
worry about it;
d) some other option -- Im sure there must be other ways out of this.
My understanding is that in the past the system dynamics literature and
some other related literatures have done (c). For what its worth, given
what I know now thats what Id vote for.
...Geo
-----------------------------------------------------------------------------
George P. Richardson G.P.Richardson@Albany.edu
Rockefeller College of Public Affairs and Policy Phone: 518-442-3859
University at Albany - SUNY, Albany, NY 12222 Fax: 518-442-3398
-----------------------------------------------------------------------------
-
"John S. Edwards"
- Junior Member
- Posts: 2
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
I think there is (still) a different issue raised by Benny Budiman,
apart from the question of "intent" that William Steinhurst and
George Richardsons recent postings have covered so well.
I will deliberately not use "SD-language" in what follows, as I do
not think I know enough about SD to deserve a vote as to whether
"feedback loops", "circular causal loops" or something else is the
best term! I hope that I can clarify the issue, though.
It is that in the "coffee cup cooling" example, the standard
treatment assumes that the room temperature is fixed, thus serving as
a "target" value for the temperature of the coffee. This is a good
enough assumption for all practical purposes, BUT, all other things
being equal, the room temperature will actually rise by a negligible
amount as the coffee cools. To illustrate better, suppose we had not
a coffee cup, but a 120-gallon fish tank full of water at say 100
degrees Fahrenheit in the room. In cooling down, that certainly
would raise the room temperature (I know this, as we keep fish!).
Thus the fish tank water temperature is aiming at a moving target,
not the effectively fixed one of the coffee cup.
The issue, on which as I said Im not fit to pronounce, is whether
this sort of equilibrium finding process is different enough from
more conventional feedback processes (like a room thermostat) that a
different name is required to cover both.
John E
j.s.edwards@aston.ac.uk
(Dr.) John S. Edwards
Operations & Information Management Group
Aston Business School
Aston University
Aston Triangle
Birmingham
B4 7ET
U.K.
Tel: +44 (0)121 359 3611 x5029
Fax: +44 (0)121 359 5271
apart from the question of "intent" that William Steinhurst and
George Richardsons recent postings have covered so well.
I will deliberately not use "SD-language" in what follows, as I do
not think I know enough about SD to deserve a vote as to whether
"feedback loops", "circular causal loops" or something else is the
best term! I hope that I can clarify the issue, though.
It is that in the "coffee cup cooling" example, the standard
treatment assumes that the room temperature is fixed, thus serving as
a "target" value for the temperature of the coffee. This is a good
enough assumption for all practical purposes, BUT, all other things
being equal, the room temperature will actually rise by a negligible
amount as the coffee cools. To illustrate better, suppose we had not
a coffee cup, but a 120-gallon fish tank full of water at say 100
degrees Fahrenheit in the room. In cooling down, that certainly
would raise the room temperature (I know this, as we keep fish!).
Thus the fish tank water temperature is aiming at a moving target,
not the effectively fixed one of the coffee cup.
The issue, on which as I said Im not fit to pronounce, is whether
this sort of equilibrium finding process is different enough from
more conventional feedback processes (like a room thermostat) that a
different name is required to cover both.
John E
j.s.edwards@aston.ac.uk
(Dr.) John S. Edwards
Operations & Information Management Group
Aston Business School
Aston University
Aston Triangle
Birmingham
B4 7ET
U.K.
Tel: +44 (0)121 359 3611 x5029
Fax: +44 (0)121 359 5271
-
gallaher@teleport.com (Ed Gallah
- Member
- Posts: 39
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
In SD-0228, George elaborates further on feedback and feedforward concepts,
and the varying interpretations of these concepts by social scientists,
engineers, and system dynamicists. Of course much more detail is provided
in his recent book as well.
We all know that teaching forces us to examine fundamental concepts, and
learn our subject more thoroughly. This is a particular challenge when we
try to take a professional, academic, scholarly topic such as SD modeling,
and present it to high school students, teachers, parents, and school
boards. We found that some of the students in the SyM Bowl missed the boat
in developing their models, in that they used the pieces of STELLA to make
calculations, but did not include feedback. This means that we as teachers
have not made this clear, and we will focus on this in the future.
The following statement:
"The state of the system (i.e. one or more amounts) affects one or more
rates."
seems to capture the essence of SD, and yet is devoid of reference to
feedback, feedforward, disciplinary biases, etc., etc.
This is important to me in my attempts to describe modeling and simulation
to my biomedical colleagues.
Clear examples of feedback and feedforward can be found in mammalian
thermoregulation.
Feedback: A setpoint exists (37 deg C). The body temp (i.e. core temp,
blood, and therefore hypothalamic temp in the brain) drops below 37 deg, an
error signal is created, a signal is sent to the muscles, and we shiver
to bring the temp back up.
Feedforward: We step out into the cold (10 deg C) and our skin sensors
send signals to the brain. We anticapate that we will get cold, and
shiver, *before* the core temperature decreases measurably.
Neither of the above (I think):
Exponential loss of body temp toward ambient after death (Newtons law of
cooling). In a simple exponential decay model, the amount (coffee temp;
water draining from a barrel; drug elimination; radioactive decay) affects
the output (the fractional loss remains the same, but the absolute loss
declines).
Such a model is readily solved with SD, especially when the inputs are
complex (drug infusion, multiple pulses (doses) at intermittent intervals,
etc), or the outputs are complex (kidney failure, competing metabolism in
the liver, etc.)
This seems to me to be the essence of system dynamics. We have rates
which affect amounts, and we have amounts that affect rates.
Exponential decay has obviously been widely studied. How do you all fit
this into the feedback, feedforward classification?
(George?)
ed gallaher
gallaher@teleport.com
and the varying interpretations of these concepts by social scientists,
engineers, and system dynamicists. Of course much more detail is provided
in his recent book as well.
We all know that teaching forces us to examine fundamental concepts, and
learn our subject more thoroughly. This is a particular challenge when we
try to take a professional, academic, scholarly topic such as SD modeling,
and present it to high school students, teachers, parents, and school
boards. We found that some of the students in the SyM Bowl missed the boat
in developing their models, in that they used the pieces of STELLA to make
calculations, but did not include feedback. This means that we as teachers
have not made this clear, and we will focus on this in the future.
The following statement:
"The state of the system (i.e. one or more amounts) affects one or more
rates."
seems to capture the essence of SD, and yet is devoid of reference to
feedback, feedforward, disciplinary biases, etc., etc.
This is important to me in my attempts to describe modeling and simulation
to my biomedical colleagues.
Clear examples of feedback and feedforward can be found in mammalian
thermoregulation.
Feedback: A setpoint exists (37 deg C). The body temp (i.e. core temp,
blood, and therefore hypothalamic temp in the brain) drops below 37 deg, an
error signal is created, a signal is sent to the muscles, and we shiver
to bring the temp back up.
Feedforward: We step out into the cold (10 deg C) and our skin sensors
send signals to the brain. We anticapate that we will get cold, and
shiver, *before* the core temperature decreases measurably.
Neither of the above (I think):
Exponential loss of body temp toward ambient after death (Newtons law of
cooling). In a simple exponential decay model, the amount (coffee temp;
water draining from a barrel; drug elimination; radioactive decay) affects
the output (the fractional loss remains the same, but the absolute loss
declines).
Such a model is readily solved with SD, especially when the inputs are
complex (drug infusion, multiple pulses (doses) at intermittent intervals,
etc), or the outputs are complex (kidney failure, competing metabolism in
the liver, etc.)
This seems to me to be the essence of system dynamics. We have rates
which affect amounts, and we have amounts that affect rates.
Exponential decay has obviously been widely studied. How do you all fit
this into the feedback, feedforward classification?
(George?)
ed gallaher
gallaher@teleport.com
-
jsterman@MIT.EDU (John Sterman)
- Senior Member
- Posts: 54
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
Re Ed G.:
The feedback loops in the example of a dead body or cup of coffee
cooling towards the ambient temperature are straightforward:
The temperature of the body or coffee affects the rate at which heat is
transferred from the body to the external environment, and the rate at
which heat is lost to the environment affects the temperature of the
body or coffee: an obvious negative feedback loop. Formally,
dH/dt = k(H* - H)
where H is the heat content of the body or coffee, H* is the ambient
temperature, and K is a constant which reflects the composition of the
material, the specific heats of air and body, the shape and composition
of the cup or clothing, and any other factors which influence the
diffusion rate of heat.
All exponential decay processes involve such a negative feedback loop.
Generically the first-order, linear negative feedback loop is given by:
dS/dt = k(S* - S)
where S is the state of the system, S* is the goal of the negative loop,
or the desired state of the system, and dS/dt is the rate of change in
the state of the system. In general, the relationship between dS/dt and
S can be more complex (higher order, nonlinear) but feedback exists
anytime the rate of change in a state depends on the state itself
(either directly or indirectly, linearly or nonlinearly). This is the
most basic concept in system dynamics, and is well treated in nearly all
the texts, including Goodmans study notes in system dynamics and
Richardson and Pughs introduction to system dynamics with dynamo.
John Sterman
jsterman@MIT.EDU
The feedback loops in the example of a dead body or cup of coffee
cooling towards the ambient temperature are straightforward:
The temperature of the body or coffee affects the rate at which heat is
transferred from the body to the external environment, and the rate at
which heat is lost to the environment affects the temperature of the
body or coffee: an obvious negative feedback loop. Formally,
dH/dt = k(H* - H)
where H is the heat content of the body or coffee, H* is the ambient
temperature, and K is a constant which reflects the composition of the
material, the specific heats of air and body, the shape and composition
of the cup or clothing, and any other factors which influence the
diffusion rate of heat.
All exponential decay processes involve such a negative feedback loop.
Generically the first-order, linear negative feedback loop is given by:
dS/dt = k(S* - S)
where S is the state of the system, S* is the goal of the negative loop,
or the desired state of the system, and dS/dt is the rate of change in
the state of the system. In general, the relationship between dS/dt and
S can be more complex (higher order, nonlinear) but feedback exists
anytime the rate of change in a state depends on the state itself
(either directly or indirectly, linearly or nonlinearly). This is the
most basic concept in system dynamics, and is well treated in nearly all
the texts, including Goodmans study notes in system dynamics and
Richardson and Pughs introduction to system dynamics with dynamo.
John Sterman
jsterman@MIT.EDU
-
bbens@MIT.EDU
- Junior Member
- Posts: 10
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
WRT (SD0252) and (SD0254)
According to John Sterman:
> The feedback loops in the example of a dead body or cup of coffee
> cooling towards the ambient temperature are straightforward:
> The temperature of the body or coffee affects the rate at which heat is
> transferred from the body to the external environment, and the rate at
> which heat is lost to the environment affects the temperature of the
> body or coffee: an obvious negative feedback loop. (snip...)
Sender: system-dynamics-approval@world.std.com
Precedence: bulk
Reply-To: system-dynamics
I dont think it is feedback at all. This is an example of "balancing
process" in which the two "bodies" in contact reach an equilibrium
temperature.
Temperature is an indicator of energy content, so molecules of a warmer
body have more (kinetic) energy than those of a cooler one. These
molecules then "transfer" their energy to their counterparts through
interactions (collisions) at the boundary. The process then continues
as the collisions "diffuse" into the originally cooler body.
Essentially, the above is NOT feedback as GPR mentioned below:
> In my earlier message on defining feedback and feedforward, I tried to
> say that people who take an endogenous point of view and think in
> circular causal terms dont have much reason to think in "back" and
> "forward" terms. System dynamicists happily talk about "feedback loops"
> and use the term to stand for all manner of circular causal system
> structures and processes. But we dont have a meaning for "feedback" by
> itself, or "feedforward" by itself.
In the realm of feedback control, we clearly distinguish causal links
(based on physics) from feedbacks. The reason is that we want to isolate
the variables that can be manipulated so that the outputs can be control-
led by manipulating variables that have causal links to the outputs. I
think it is important in socio-economic/socio-political systems to
distinguish causal links from feedback/feedforward. This way we can
clearly distinguish levers to impact the behaviors of the system.
GPR also wrote:
> I understand that the "feedforward" terminology has an important use in
> behavioral decision theory, where it signals (I think) the use of a model or
> tool to project into the future as an aid to making a decision that will
> influence that future. But that seems to be a very different use from
> the engineers notion of feeding the error signal in a control process
> ahead in a plant/controller sequence to improve the controllers
> capabilities (thats Eds example of skin sensors and body temperature).
Engineers also use similar technique in adaptive control. In adaptive
control, control engineers assume a certain structure as a model and
then adjust the parameters on-line based on measurements. This model
is then used as an estimator for the control action!
Just my $0.02 from an engineering perspective of system dynamics and
control,
-Benny Budiman-
bbens@MIT.EDU
Graduate Student
According to John Sterman:
> The feedback loops in the example of a dead body or cup of coffee
> cooling towards the ambient temperature are straightforward:
> The temperature of the body or coffee affects the rate at which heat is
> transferred from the body to the external environment, and the rate at
> which heat is lost to the environment affects the temperature of the
> body or coffee: an obvious negative feedback loop. (snip...)
Sender: system-dynamics-approval@world.std.com
Precedence: bulk
Reply-To: system-dynamics
I dont think it is feedback at all. This is an example of "balancing
process" in which the two "bodies" in contact reach an equilibrium
temperature.
Temperature is an indicator of energy content, so molecules of a warmer
body have more (kinetic) energy than those of a cooler one. These
molecules then "transfer" their energy to their counterparts through
interactions (collisions) at the boundary. The process then continues
as the collisions "diffuse" into the originally cooler body.
Essentially, the above is NOT feedback as GPR mentioned below:
> In my earlier message on defining feedback and feedforward, I tried to
> say that people who take an endogenous point of view and think in
> circular causal terms dont have much reason to think in "back" and
> "forward" terms. System dynamicists happily talk about "feedback loops"
> and use the term to stand for all manner of circular causal system
> structures and processes. But we dont have a meaning for "feedback" by
> itself, or "feedforward" by itself.
In the realm of feedback control, we clearly distinguish causal links
(based on physics) from feedbacks. The reason is that we want to isolate
the variables that can be manipulated so that the outputs can be control-
led by manipulating variables that have causal links to the outputs. I
think it is important in socio-economic/socio-political systems to
distinguish causal links from feedback/feedforward. This way we can
clearly distinguish levers to impact the behaviors of the system.
GPR also wrote:
> I understand that the "feedforward" terminology has an important use in
> behavioral decision theory, where it signals (I think) the use of a model or
> tool to project into the future as an aid to making a decision that will
> influence that future. But that seems to be a very different use from
> the engineers notion of feeding the error signal in a control process
> ahead in a plant/controller sequence to improve the controllers
> capabilities (thats Eds example of skin sensors and body temperature).
Engineers also use similar technique in adaptive control. In adaptive
control, control engineers assume a certain structure as a model and
then adjust the parameters on-line based on measurements. This model
is then used as an estimator for the control action!
Just my $0.02 from an engineering perspective of system dynamics and
control,
-Benny Budiman-
bbens@MIT.EDU
Graduate Student
-
jsterman@MIT.EDU (John Sterman)
- Senior Member
- Posts: 54
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
My friend Benny is right that it is important to distinguish between
those aspects of a system one can influence (the possible policy levers)
and those one cannot (the embedded physics), but he is wrong to say that
the example of a cup of coffee cooling to thermal equilibrium is not an
example of feedback or a feedback loop. The feedback loop structure of
the coffee cooling and a thermostatically controlled environment are
identical (both negative loops). The only difference is that in one
case the set point (goal) of the system is determined by the room
temperature and in the other it is set by a human being. If we ignore
the fact that most real furnaces are discrete on/off devices, and
instead imagine a heating system which continuously delivers heat in
proportion to the gap between the thermostat setting and the room
temperature, then not only will the feedback loop structure of the two
systems be the same, but the differential equations describing them will
also be the same (except for the value of parameters). The difference
is further obscured when one realizes that the temperature of the room
might in fact be under the control of a human - perhaps even the same
one who is setting the thermostat. So the issue of what is the
controller and what is the plant (to use control theory jargon for the
policy levers and the physics of the system) is ambiguous, and depends
on the model boundary. The existence, polarity, and other
characteristics of the feedback loops in the system is not similarly
ambiguous: The coffee cools by negative feedback whether or not a human
being is responsible for the design of the loop or for the value of the
equilibrium reached.
It is useful to make the distinction between implicit and overt
decisions (enunciated by Forrester in Industrial Dynamics, chapter 10).
The control theory usage focuses on feedbacy by overt decisions
(decisions made explicitly by a person or by a controller which has been
designed for the purpose). In system dynamics, we recognize that many
feedbacks arise through implicit decisions, i.e. those carried out as a
result of the physics of a system.
John Sterman
jsterman@mit.edu
those aspects of a system one can influence (the possible policy levers)
and those one cannot (the embedded physics), but he is wrong to say that
the example of a cup of coffee cooling to thermal equilibrium is not an
example of feedback or a feedback loop. The feedback loop structure of
the coffee cooling and a thermostatically controlled environment are
identical (both negative loops). The only difference is that in one
case the set point (goal) of the system is determined by the room
temperature and in the other it is set by a human being. If we ignore
the fact that most real furnaces are discrete on/off devices, and
instead imagine a heating system which continuously delivers heat in
proportion to the gap between the thermostat setting and the room
temperature, then not only will the feedback loop structure of the two
systems be the same, but the differential equations describing them will
also be the same (except for the value of parameters). The difference
is further obscured when one realizes that the temperature of the room
might in fact be under the control of a human - perhaps even the same
one who is setting the thermostat. So the issue of what is the
controller and what is the plant (to use control theory jargon for the
policy levers and the physics of the system) is ambiguous, and depends
on the model boundary. The existence, polarity, and other
characteristics of the feedback loops in the system is not similarly
ambiguous: The coffee cools by negative feedback whether or not a human
being is responsible for the design of the loop or for the value of the
equilibrium reached.
It is useful to make the distinction between implicit and overt
decisions (enunciated by Forrester in Industrial Dynamics, chapter 10).
The control theory usage focuses on feedbacy by overt decisions
(decisions made explicitly by a person or by a controller which has been
designed for the purpose). In system dynamics, we recognize that many
feedbacks arise through implicit decisions, i.e. those carried out as a
result of the physics of a system.
John Sterman
jsterman@mit.edu
-
bbens@MIT.EDU
- Junior Member
- Posts: 10
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
IRT SD0261
> In his 1948 control theory text, Gordon Brown (Jays mentor, and so in
> some sense ours as well) wrote: "A closed-loop control system is thus an
> error-sensitive system and, being such, it acquires certain pecularities
> and idiosyncrasis which, in large measure, are the reasons for this book."
> And for our entire field, Im sure he would be happy to add now.
Exactly my point!!! A closed-loop control system is so because there is a
means to compare output and reference through measurement. This comparison
is in essence "feeding back" the information upon which a control action is
generated to drive error to zero. Note that there are two subsystems in the
so-called closed-loop control system: the plant to be controlled and the
controller. Lets side track a bit...
In many disciplines of engineering, the field from which System Dynamics
has evolved, engineers do modeling for a number of reasons--the main ones,
I think, being the desire to estimate the behavior without having to build
prototypes (very useful in design stage) and being the tool in designing a
controller without causing "damages" to the actual plant.
Back to my point, modeling efforts--as system dynamicists do for social,
economic and political systems--try to capture the behavior of a
particular system by understanding the structure that is in place to cause
the system to behave a certain way. I think it is universal that there are
causal links in the structure among the elements. Please keep in mind,
however, that these causal links are "fixed" until redesigned. Of course
nature has her closed-loop control subsystems as many biologists have seen.
But, there is no way to alter it without changing the intrinsic structure
of these systems. SO, I tend to call the links these closed-loop subsystems
as causal links as they are embedded in the "plant."
My point of separating causal links intrinsic in the system from feedback
or feedforward links that are the result of policy design and implementation
is, I think, helpful in identifying (long) levers that are able to influence
the behavior of the system. If, for any unfortunate reasons, no levers long
enough are to be found, such an understanding will be helpful in redesigning
the system. I have witnessed the success of this approach in engineering
field. The question is "Why does it not work in other systems?"
> We need a term for the circularities that are so crucial to an endogenous
> view of a systems dynamics. Forrester wrote "Systems of information
> feedback control are fundamental to all life and human endeavor, from the
> slow pace of biological evolution to the launching of the latest satellit.
> A feedback control systems exists whenever the environment causes a
> decision which in turn affects the original environment" (Forrester 1958,
> p. 4).
I agree wholeheartedly to Forresters statement above. But, do keep in
mind that feedbacks exist when people are comparing their goal (desire) and
the actual outcome of their effort. The purpose is, I think everyone would
agree to this, to formulate what the next step to close the gap (error)
should be. Note that Forrester acknowledged "the environment causes a
decision which in turn affects the original environment." This means that
there is an intrinsic "measuring" and "comparing" actions in place in the
system. The key is to recognize these actions. Using the same terminology
of feedbacks for anything ranging from causal links to loops of action can
be confusing as we all have seen by the number of questions and discussions
on this list.
> Lets also agree that some may want to identify, for their own purposes,
> a given link as a "feedback link" and to distinguish that kind of link
> from some other kinds which they would say are not "feedback links." But
> lets insist that if a causal loop is formed, we will all call it a
> "feedback loop," not matter how its links are labeled, or by whom.
> Similarly, lets agree that some may find a good use for the term
> "feedforward link." For both these link terms, however, we would require
> much stricter definitions than weve seen so far, and many of us would
> probably want to see strong reasons for paying attention to the
> distinctions in the dynamic policy systems we are addressing.
This terminology is fine for some people, but confusing for others. As I
have heard a joke about systems dynamicists. The joke was that system
dynamicists just add more and more feedback loops whenever the model
didnt match historical data such that everything was nothing more than
a spaghetti!! This spaghetti-looking diagram may be one factor that
impedes the progress of system dynamics as a tool or methodology.
> And let us again focus on what is most important here to ourselves and to
> the larger world which for the most part does not share our perspective:
> we are taking an endogenous view (and internal view, a "system as cause"
> view) of system dynamics, and that forces us to see causality as
> circular. It is the loop view that is crucial, not the name we attach to
> any particular link.
What is important to ourselves here is the ability to clearly understand
that the structure of any system causes the system to behave a certain
way. Also of importance is the ability to design policies or even the
system itself from the "structure as cause" view. While the name may not
matter, it may create confusion that should be avoided if ever possible.
Just a thought or two,
-Benny Budiman-
bbens@MIT.EDU
> In his 1948 control theory text, Gordon Brown (Jays mentor, and so in
> some sense ours as well) wrote: "A closed-loop control system is thus an
> error-sensitive system and, being such, it acquires certain pecularities
> and idiosyncrasis which, in large measure, are the reasons for this book."
> And for our entire field, Im sure he would be happy to add now.
Exactly my point!!! A closed-loop control system is so because there is a
means to compare output and reference through measurement. This comparison
is in essence "feeding back" the information upon which a control action is
generated to drive error to zero. Note that there are two subsystems in the
so-called closed-loop control system: the plant to be controlled and the
controller. Lets side track a bit...
In many disciplines of engineering, the field from which System Dynamics
has evolved, engineers do modeling for a number of reasons--the main ones,
I think, being the desire to estimate the behavior without having to build
prototypes (very useful in design stage) and being the tool in designing a
controller without causing "damages" to the actual plant.
Back to my point, modeling efforts--as system dynamicists do for social,
economic and political systems--try to capture the behavior of a
particular system by understanding the structure that is in place to cause
the system to behave a certain way. I think it is universal that there are
causal links in the structure among the elements. Please keep in mind,
however, that these causal links are "fixed" until redesigned. Of course
nature has her closed-loop control subsystems as many biologists have seen.
But, there is no way to alter it without changing the intrinsic structure
of these systems. SO, I tend to call the links these closed-loop subsystems
as causal links as they are embedded in the "plant."
My point of separating causal links intrinsic in the system from feedback
or feedforward links that are the result of policy design and implementation
is, I think, helpful in identifying (long) levers that are able to influence
the behavior of the system. If, for any unfortunate reasons, no levers long
enough are to be found, such an understanding will be helpful in redesigning
the system. I have witnessed the success of this approach in engineering
field. The question is "Why does it not work in other systems?"
> We need a term for the circularities that are so crucial to an endogenous
> view of a systems dynamics. Forrester wrote "Systems of information
> feedback control are fundamental to all life and human endeavor, from the
> slow pace of biological evolution to the launching of the latest satellit.
> A feedback control systems exists whenever the environment causes a
> decision which in turn affects the original environment" (Forrester 1958,
> p. 4).
I agree wholeheartedly to Forresters statement above. But, do keep in
mind that feedbacks exist when people are comparing their goal (desire) and
the actual outcome of their effort. The purpose is, I think everyone would
agree to this, to formulate what the next step to close the gap (error)
should be. Note that Forrester acknowledged "the environment causes a
decision which in turn affects the original environment." This means that
there is an intrinsic "measuring" and "comparing" actions in place in the
system. The key is to recognize these actions. Using the same terminology
of feedbacks for anything ranging from causal links to loops of action can
be confusing as we all have seen by the number of questions and discussions
on this list.
> Lets also agree that some may want to identify, for their own purposes,
> a given link as a "feedback link" and to distinguish that kind of link
> from some other kinds which they would say are not "feedback links." But
> lets insist that if a causal loop is formed, we will all call it a
> "feedback loop," not matter how its links are labeled, or by whom.
> Similarly, lets agree that some may find a good use for the term
> "feedforward link." For both these link terms, however, we would require
> much stricter definitions than weve seen so far, and many of us would
> probably want to see strong reasons for paying attention to the
> distinctions in the dynamic policy systems we are addressing.
This terminology is fine for some people, but confusing for others. As I
have heard a joke about systems dynamicists. The joke was that system
dynamicists just add more and more feedback loops whenever the model
didnt match historical data such that everything was nothing more than
a spaghetti!! This spaghetti-looking diagram may be one factor that
impedes the progress of system dynamics as a tool or methodology.
> And let us again focus on what is most important here to ourselves and to
> the larger world which for the most part does not share our perspective:
> we are taking an endogenous view (and internal view, a "system as cause"
> view) of system dynamics, and that forces us to see causality as
> circular. It is the loop view that is crucial, not the name we attach to
> any particular link.
What is important to ourselves here is the ability to clearly understand
that the structure of any system causes the system to behave a certain
way. Also of importance is the ability to design policies or even the
system itself from the "structure as cause" view. While the name may not
matter, it may create confusion that should be avoided if ever possible.
Just a thought or two,
-Benny Budiman-
bbens@MIT.EDU
-
"Robert M. Kane"
- Junior Member
- Posts: 7
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
I started this message and then set it aside but now feel compelled to send it
along:
Benny Budiman made some comments that I may need to think through in greater
depth but he closed on a point close to my heart:
> What is important to ourselves here is the ability to clearly
> understand that the structure of any system causes the system to behave
> a certain way. Also of importance is the ability to design policies
> or even the system itself from the "structure as cause" view. While
> the name may not matter, it may create confusion that should be avoided
> if ever possible.
The name is critically important. I was much impressed by Jeremy Campbells
book Grammatical Man not because of his attempt to connect DNA and language
but because it suddenly occurred to me that every attempt at analysis centers
on finding a grammar to express the otherwise unexpressible.
I believe that I have witnessed attempts to undermine a nascent grammar by
those who cant take the time to understand it. The words and meanings
associated with each word (and more complex grammatical structures (phrases))
are very important, and anyone interested in participating has a
responsibility to try very hard to conform to the grammar. Without it we have
no real exchange of ideas.
Additionally:
A grammar is not truth-driven proposition. Rather it provides a path to
arrive at some end. Budians desire to de-emphasize natural systems seems to
say, "lets not spend time studying phenomena that nature already manages
efficiently". Instead, lets focus on controls that guide natural systems to
states that are optimal for human survival which dont correspond with the
physics of dissipation of energy. The reason for this arbitrary grammatical
choice is to guide the discussion toward profitable ends. It is not because
there is something undesireable about natural processes.
Bob Kane, North American Coordinator
Institute for Sustainable Agriculture in the Tropics
Jaen, Peru
robert_kane@ds.Cubic.COM
along:
Benny Budiman made some comments that I may need to think through in greater
depth but he closed on a point close to my heart:
> What is important to ourselves here is the ability to clearly
> understand that the structure of any system causes the system to behave
> a certain way. Also of importance is the ability to design policies
> or even the system itself from the "structure as cause" view. While
> the name may not matter, it may create confusion that should be avoided
> if ever possible.
The name is critically important. I was much impressed by Jeremy Campbells
book Grammatical Man not because of his attempt to connect DNA and language
but because it suddenly occurred to me that every attempt at analysis centers
on finding a grammar to express the otherwise unexpressible.
I believe that I have witnessed attempts to undermine a nascent grammar by
those who cant take the time to understand it. The words and meanings
associated with each word (and more complex grammatical structures (phrases))
are very important, and anyone interested in participating has a
responsibility to try very hard to conform to the grammar. Without it we have
no real exchange of ideas.
Additionally:
A grammar is not truth-driven proposition. Rather it provides a path to
arrive at some end. Budians desire to de-emphasize natural systems seems to
say, "lets not spend time studying phenomena that nature already manages
efficiently". Instead, lets focus on controls that guide natural systems to
states that are optimal for human survival which dont correspond with the
physics of dissipation of energy. The reason for this arbitrary grammatical
choice is to guide the discussion toward profitable ends. It is not because
there is something undesireable about natural processes.
Bob Kane, North American Coordinator
Institute for Sustainable Agriculture in the Tropics
Jaen, Peru
robert_kane@ds.Cubic.COM
-
jsterman@MIT.EDU (John Sterman)
- Senior Member
- Posts: 54
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
Paul atkins is right that the molecular level "explains" heat transfer
"perfectly" - except that no one can solve analytically or by simulation
in reasonable time (i.e. less than the life of the universe) a molecular
interaction model of a cup of coffee cooling to room temperature - much
less a more interesting system like drug diffusion in an organism. So
while the theory is perfect, it is useless for the problem that was
posed. To understand the phenomenon of coffee cooling, we need models
at a higher level of aggregation. It is the same in other fields - the
Navier-Stokes equations "perfectly" describe fluid dynamics from first
principles (Newtons laws) - except no one can solve them for realistic
systems (including any turbulent flows). All fluid dynamics in practice
rests therefore on approximations and models at higher levels of
aggregation.
These models not only gain something compared to models of the
individual molecules interacting, they are currently the only way we
gain knowledge about the dynamics of aggregates like pressure,
temperature, etc. in complex systems.
For those interested in the individual entity level of modeling, there
is exciting progress being made. Steen Rasmussen at Los Alamos/Santa Fe
Institute is developing terrific models of bio-chemical interactions at
the level of the individual molecules. These models cant simulate in
reasonable time the interactions by solving e.g. Schrodingers equation,
but instead use a variety of simplifications and heuristics to capture
the interactions of the particles. They can simulate the formation of
small structures at present, but a cup of coffee will almost certainly
forever remain outside the reach of detailed micro-molecular modeling.
John Sterman
jsterman@mit.edu
"perfectly" - except that no one can solve analytically or by simulation
in reasonable time (i.e. less than the life of the universe) a molecular
interaction model of a cup of coffee cooling to room temperature - much
less a more interesting system like drug diffusion in an organism. So
while the theory is perfect, it is useless for the problem that was
posed. To understand the phenomenon of coffee cooling, we need models
at a higher level of aggregation. It is the same in other fields - the
Navier-Stokes equations "perfectly" describe fluid dynamics from first
principles (Newtons laws) - except no one can solve them for realistic
systems (including any turbulent flows). All fluid dynamics in practice
rests therefore on approximations and models at higher levels of
aggregation.
These models not only gain something compared to models of the
individual molecules interacting, they are currently the only way we
gain knowledge about the dynamics of aggregates like pressure,
temperature, etc. in complex systems.
For those interested in the individual entity level of modeling, there
is exciting progress being made. Steen Rasmussen at Los Alamos/Santa Fe
Institute is developing terrific models of bio-chemical interactions at
the level of the individual molecules. These models cant simulate in
reasonable time the interactions by solving e.g. Schrodingers equation,
but instead use a variety of simplifications and heuristics to capture
the interactions of the particles. They can simulate the formation of
small structures at present, but a cup of coffee will almost certainly
forever remain outside the reach of detailed micro-molecular modeling.
John Sterman
jsterman@mit.edu
-
jsterman@MIT.EDU (John Sterman)
- Senior Member
- Posts: 54
- Joined: Fri Mar 29, 2002 3:39 am
Feedback -forward?
Re john edwards and the cup of coffee cooling down:
The fact that in real life the air around a cup of coffee has finite
specific heat, and therefore warms slightly as the coffee cools, is
ABSOLUTELY NOT inconsistent with the existence of the feedback loop by
which coffee cools to the surrounding temperature. The implicit
assumption in the simplest coffee cooling model is that the surrounding
medium has infinite heat absorption capacity, so that the ambient
temperature stays constant even as heat is transferred from coffee to
air. This is a pretty good assumption for most purposes since
convection currents carry the heat away from the cup, drawing cooler air
in to take its place. However, if one wanted to recognize in the model
the warming effect on the air, the previously exogenous variable T* (the
ambient temperature around the cup) becomes an endogenous variable,
introducing NEW feedback loops. The old loop continues to exist and
operate, but a new negative feedback (the warming of the air to the
temperature of the coffee) is added. Both loops are conventional
negative feedbacks, and no new names or concepts are required to
understand the process or its dynamics.
John Sterman
jsterman@mit.edu
The fact that in real life the air around a cup of coffee has finite
specific heat, and therefore warms slightly as the coffee cools, is
ABSOLUTELY NOT inconsistent with the existence of the feedback loop by
which coffee cools to the surrounding temperature. The implicit
assumption in the simplest coffee cooling model is that the surrounding
medium has infinite heat absorption capacity, so that the ambient
temperature stays constant even as heat is transferred from coffee to
air. This is a pretty good assumption for most purposes since
convection currents carry the heat away from the cup, drawing cooler air
in to take its place. However, if one wanted to recognize in the model
the warming effect on the air, the previously exogenous variable T* (the
ambient temperature around the cup) becomes an endogenous variable,
introducing NEW feedback loops. The old loop continues to exist and
operate, but a new negative feedback (the warming of the air to the
temperature of the coffee) is added. Both loops are conventional
negative feedbacks, and no new names or concepts are required to
understand the process or its dynamics.
John Sterman
jsterman@mit.edu