# Ashby and Control Theory

**URL:** <http://discourse.iapct.org/t/ashby-and-control-theory/1324>\
**Category:** CSG2009\
**Created:** [November 30, 2009, 10:37pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324 "2009-11-30T22:37:22Z")\
**Posts on this page:** 11\
**Page:** 2

<div class="post-metadata">

**Author:** ![Bill\_Powers1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/8e7dd6/32.png) [@Bill\_Powers1](http://discourse.iapct.org/u/Bill_Powers1)\
**Post date:** [December 5, 2009, 3:53pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/21 "2009-12-05T15:53:52Z")

</div>

> Hi to all !
> 
> BH: Thanks for your answers. It doesn’t seem to me that I was trying to  
> define anything, I just used the definition I found in B:CP about  
> feedforward and try some examples (experiences) from everyday life (or  
> with free-living
> 
> organisms) to find out, whether feedforward correspond to evolution
> 
> necessity of development in histroy of organisms or not.

BP: It’s important to define feedforward, because it’s easy to forget  
that there is no feedback from the result when feedforward is the only  
method of control. If the feedforward action is wrong, it will continue  
to be wrong every time. If there is anything wrong with the computation  
used to calculate the action that is needed, it will never be corrected.  
If there is a disturbance that makes the result different from the  
calculated result, that difference will never be corrected.

> BH: I offered 4 cases and  
> neither was answered directly. Only some abstract
> 
> definitions.

BP: I thought I would point out some of the details about feedfoward so  
you could discover for yourself any problems in your examples, rather  
than just telling you about them. I find that to be a useful way to find  
out when I have made a mistake, and maybe less embarrassing than having  
someone else point it out.

> BH: I still think that  
> however we define feedforward (anticipation,
> 
> prediction, reflex) it has it’s function in maintaining the  
> essential
> 
> variables in their limits.

BP: Yes, that’s the result we want, but unfortunately in feedforward  
control the “limits” have no effect, because by definition the  
feedforward system never learns the results of its actions and has no way  
of correcting errors. We can hope that the feedforward system  
“maintains essential variables in their limits,” but if it does  
not, it will not change its behavior. Only feedback systems sense the  
results of their behavior and change their actions if the results are not  
right.

> BH: So I still think that  
> development of feedforward was somehow necessary from the perspective of  
> evolution. I’ll still remain with definition of feedforward given in B:CP  
> (p.48, 2005)

BP: So will I. But you have to know more about the details to arrive at  
correct conclusions. Feedforward means acting without feedback. If the  
system somehow is adjusted by some external agency so it acts correctly,  
all will be well – until something changes in the environment or inside  
the system itself. Then the actions will generate the wrong result, and  
will continue to do so.

> BH: 1. CASE ­ NATURE AND FREE  
> LIVING ORGANISMS : It was about how free-living
> 
> organisms in wild nature respond to some signs of predator with
> 
> feed-forward. The problem was how could we describe what’s happening  
> inside
> 
> animals, for example antilope, who senses the predator (probably  
> possible
> 
> actual disturbance to essencial variables) and starts suddenly running  
> away
> 
> and there is no attack of predator (for example lion) or the  
> predator
> 
> attacks anyway, after she starts running.

BP: Here the problem is, how does the feedforward system know which  
direction is ‘away’? If the antelope just starts running in some random  
direction, it is as likely to be running toward the danger as away from  
it. This would save the antelopes who happened to be running mostly away  
from the danger, but not the others. That’s better, evolution-wise, than  
having all the antelopes killed, but it doesn’t give the next generation  
any advantage because survival was by pure chance, not because of any  
mutation in the animal’s behavior.

It would be much better for antelopes if they could learn to perceive a  
relationship between their own positions and the positions of predators,  
and control this relationship so as to be sure of increasing the distance  
between them and the predator every time, no matter where the predator  
is, and no matter whether there is more than one predator (it doesn’t  
help much to run away from one predator if that means running toward  
another).

> BH: I still think that  
> feed-forward saves life to antilope. We can see that practically everyday  
> on NG WILD.

BP: No. What you see is antelopes running away from predators. You  
_imagine_ that this is being done blindly, without regard to the  
relationship between the direction of running and the direction of the  
predator. In wildlife programs I have seen, the prey can be seen altering  
its path as predators approach and try to cut off its escape, which  
suggests that the actions are being varied to counteract a perceived  
decrease in distance from the predators.

> BH: If I try to follow  
> definition in B:CP (2005) can we say that antilope is
> 
> sensing the cause of the disturbance (lion) and she anticipates it’s  
> effects
> 
> (deadly attack of the lion) thus improving control.

BP: This does not improve control of the relationship to the lion; it  
improves the decision about when to start controlling it, a higher-level  
process. Once the antelope start to evade the lion, feedback control is  
essential; all the animals that simply rushed around without looking to  
see the results have been eaten by now, so none are left to  
reproduce.

> BH: What happens in  
> antilope? Can we simply say that antilope is anticipating (predicting)  
> future dangerous situation on the bases of smell and counteract to  
> something what will acctually occur or never occurs.

BP: Reacting on the basis of smell is very chancy, since smell carries no  
directional information (except perhaps that the source is somewhere in  
the direction from which the wind is blowing, if there is one). The best  
feedforward action that might occur is to start looking around more  
alertly, trying to locate where the lion is, so that the action can be  
adjusted appropriately to the error.

> BH: Whatever happens on  
> the basis of feedforward, it’s more efficiently than pure control. It’s  
> maintaining antilope’s essential variables stable, what would not be the  
> case with only classic control.

BP: I don’t understand why you say that. Feedback control is always more  
efficient than feedfoward, when it is possible. If it’s not possible (the  
antelope can smell the lion but can’t see it), feedforward may offer a  
better chance of survival than doing nothing would offer, but it’s still  
not a very good chance. You just assume that feedforward would maintain  
the antelope’s essential variables stable, but that’s not enough: you  
have to give some reason for saying that this desireable result would  
actually be achieved. Under some conditions feedforward can reduce the  
variations in essential variables that would otherwise occur, at least  
with some probability. But feedback control, when possible, will always  
do that better.

> BH: Whatever is happening, it’s  
> saving her life.

BP: You assume it saves her life, but there is a quite large chance that  
it won’t.

> BH: If she stays and the attack  
> of predator begins (IN CONTROL MODE OF ANTILOPE) she has less chances to  
> escape and so to survive or there is even no chances.

BP: But if she sees the lion and is able to run in a direction that is  
actually away from the lion, she then has an even better chance of  
surviving.

> BH: 2. CASE : was about the  
> drivers and pedestrians. If the driver is
> 
> controlling only speed of the car and course of the car and turn right  
> in
> 
> the crossing, he could run over the pedestrian. If he is controlling  
> the
> 
> speed of the car and the speed of the pedestrian, they both can come  
> to
> 
> crossing at the same time and driver would be forced to stop the car
> 
> immediately so risking that some driver behind him bump into his car.
> 
> If the driver is anticipating (predicting) the course and speed of his  
> car
> 
> and course and speed of the pedestrian, he improves control and make  
> save
> 
> deccision whether to stop or accelerate and safely turn in  
> crossing.

BP: But now you’re assuming a control system. Feedforward systems don’t  
sense the results they produce and change their behavior  
accordingly.

> BH: So if the driver is sensing  
> the cause of distrubance (pedestrian) and
> 
> anticipating it’s effects (possibility of running over the pedestrian),  
> he
> 
> is improving his control.

BP: The problem with this mode of “proof” is that you assume  
the result before you have shown that this result would actually occur.  
You’re imagining that the driver is blind to the pedestrian until it’s  
too late to do anything (which does happen occasionally and accounts for  
quite a few pedestrian deaths). What about controlling the perceived  
position of the car in relation to the pedestrian? Are you declaring that  
this is impossible? I realise that the appearance of collision avoidance  
is that the controller is predicting the path of the other object, but  
this is a very poor way to avoid collisions, though it might work if you  
could perceive and compute very rapidly and accurately. A much better way  
is just to make sure that the direction from you to the other object,  
relative to your direction of travel, is changing at a reasonable rate.  
If the direction is not changing, there will be a collision. So alter  
your direction until the bearing of the other object starts to change  
(either way), and you will most probably miss it.

> BH: 3. CASE : In every sport I  
> mentioned, anticipating (predicting) what
> 
> opponent will do is huge advantage and ussually means winning.  
> Better
> 
> anticipation of players means more chances for winning.

BP: You assume that it’s the anticipation of the opponent’s move that  
means winning, but you doing this in order to show that anticipation is  
what allows winning. That is a logical error, with various names, one  
being “begging the question.”  
[  
http://en.wikipedia.org/wiki/Circular\_logic](http://en.wikipedia.org/wiki/Circular_logic)

Opponents also use your anticipation to get past you, don’t they? In  
football it’s called “faking the defender out of his shoes.”  
You display the start of a move to the left, and the instant the opponent  
anticipates your move, you go the other way.

BH: Some actions of players I think it can be done only with presence of  
feed-forward (throw, handing the ball over).

BP: This is something else: you control the way you throw in relation to  
the target, so you feel the right accelerations and velocities. That’s  
feedback control. But you have no control of any kind after you let the  
ball go. You do see the result, so you can make changes in the way you  
throw, but not until the next time. That is still feedback  
control.

> BH: The goal of the game is more  
> efficiently achieved if the moving of players are anticipated (predicted,  
> planned). It’s so called tactics.

BP: True, and I would call this feedforward. If it worked as well as  
feedback, the game would be over at the start because the best strategy  
would always win. But it doesn’t.

> BH: So sensing the cause of  
> disturbance (opponent) and anticipation of effects (opponent’s action)  
> players are improving control.

BP: Yes. But the whole question rests on just how much improvement you  
can get from feedforward, especially when feedback control is  
possible.

> BH: So feed-forward is somehow  
> neccessary because everybody wants to win, and
> 
> most players and teams make tactics and improve other elements of  
> training.

BP: No, it’s necessary in order to win only when feedback control is  
impossible.

> BH: Trainings contains also  
> the improvement of feedforward actions, so to
> 
> achieve their goals : to WIN, thus have more money and higher quality  
> of
> 
> life (survival) or in if we say it in Ashby’s language to maintain  
> essential
> 
> variables in the limits.

BP: You can’t make feedforward control any more effective by pointing out  
how important it is to achieve a goal, have more money, or have a higher  
quality of life. Those are not goals of feedforward because feedforward  
systems have no goals. They only react to inputs with a precalculated  
action. It makes no difference to them if the action takes them  
closer to some goal assumed by an observer or away from it.

> BH: 4. CASE : putting all kind  
> of vegetables in water is by my oppinion
> 
> feed-forward act (anticipating the taste of the soup), and testing it  
> while
> 
> cooking is control act.

BP: I would say that putting vegetables in water is a control process. Do  
you think you could do that blindfolded and without being allowed to feel  
where the vegetables and the water are? Doing this is part of a  
higher-level process that controls a sequence called “preparing  
dinner.” Each step in the sequence is brought about in the correct  
order at the correct time to achieve the goal. If there are disturbances  
of the sequence, they are corrected; parts of the sequence are repeated  
in the correct order, or the correction is made before the next step  
takes place.

I think what you’re missing here is the main point: Feedforward is simply  
a fancy name for a stimulus causing a response. If the result of the  
response is sensed and the next forward act is modified according to the  
result of the previous response, we have a feedback system, not a  
feedforward system. The only time you have a real feedforward system is  
when the response to the stimulus remains the same every time no matter  
what the result of the response is. If the stimulus of dinnertime calls  
for putting the celery in a bowl of water, and you find the celery  
already in a bowl of water, the feedforward system would put it in  
another bowl of water. Timed lawn sprinkers are feedforward systems which  
react to a timer by turning on the water even in a rainstorm. A  
feedforward system puts the meat in a pan and turns the gas knob even if  
the meat has been cooked already, and even if the gas has been turned off  
at the main entry to the house. If the gas has been turned off, the  
feedforward system still waits 15 minutes, then takes the meat out of the  
pan and serves it (if the plate is exactly where it is supposed to be and  
not a foot to the left or right).

> BH : So we agree that building  
> houses, skyscrapers, bridges with plans
> 
> (feed-forward) very much improve control.

BP: No, we do not agree; YOU agree. The fact that the outcome of a  
control process takes time to appear, and appears in the future, does not  
make it a feedforward process. Building a house would be a feedforward  
process if the pieces of wood were put in place and the hammers  
were swung at the nails and the walls were pushed upright all without  
ever perceiving them or comparing the results with the  
blueprint.

> BH: Plans in constructing  
> anything (probably also when engineers construct machines) are  
> necessary.

BP: Yes. They are specifications for what is to be perceived, not for the  
actions that are to be performed. The actions used in building two  
identical houses are very different from one house to another. The  
carpenters have a saying: “measure twice, cut once.” Building  
anything is a process of varying actions to achieve a precise result, and  
the precise result is achieved by comparing perceptions continually to  
reference perceptions. Feedforward systems have no reference  
perceptions.

> BH: So the LAW
> 
> says. Despite this, some smart guys build houses without plans (only
> 
> control).

BP: That is not control, it’s feedforward.

> BH: The consequences are  
> ussually seen. Speccially on muddy ground
> 
> like in our country. The houses simply sink into the ground. There are  
> many
> 
> things to predict when building houses.

BP: If a house was built only with feedforward, there would be no way to  
predict the result. It would probably be a random pile of lumber with  
bent nails lying all over the ground and no structure higher than a  
carpenter’s waist.

> BP earlier: The safest thing to  
> do is trust the feedforward, but verify it with feedback control to  
> handle the many situations in which feedforward fails to help or makes  
> matters worse.
> 
> BH : So I think we solve the problem. Can we agree that feed-forward  
> is
> 
> improving control and it’s mostly necessary with activities living  
> organisms
> 
> do, along with feed-back control ?

BP: I said “the many situations in which feedforward fails to help  
or makes matter worse.” Is that agreement? I think feedforward does  
occur, but that it is too crude to be relied on. If you want predictable  
results and precise results and fast results, you need feedback control  
systems. We use feedforward only when there isn’t any way to monitor the  
results and make corrections as we proceed. If necessary, we’ll inflate a  
flat tire without a pressure gauge, but as soon as we can we’ll stop and  
measure the pressure before going on. Only a brave man would fill his  
bathtub with water and jump into it without feeling the temperature while  
filling it.

> BH: And can we agree that some  
> combination of feed-back and feed-forward is ussually present as Arthur  
> proposed : “Is the combination of feedback and feedforward not  
> preferable in most situations ?”

BP: I have built many control systems, fixed many others, and observed  
still more of them, and have never found that feedforward was a good way  
to improve control by more than a very small amount. The most common  
result of adding some feedforward to get faster responses was to make the  
system unstable and cause oscillations, as well as making the system  
respond needlessly to random noise.

I really can’t see why feedforward is getting so much support. Do people  
just like the word? Is it to achieve some kind of symmetry with feedback?  
Would people be sad or disappointed if we just stopped using the word? Or  
is it, as I rather suspect, a throwback to the days when ALL behavior was  
thought to be feed-forward, cause-effect, antecedent-consequent,  
stimulus-response?

I think that if we just keep in mind that feedforward is an action taken  
in response to some stimulus, and taken blindly with no modification on  
the basis of its results, we can put feedforward in its proper place. It  
really doesn’t work any better in conjunction with control theory than it  
did when it was the only theory of behavior.

Best,

Bill P.

> **···**
>
> At 03:13 AM 12/5/2009 -0600, you wrote:

---

<div class="post-metadata">

**Author:** ![Arthur\_Dijkstra](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/a/f9ae1b/32.png) [@Arthur\_Dijkstra](http://discourse.iapct.org/u/Arthur_Dijkstra)\
**Post date:** [December 5, 2009, 5:16pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/22 "2009-12-05T17:16:30Z")

</div>

From Arthur Dijkstra 2009-12-05

What do you think of this use of feedback and feedforward ?

[http://www.ida.liu.se/~eriho/ECOM\_M.htm](http://www.ida.liu.se/~eriho/ECOM_M.htm)

Arthur

---

<div class="post-metadata">

**Author:** ![Bill\_Powers1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/8e7dd6/32.png) [@Bill\_Powers1](http://discourse.iapct.org/u/Bill_Powers1)\
**Post date:** [December 5, 2009, 5:21pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/23 "2009-12-05T17:21:18Z")

</div>

[From Bill Powers (2009.12.05.0945 MDT)]

Arthur Dijkstra 2009-12-05 –

> What do you think of this use of  
> feedback and feedforward ?
> 
> [  
> http://www.ida.liu.se/~eriho/ECOM\_M.htm](http://www.ida.liu.se/~eriho/ECOM_M.htm)

Erik Hollnagel’s model looks a lot like mine, in that higher levels act  
by setting reference conditions for lower levels. However, he missed  
seeing that the perceptions in a higher level system are derived from  
lower-level perceptions, so he didn’t see that all the levels are  
closed-loop systems. He didn’t see that the higher system sets the goals  
for lower systems or that it does so _as a means of controlling its own  
perception_.

If he had seen this, he would have seen that “Targeting” is not  
merely a goal-setting system, open-loop or feedforward. The behavior of  
the lower systems has effects perceived at the top level as a change in  
the “situation” and its assessment, and that closes the  
loop.

Also, Hollnagel couldn’t see how a higher system could adjust the goal of  
a lower one without disconnecting the lower system’s target, goal, or  
plan so as to substitute its own target etc… This is the same problem  
that Rodney Brooks has with his “subsumption” architecture: how  
to let the higher system control when there are lower control systems  
already working, which would fight back if the higher system tried to  
alter the behavior of the lower system. I saw that problem when thinking  
about the spinal control systems: how could any higher system change the  
muscle tensions if the spinal systems were already controlling them?

My solution was to realize that the higher system acted not by sending  
outputs to the muscles but by varying the reference signals of the spinal  
systems – that is, not by substituting their outputs for the actions of  
lower systems. The higher systems use the lower systems as part of their  
output equipment. This idea allowed all levels to be operating at the  
same time, with no conflicts between levels.

Hollnagel’s four levels are named for three synonyms of controlling and  
one of perceiving: Tracking, Regulating, Monitoring, and Targeting. His  
examples of these levels seem OK in terms of my levels of perception,  
though they’re not clearly tied to my scheme.

Here’s his diagram:

 ![10aecd3.jpg](http://discourse.iapct.org/uploads/default/original/1X/4e371bf07f1d0e34b2b9c2b667da04f0d4db785b.jpeg)

As usual in this field, he cites the TOTE unit of Miller, Galanter, and  
Pribram (1960) but not Powers, Clark, and MacFarland (1960) in which this  
whole hierarchical control scheme was laid out. A lot of people seem to  
think that my first publication on this subject was in 1973, or 1985, or  
2001, or whatever the date was on the journal in which they found it. Oh,  
well.

Feedforward is perfectly OK with me, but I just don’t think it’s very  
important. Most of the examples I see are feedback systems, the authors  
forgetting that the actions commanded by the system alter the perceptions  
on which the action is based, which turns “forward” into  
“back.”

Did Hollnagel ever turn his descriptions here into a working  
simulation?

Best,

Bill P.

---

<div class="post-metadata">

**Author:** ![\_Boris\_Hartman](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/_/e8c25b/32.png) [@\_Boris\_Hartman](http://discourse.iapct.org/u/_Boris_Hartman)\
**Post date:** [December 5, 2009, 7:10pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/24 "2009-12-05T19:10:21Z")

</div>

Bill !

BP: It's important to define feedforward

BP : Sensing the cause of a disturbance and anticipating it's effects can  
sometimes improve control (feed-forward) but is by no mean necesary and  
usually it's not even possible. (B:CP, p. 48, 2005)

BH : O.K. please explain to me, what did you mean with this definition ?  
Maybe you could add some examples.

BP : The safest thing to do is trust the feedforward, but verify it with  
feedback control to handle the many situations in which feedforward fails to  
help or makes matters worse

BH : What did you mean with this relation between feedforward and feedback ?  
Is this happening in the same control system ?

BP : As to other kinds of feedforward such as making plans or taking  
precautions, those belong to higher-level kinds of control which involve  
much slower changes and longer delays, and as Robert Burns said, often go  
wrong or "gang aft agley". We try to anticipate problems, but the unexpected  
happens anyway, and we just have to rely on our good old negative feedback  
control systems to deal with the many uncorrected errors that are left when  
the feedforward has done all it can do.

BH : Can you explain to me this relation between feedforward and feedback ?  
Maybe some examples ?

BH : So the LAW says. Despite this, some smart guys build houses without  
plans (only control).  
You wrote it is feedforward ? Can you explain ?

Boris

---

<div class="post-metadata">

**Author:** ![rsmarken](http://discourse.iapct.org/user_avatar/discourse.iapct.org/rsmarken/32/3553_2.png) [@rsmarken](http://discourse.iapct.org/u/rsmarken)\
**Post date:** [December 5, 2009, 7:26pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/25 "2009-12-05T19:26:32Z")

</div>

[From Rick Marken (2009.12.05.1130)]

> If I understood you right, living organisms are likely to produce only one  
> kind of feed-forward control :

Actually, I think only kind of feed-forward control people do is when they act to have imagined, predicted results, like when they reach in the dark for their glasses. It’s not really control because people don’t consistently produce the intended result under these circumstances.

Much of what seems like feedforward control is, I would argue, actually feedback control of higher level variables. This can be seen in my model of catching a baseball ([http://www.mindreadings.com/ControlDemo/CatchXY.html](http://www.mindreadings.com/ControlDemo/CatchXY.html)). It looks like the fielder is predicting the future position of the ball and moving there by feed forward control. But, in fact, the player moves to the future position of the ball via feedback control of vertical optical velocity and horizontal displacement from the center of the field of view.

> BH : Oh. I almost forgot. I wanted to ask you something.
> 
> RM : I think the E.coli “reorganization” model of evolution can be  
> considered somewhat Lamarckian…
> 
> BH : Who’s oppinion is this ?

think this is reasonably good quick and dirty summary of the control model of evolution, as Bill described it some time ago.

Best

Rick

> **···**
>
> 2009/12/5 Boris Hartman [boris.hartman@masicom.net](mailto:boris.hartman@masicom.net):
> 
> –  
> Richard S. Marken PhD  
> [rsmarken@gmail.com](mailto:rsmarken@gmail.com)  
> [www.mindreadings.com](http://www.mindreadings.com)

---

<div class="post-metadata">

**Author:** ![Bill\_Powers1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/8e7dd6/32.png) [@Bill\_Powers1](http://discourse.iapct.org/u/Bill_Powers1)\
**Post date:** [December 5, 2009, 10:57pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/26 "2009-12-05T22:57:06Z")

</div>

> Bill !
> 
> BP earlier: It's important to define feedforward
> 
> BP earlier : Sensing the cause of a disturbance and anticipating  
> it's effects can sometimes improve control (feed-forward) but is by  
> no mean necesary and  
> usually it's not even possible. (B:CP, p. 48, 2005)
> 
> BH : O.K. please explain to me, what did you mean with this definition ?  
> Maybe you could add some examples

BP: That wasn't a definition, it was a statement that sometimes  
feedforward helps a little but feedback will work without it ("by no  
means necessary") and usually the cause of a disturbance can't be  
observed, or observed accurately enough, ("not even possible") so  
there isn't enough information to use as the basis of feedforward.  
When you drive a car in a crosswind, the wind disturbs the path of  
the car but it can't be seen. Most disturbances are first known as an  
unexpected change in the controlled variable. When that happens it's  
too late for feedforward to help even a little.

> BP earlier: The safest thing to do is trust the feedforward, but  
> verify it with feedback control to handle the many situations in  
> which feedforward fails to help or makes matters worse
> 
> BH : What did you mean with this relation between feedforward and feedback ?  
> Is this happening in the same control system ?

BP: I was thinking more in terms of the designer of the control  
system. If the control problem is so simple that you can use  
feedforward alone (like turning the streetlights on when a photocell  
shows that night is falling), you use it because it's cheap. If it's  
more complex, first you design a control system with feedback  
control; then you see if adding some feedforward reduces the errors  
enough to make the expense worthwhile.

> BP : As to other kinds of feedforward such as making plans or taking  
> precautions, those belong to higher-level kinds of control which involve  
> much slower changes and longer delays, and as Robert Burns said, often go  
> wrong or "gang aft agley". We try to anticipate problems, but the unexpected  
> happens anyway, and we just have to rely on our good old negative feedback  
> control systems to deal with the many uncorrected errors that are left when  
> the feedforward has done all it can do.
> 
> BH : Can you explain to me this relation between feedforward and feedback ?  
> Maybe some examples ?

The attached figure shows how feedforward control can be added to a  
feedback control system (feedback is in the lower part of the  
figure). The comments and arrows in red were added by me; the figure  
is from a control-system engineer, Dr. Douglas Cooper, and is used  
with permission. I show where added disturbances can enter that the  
feedforward doesn't take into account. The feedback part can correct  
most of the effects of all the disturbances, known and unknown, on  
the "process variable," which is the controlled variable.

Feedback is based on observing the variable that is to be controlled,  
comparing its state with a reference condition, and acting on the  
basis of the difference to oppose the effects of a disturbance.

Feedforward is based on observing other variables ("Disturbance")  
that will cause (through the box labeled "disturbance process") a  
future change in the controlled variable. The cause of the  
disturbance is sensed ("Disturbance Sensor/transmitter") and the  
sensory signal goes through some computing function ("Feedforward  
Computational Element") that applies an effect of the opposite sign  
through the remaining part of the output process to the controlled  
variable. The feedforward elements do not use any information about  
the present state of the controlled variable.

> BH : So the LAW says. Despite this, some smart guys build houses without  
> plans (only control).  
> You wrote it is feedforward ? Can you explain?

Building a house without any plan means building it without any  
reference condition. A plan specifies perceptions that are to be  
created by the actions of the builder; it does not specify what the  
actions should be, but only what their consequences should be. If the  
plan is on paper, it shows how the finished house will look; if it is  
only in the builder's imagination, it shows the same thing, but can't  
be checked by anyone else (which is why there are laws about having  
plans drawn on paper). A feedback control system is needed to make  
the actual perceptions match the plans, which are reference signals.

If a house is built without any plans, either on paper or in  
imagination, it could only be built by carrying out precalculated  
actions without observing their effects. That is feedforward. I doubt  
very much that a house could be built that way.

Best,

Bill P.

![feedforward.jpg](http://discourse.iapct.org/uploads/default/original/1X/ba98ad609b578aa18428b7a8ea8ffae8e687075d.jpeg)

> **···**
>
> At 01:10 PM 12/5/2009 -0600, you wrote:

---

<div class="post-metadata">

**Author:** ![\_Boris\_Hartman](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/_/e8c25b/32.png) [@\_Boris\_Hartman](http://discourse.iapct.org/u/_Boris_Hartman)\
**Post date:** [December 6, 2009, 10:44am UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/27 "2009-12-06T10:44:10Z")

</div>

Bill !

BP : Sensing the cause of a disturbance and anticipating it's effects can  
sometimes improve control (feed-forward) but is by no mean necesary and  
usually it's not even possible. (B:CP, p. 48, 2005).

BP earlier : That wasn't a definition, it was a statement that sometimes  
feedforward helps a little but feedback will work without it ("by no  
means necessary") and usually the cause of a disturbance can't be  
observed, or observed accurately enough, ("not even possible") so  
there isn't enough information to use as the basis of feedforward.  
When you drive a car in a crosswind, the wind disturbs the path of  
the car but it can't be seen. Most disturbances are first known as an  
unexpected change in the controlled variable. When that happens it's  
too late for feedforward to help even a little.

BH : So if I understand right when crosswind suddenly disturb controlled  
quantity, probably control of direction of a car on the road, sensed  
disturbance is opposed with negative feed-back alias sensed value of  
disturbance is compared to reference value in comparator (where the car  
should be). The difference is "error" to effector, and so onďż˝  
But in the moment the wind disturbs the course of the car, I'm pretty sure  
that I can't sense the effect of my output opposing the distrubance. As I  
experienced the sudden "wind" on my steering wheel I "reacted" much more  
faster than I could think. I was somehow surprised. I felt as stimulus went  
directly to output.  
I would say it was some kind of reflex reaction. Can you explain what reflex  
is from PCT view ? I mean is it direct connection between sensor and  
effector or feedback control system ? What's about the reaction of organism  
when it's burned with fire ?

BP earlier : As to other kinds of feedforward such as making plans or taking  
precautions, those belong to higher-level kinds of control which involve  
much slower changes and longer delays, and as Robert Burns said, often go  
wrong or "gang aft agley". We try to anticipate problems, but the unexpected  
happens anyway, and we just have to rely on our good old negative feedback  
control systems to deal with the many uncorrected errors that are left when  
the feedforward has done all it can do. The safest thing to do is trust the  
feedforward, but verify it with feedback control to handle the many  
situations in which feedforward fails to help or makes matters worse.

BH : Can you explain what you meant ?

Boris

---

<div class="post-metadata">

**Author:** ![\_Boris\_Hartman](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/_/e8c25b/32.png) [@\_Boris\_Hartman](http://discourse.iapct.org/u/_Boris_Hartman)\
**Post date:** [December 6, 2009, 10:54am UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/28 "2009-12-06T10:54:06Z")

</div>

Hi, Richard

RM : Actually, I think only kind of feed-forward control people do is when  
they act to have imagined, predicted results, like when they reach in the  
dark for their glasses. It's not really control because people don't  
consistently produce the intended result under these circumstances.

BH : Can I understand your explanation like : forming an imagined, predicted  
result in the heads of people as feedforward. Could we say that imagined,  
predicted result is "plan" for looking the glasses in the dark. For example  
: people with glasses who put it away somewhere and can't find it in the  
dar, form imagined picture where they could be. When picture is forming or  
formed, they start to search them in the dark, touching with hands so  
performing search with feedback alias matching the sensed effects of their  
own output with reference they formed in their heads on the basis of  
feedforward.

RM : Much of what seems like feedforward control is, I would argue, actually  
feedback control of higher level variables. This can be seen in my model of  
catching a baseball ([http://www.mindreadings.com/ControlDemo/CatchXY.html](http://www.mindreadings.com/ControlDemo/CatchXY.html)).  
It looks like the fielder is predicting the future position of the ball and  
moving there by feed forward control. But, in fact, the player moves to the  
future position of the ball via feedback control of vertical optical  
velocity and horizontal displacement from the center of the field of view.

BH : But how can we explain with your model, actual action of the fielder  
who is observing the flying ball and suddenly turn back to the ball and  
start running in some direction that probably only he knows why. So how can  
we describe with PCT means what is happening in fielder ? Could we say that  
on the basis of sensory data (retina) of vertical and horizontal velocity,  
fielder formed imagined, predicted result of where the ball will fall. So he  
simply turned into that direction and with imagined plan in his head he  
strats running so to match the sensed effects of his output with imagined  
(predicted) spot (reference) where the ball could fall.  
Well nice. It seems to me that you improved demonstration from the times we  
were talking about it.

RM : I think the E.coli "reorganization" model of evolution can be  
considered somewhat Lamarckian... think this is reasonably good quick and  
dirty summary of the control model of evolution, as Bill described it some  
time ago.

BH : Interesting destiny of evolution model, but why dirty summary. It's  
seems to me like realistic estimation of real value of that "control  
evolution theory". But anyway I agree with your "good quick and dirty  
summary", so I must say that I had bad feelings in that times, because I  
thouhgt that designing the evolution theory on the basis of bacteria is very  
slippery act. As far as I'm acquanted with this matter bacteria shows only  
through long term experiments Darwinian characteristics of selection. I  
think Bill would have much more chances with Amoeba as starting point of  
theoretical background for control evolution theory. But what's done is done.

Best,

Boris

---

<div class="post-metadata">

**Author:** ![Bill\_Powers1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/8e7dd6/32.png) [@Bill\_Powers1](http://discourse.iapct.org/u/Bill_Powers1)\
**Post date:** [December 6, 2009, 3:50pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/29 "2009-12-06T15:50:54Z")

</div>

[From Bill Powers (2009.12.06.0716 MDT)]

> BH : So if I understand right  
> when crosswind suddenly disturb controlled
> 
> quantity, probably control of direction of a car on the road, sensed
> 
> disturbance is opposed with negative feed-back alias sensed value of
> 
> disturbance is compared to reference value in comparator (where the  
> car
> 
> should be).

BP: This is a problem in the English language. I talk about it frequently  
but somehow it keeps coming back. “Disturbance” has two  
meanings. In the present example, one meaning is “wind force applied  
to car”. That is, the disturbance is a physical variable different  
from the direction of the car. It’s a force, not a direction. The driver  
can’t detect the force that the wind is applying to the car.  
The other meaning is “change in the direction of the car caused by  
the wind.” The direction of the car is a variable that the  
driver is controlling all the time. When a wind applies a force to the  
car, the car begins to move sideways. We say that there is a disturbance  
of the car’s direction, now using the word disturbance to mean the change  
in direction itself, rather than the wind force that is causing it. So we  
are using the same word, disturbance, to mean the cause (the force of the  
wind on the car) and the effect (the change in the direction of the car).  
It’s very important to be clear about which meaning you’re using. I  
suspect other languages have similar problems.  
Feedforward depends on sensing the _cause_ of a change in direction  
of the car, not the change itself. The cause of the change in direction  
is a force applied by the wind to the car. Sensing the change itself, the  
actual change in direction, is what a control system does, but a  
feedforward connection does not do that. That’s the main reason that  
feedforward is not a good way to achieve control. It produces actions  
blindly, whether they are needed or not, and does not correct them if the  
result is not the one desired.

You have to think very carefully about word meanings in talking about  
these matters. For example, you say "… crosswind suddenly disturb  
controlled

quantity, probably control of direction of a car on the road." That  
phrase is incorrect, because it says the controlled quantity is  
“control of the direction of the car.” The controlled quantity  
is not “control”, it is “the direction of the car”.  
So you should have said “crosswind suddenly disturb controlled  
quantity, probably direction of a car on the road.”

When you say it exactly right, it is clear that the wind force and the  
direction of the car are two different variables. The wind force is  
measured in kilograms, the direction of the car in degrees, or from the  
driver’s point of view the distance of car left or right from the center  
of the road, in centimeters.

Feedforward would help the driver by a small amount. If the driver could  
sense a sudden change in the wind velocity outside the car, or the force  
it is applying to the car, and react to it by turning the steering wheel  
before the car changes direction enough to see, he could keep the change  
in direction smaller than it would be if he based the initial steering  
effort on the change in direction. I am assuming that his reaction to the  
change in wind force would not have a longer reaction time than the  
reaction to the car’s deviations.

Normally, however, winds do not change their speed instantly; the speed  
rises and falls smoothly, so the wind force on the car would increase and  
decrease smoothly, not suddenly. This means that the sideways movements  
would be smaller and slower even if the final change in wind force were  
the same, and the normal feedback sensing would probably not be  
measurably improved by feedforward.

> BH: The difference is  
> “error” to effector, and so on.
> 
> But in the moment the wind disturbs the course of the car, I’m pretty  
> sure
> 
> that I can’t sense the effect of my output opposing the  
> distrubance.

That’s true, because the change in direction isn’t instantly observable,  
and your perceptual lag means you will not start turning the steering  
wheel for about 100 to 200 milliseconds even after you observe it.  
However, if you had a sensor that could detect a sudden change in the  
wind force, you would have a similar reaction time before you started  
turning the wheel, so the feedforward wouldn’t help much, unless the  
feedforward sensor were located 20 or 50 meters upwind so it could detect  
a change in the wind before it got to the car, or the car got to it. That  
would be hard to accomplish, and it wouldn’t be as accurate as measuring  
the actual force on the car due to the wind. You’d have to compute the  
force on the car using aerodynamic equations, and that would probably  
take somewhat longer than just waiting to feel and see the car’s change  
of direction.

> As I experienced the sudden  
> “wind” on my steering wheel I “reacted” much  
> more
> 
> faster than I could think. I was somehow surprised. I felt as stimulus  
> went
> 
> directly to output.I would say it was some kind of reflex reaction. Can  
> you explain what reflex is from PCT view ? I mean is it direct connection  
> between sensor and effector or feedback control system ? What’s about the  
> reaction of organism when it’s burned with fire ?

You’re forgetting about levels of control. The lowest level control  
systems react much faster than higher systems, especially systems  
involved in “thinking.” In tracking experiments, we find that  
the overall delay in reacting to a visual error is about 0.13 seconds,  
130 milliseconds. That is much faster than you can do if you’re trying to  
think about what you’re doing before you act. Yes, this is what has been  
called a “reflex,” which is actually a low-level control  
system; see Fig. 7.3 in B:CP for a diagram of a spinal reflex (actually  
several of them – tendon reflex and stretch reflex). Reflexes are just  
low-level control systems.

I think you’re also forgetting that just “reacting” is not  
enough to achieve good control. You have to react in exactly the right  
direction, exactly by the right amount, and exactly for the right length  
of time. Control depends on quantitative accuracy, not just qualitative  
“movements.”

> BP earlier : As to other kinds  
> of feedforward such as making plans or taking
> 
> precautions, those belong to higher-level kinds of control which  
> involve
> 
> much slower changes and longer delays, and as Robert Burns said, often  
> go
> 
> wrong or “gang aft agley”. We try to anticipate problems, but  
> the unexpected
> 
> happens anyway, and we just have to rely on our good old negative  
> feedback
> 
> control systems to deal with the many uncorrected errors that are left  
> when
> 
> the feedforward has done all it can do. The safest thing to do is trust  
> the
> 
> feedforward, but verify it with feedback control to handle the many
> 
> situations in which feedforward fails to help or makes matters  
> worse.
> 
> BH : Can you explain what you meant ?

That was a long paragraph dealing with several subjects. Which one are  
you asking about? Basically, I’m agreeing that we do make plans and try  
to be ready to meet disturbances before they happen. But such plans work  
only if there are no unexpected disturbances or changes in the  
environment, which is not true very often. It’s good to be prepared, but  
you also have to be ready to handle errors as they arise, by normal  
feedback control. Even if feedforward does reduce the effects of some  
disturbing events, it can never counteract them accurately, so the  
feedback control systems still have to act. Would you say that your fast  
reaction to a sudden wind disturbance was as accurate as your normal  
steering process?

By combining feedforward with feedback, as in the diagram I posted  
yesterday, it is possible to generate some approximately correct actions  
during the time before the negative feedback system starts to act  
(assuming that the feedforward sensor has less lag than the feedback  
sensor). But that’s about the only kind of advantage one gains from  
feedforward, as far as I can see. And it works only in the few cases  
where the external cause of a change in the controlled variable can be  
sensed with any degree of accuracy and without any normal sensory  
lags.

Best,

Bill P.

> **···**
>

---

<div class="post-metadata">

**Author:** ![Bill\_Powers1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/b/8e7dd6/32.png) [@Bill\_Powers1](http://discourse.iapct.org/u/Bill_Powers1)\
**Post date:** [December 6, 2009, 4:10pm UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/30 "2009-12-06T16:10:37Z")

</div>

[From Bill Powers (2009.12.06.0857 M\<DT)]

> BH (to Rick Marken) : But how can we explain with your model, actual action of the fielder who is observing the flying ball and suddenly turn back to the ball and start running in some direction that probably only he knows why. So how can we describe with PCT means what is happening in fielder ? Could we say that on the basis of sensory data (retina) of vertical and horizontal velocity, fielder formed imagined, predicted result of where the ball will fall. So he simply turned into that direction and with imagined plan in his head he strats running so to match the sensed effects of his output with imagined (predicted) spot (reference) where the ball could fall.

You really should start this question by saying "I don't believe what Bill Powers and Rick Marken have said about how fielders catch balls." And then you should say why you don't believe it. Just ignoring what we have said doesn't work as an argument; it just leads to extreme frustration.

What I have said is that fielders who turn their backs on the ball and run to where they think it will come down fail to catch the ball most of the time. I have watched this happen many times during television broadcasts of baseball games. We tend to remember miraculous catches of this kind precisely because they are so rare.

A fielder take his eyes off the ball only when it is clearly going to pass over his head and he can't run fast enough to keep its image under control (that is, stationary or slowing rising). Usually when the fielder does this, he does not catch the ball because his estimate of where it will come down was not accurate enough. He runs too far or not far enough, and not quite in the right direction. This is why coaches scream at their players, "Keep your eye on the ball!" A PCT coach would yell, "Dammit, keep the loop closed!"

Best,

Bill P.

---

<div class="post-metadata">

**Author:** ![\_Boris\_Hartman](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/_/e8c25b/32.png) [@\_Boris\_Hartman](http://discourse.iapct.org/u/_Boris_Hartman)\
**Post date:** [December 7, 2009, 7:30am UTC](http://discourse.iapct.org/t/ashby-and-control-theory/1324/31 "2009-12-07T07:30:25Z")

</div>

Bill !

BP (in name of Rick Marken) : You really should start this question by  
saying "I don't believe what Bill Powers and Rick Marken have said about how  
fielders catch balls." And then you should say why you don't believe it.  
Just ignoring what we have said doesn't work as an argument; it just leads  
to extreme frustration.

BH : So you know what others think. Interesting. And in the name of Richard.  
You probably know what he is thinking, too. And what did you say that I  
ignored ?

Whatever you said are yours believes, yours perceptions of the world  
outside. If you feel frustrated it's your problem.  
It' not a problem whether I believe you or not. It's about what I think  
about reality and what is true for me and what's the difference between our  
perceptual worlds (analogs of physical world). And I see "reality" a little  
different from you. I have my controlled perceptions. I don't believe that  
you want us to be "slaves" of your perceptual world. Remember David and his  
10 WTP commandments ? You pointed out that we are all control systems and we  
act like control systems.

Your model has to prove it's validity in every perceptual situation and in  
every human behavior or behavior of living control systems if it's to be  
believed 100%. But it's not for now. PCT is a good model (probably the best  
nowadays) but it need more improvement, like this one in baseball,  
evolution, learning, etc. There's enough mystery left for all of us to  
discover.

So it's not a problem whether I believe you or not, but do I percept reality  
as you do. Are my experiences equal to yours. With your theory you showed  
me, that reorganization is creative process and people are creative living  
organisms. So control activity of the brains are showing diversity.  
I understand you want to control all people as that seems to be one of  
tendency of control systems. It showed it's "face" through all the history  
of living beings, and is showing it's "face" today through terrorist  
attacks, hierarchy of organizations and also through your activity. All  
control system probably have that tendency.  
But we are social beings. We are also different in control. And I think we  
must know how to direct our tendency of control of other people with all  
forms of interpersonal control or nocontrol in the shared environment. You  
and McClelland showed that very clearly.

BP : What I have said is that fielders who turn their backs on the ball and  
run to where they think it will come down fail to catch the ball most of the  
time. I have watched this happen many times during television broadcasts of  
baseball games. We tend to remember  
miraculous catches of this kind precisely because they are so rare.

BH : Exactly. That's the way how real players work. That's real life. Why is  
that so ? Why they fail to catch the ball ? What in that situation make them  
behave as they do ? Why I as a player failed to work on "written" principles  
what is right ? That's what I'm interested in, not whether I believe you or  
Richard. We have all sensors and we all can perceive playing situations or  
real life situations on our own way and we all have a little different  
brains who saw it a little bit different. So probably there is no objective  
reality. There are probably only our perceptions and exchange of our  
perceptions. I think that's also one of the main believings of VonFoerster  
as Arthur already pointed on him.  
BP : A fielder take his eyes off the ball only when it is clearly going to  
pass over his head and he can't run fast enough to keep its image  
under control (that is, stationary or slowing rising). Usually when  
the fielder does this, he does not catch the ball because his  
estimate of where it will come down was not accurate enough. He runs  
too far or not far enough, and not quite in the right direction. This  
is why coaches scream at their players, "Keep your eye on the ball!"

BH : Screaming on players doesn't mean much. Such a coaches are bad coaches.  
They should solve this problem on the training. In the game it won't help  
much. That's probably another characteristics of control systems in real  
world. I was a player in national leagues for a long time. And I know what  
is to experience playing situation and what is to scream from side.  
But I think that we are coming slowly to the solution of our problem. What's  
the combination of feedforward and feedback in real behavior of real people  
(all living control systems) whom we describe as HP control systems who try  
to maintain their essential variables in their limits with reorganization  
and various behaviors.

BP : A PCT coach would yell, "Dammit, keep the loop closed!"

BH : Or : "Dammit, switch from 'feedforward mode' to control mode".

Boris

[Previous page](http://discourse.iapct.org/t/ashby-and-control-theory/1324.md?page=1)