Experiences of positive feedback

Evadeh · December 3, 2021, 10:29am

I wrote this on my site (in Dutch: Stop: je maakt het alleen maar erger! – Perceptual Control Theory & Method of Levels). English version: Stop: you’re only making it worse! – Perceptual Control Theory & Method of Levels

Curious what you think: do I get the mechanics right? Does this resonate with your experience? I should note the actual experience is not just positive feedback, it’s also conflict, other ways to try to control, reorganization. But the idea that things turn bad really quickly, is interesting from the positive feedback point of view.

Positive feedback: stop it, you’re only making it worse!

“I only give negative feedback” is printed on a shirt my friend gave me to wear to my classes. A nerd joke, on a nerd-style T-shirt, which doesn’t fit well, obviously. But what it says is correct. I only give negative feedback. I read my students’ reports, notice where things go wrong, and provide feedback with the aim of reducing the error in the next version. That is the principle of negative feedback: you do what you can to reduce the error, the difference between how it is and how it should be. The negative feedback loop is how controlsystems work and how all the controlsystems we have - or rather which we are - continually reduce their error and achieve their goals.

I always get a hard-to-suppress, cringe-like feeling when I hear colleagues talking about how important it is as a teacher to give positive feedback. “Don’t do it!” I’d cry. But I hold back: even without wearing a nerd shirt, I’m all too often the one who knows better. But I’m right anyway: it’s a stupid plan to give positive feedback.

Positive feedback is the reverse of negative feedback. Where negative feedback makes the error smaller, positive feedback makes the error larger. Everything you do increases the difference between how it is and how it should be. That creates a huge problem in no time.

You know that shrill howling sound of a microphone at a conference? The speaker speaks into the microphone, and instead of hearing the speaker, you hear an increasingly loud, higher-pitched sound that makes all further communication impossible. This positive feedback is caused by the fact that the output of the microphone, the amplified sound from the speakers, is fed back as the input of the microphone. The microphone amplifies its own amplified sound in a loop - it amplifies and amplifies and amplifies. The first thing you do is pull the plug so that the sound stops. Next, you need to make sure that the microphone only picks up the speaker’s voice, and not the amplified sound. By connecting the plugs better, or by not standing next to the speakers while you speak.

Figure 1. Simplified image of a control system.

Positive feedback loops are actually always undesirable. After all, in which situation do you want a perception to spiral out of control? Even if you fall into a positive feedback loop of happiness, be careful not to become manic. Positive feedback means loss of control: with positive feedback you will never succeed in keeping a perception at a desired value.

Yet you regularly experience positive feedback. Some examples:

If you have a rising sense of panic, and you try not to panic, this trying not to panic often makes it worse. You notice that it doesn’t work, that makes you panic more, which makes you less successful in not panicking, so that your goals get even further out of sight and you panic about them even more.
When my son is bored, I often say: go play with Lego. Draw something. Practice guitar. But all those things make it worse: he doesn’t feel like it, the feeling of boredom only increases. Also without my ‘help’, anything he invents with the aim of reducing boredom only makes him feel more annoyed because he finds that he doesn’t feel like doing anything.
I had a conflict with someone I worked with. It wasn’t really about anything important, it seemed to me mainly incomprehension and miscommunication. I tried to save the collaboration, but everything I did to make sure he understood me (explain, argue, inquire) only made the misunderstanding worse. The more vigorously I argued, the greater the misunderstanding and thus the problem became.

In all these situations you can do the same as with the microphone: Pull the plug, to make sure you don’t experience that self-amplifying error signal. You can stop the panic by taking a cold shower; stop the boredom by sending your child into the garden, stop the quarrel by ending the collaboration.

But that still doesn’t get you to the root of the problem. To do that, you need to know what started that positive feedback loop. Just as it happens with a microphone because the microphone picks up the wrong signal, you get positive feedback loops - those wholly out of control situations - as you try to control the wrong perception.

If you’re trying to control panic, that means your controlled variable is the panic level . You want to get that down to a low level of panic. However, your current level is high. Therefore you experience error, and that error is experienced as the feeling of panic. You’re trying to control something that shouldn’t be under control itself (the error signal), so that’s why it doesn’t work. It helps to look for something that you can control. Your breathing. Your sense of sensations in your arms or legs. One of the other many goals you are trying to achieve. Have a cup of tea, walk around the block. That’s how you eventually end the panic.
If you try to control boredom, and it doesn’t work, you feel even more bored. You can’t control that bored feeling yourself. The feeling itself is the error that arises when it is not possible to gain control over certain important perceptions. The problem will resolve itself if you succeed in pursuing a different, attainable goal. Pet the cat. Feed the cat. Do something and notice that you feel like it.
Now when I look back at the problem with my colleague, I can see how his plan to pull the plug on the partnership is an attempt to stop the positive feedback - it just kept getting worse. By stopping the collaboration, the cycle of increasing misunderstanding stops. But when we stopped there, the goals we had together have also been lost. If this happens to you, and there’s still room to save the collaboration, it’s helpful to get the system back on track by controlling the controlled variables. The signal we controlled - the misunderstanding - is apparently itself a form of error. It is the result of another control process that is not going well. So that’s what you have to look for: what are you trying to control that doesn’t work? What goals do you want to achieve? Once you have that clear and the technology is back in order, you can plug it back in.

You recognize a situation of positive feedback because it is getting worse. Conflict keeps you stuck: you don’t achieve your goals, no matter how hard you try, but it doesn’t get out of hand either. The condition remains the same regardless of the effort. When it comes to positive feedback, things spiral out of control, and everything you do seems to make it worse.

In a conflict, it is important to stand still, focus on the relevant aspects of the problem and let the reorganization process do its work. Then you can find a way to move on from higher goals. You also have to think about positive feedback. First by acting quickly and pulling the plug. Then by checking what you’re trying to control, and finding out that you’re trying to control something that you can’t control at all. You try to keep the error at a certain level, and it doesn’t work that way. The error - the unpleasant feeling, the feeling of panic, the incomprehension between two parties - is itself part of the control system, but does not give you a foothold to improve control.

You have to look for a goal that you can achieve in order to regain control. If your microphone is buzzing, make sure you’re getting the right input: the voice, not the amp. In case of panic: make sure you focus on a small, achievable target, then you will regain control. In case of boredom: keep messing around until you find something you feel like. If there is pain in a relationship, misunderstanding and frustration, talk about what is important to both of you and focus on that.

Because I ended up in a positive feedback loop, I noticed how little control I had when things went wrong. But I also learned how this type of control works. If you try to keep the error signal under control without having an eye for the higher goals that matter, then things go wrong. That makes it important to give all emotions (error signals) room to do their job in difficult situations. If you try to keep the error small, any deviation from it is a sign of failure, increasing the error again. So focus your attention on the higher goals, and give your feelings the space to show you the direction.

wmansell · December 3, 2021, 12:00pm

Love it!

bnhpct · December 3, 2021, 5:42pm

A lovely use of metaphor for guidance how to stop social polarization from escalating!

Gregory Bateson introduced the term schismogenesis to anthropology in the 1930s. Although a bit abstract (his readers would have had particular communities and culture-contact situations in mind), it is well worth reading.

What makes the feedback negative is the comparator within the recipient.

What people think is negative or positive about the perceptions that you bring to their attention is so for them because of their emotional judgement perception of it. An observer may also perceive your intervention as negative or positive according as how they imagine the intended recipient perceives it.

The control mechanisms for emotional judgements were featured in my presentation in October.

Technical nit, not relevant to your main point here. The howl of feedback from speaker to microphone is because of the lag between transmission from the speaker and receipt to the microphone. Two more or less identical sound waves are separated by a short gap.

In the image you can see that where two peaks of a sine wave are separated by a short gap, there is a brief drop in air pressure between the duplicate peaks. This transition in air pressure of course persists throughout the extent of the sound signal (whether a sine wave or a more complex combination of sine waves). The transition in air pressure itself constitutes a sine wave of air pressure which we perceive as a high-pitch howl.

This is irrelevant to your point about what you can do to interrupt the feedback loop and stop social polarization from escalating.

richardhpfau · December 3, 2021, 9:27pm

Eva, A great article! I think that I’ll order one of those negative feedback t-shirts. I discovered that they are available by searching online using these terms: t-shirt, negative feedback.

MartinT · December 3, 2021, 10:58pm

Eva, you describe what I called “The Bomb in the Hierarchy” (Section 6.5 in PPC). In the context of interpersonal (mis)communication, I began to address it in Sections 25.9 and 25.10 (the latter, I find, is as yet largely unwritten). The question addressed in your failed interaction with your colleague is "What component of the protocol, at what level up and down the hierarchy of protocols, and in what supporting protocol are the channels that normally lead to negative feedback blocked?

Every instance is different, but one issue you highlight is that your colleague eventually changed a higher level reference for perceiving you to be a valuable communication partner to the reverse. Why? Something else was being controlled by one of you, at high level, and the error in that perception was growing. Self-image, perhaps? Who knows, in any specific instance. Anyway, the solution to the Bomb Avalanche is to shore up the bottom, though it is always easier to start an avalanche than to clean up afterwards.

Evadeh · December 4, 2021, 1:36pm

Thanks all, that’s interesting and nice replies to my blogpost.

@MartinT thank you for your pointers to PPC. So if I understand you right, what I was expecting with my colleague was a protocol, in which usually one partner voicing some error over how things are going, leads to the other listening and finding a way to solve that error together. In this case, that protocal was left immediately. Me voicing my discontent was interpreted as a sign of treason, which didn’t fit the collaboration protocol, obviously. Interesting to think of what exactly blocked the channels (can I interpret these as ‘contact’?) that usually worked in our conversations.

I remember clearly being disturbed by the different interpretations we had of our conversation. I wondered: what does it mean for control between two control systems, if they have such a difference in interpretations? Both of us had very different higher level control systems actively maintaining control over what we both interpreted as ‘truth’. Our truths however, were very different.

There’s a parallel with this example and polarisation in society, right? It seems that different people have very different truths and interpret the same situations in totally different ways. How do you get these people to interact and broaden their views, to get a better and more flexible way to control?

So to shore up the bottom would mean what exactly? Can you elaborate a bit, Martin, on your view how to get out of a reinforcing feedback loop once you find yourself in one?

MartinT · December 4, 2021, 7:10pm

Yes, I think the parallel is exact, interchanging Collectives for individuals.

I don’t address how to solve the problem, but I do have a view of the “different truths” leading to conflict in Section 29.7. But I suspect that to make sense of what I say there, you probably would need to read something about the crumpling metaphor and category perception, meaning at least skimming Chapters 18 and 19.

Fixing the conflict is a problem to which there can be no pat answer, because it hinges on finding the variable(s) each of you are controlling. Both appear to me to be using control of a truth in an autocatalytic homeostatic loop (Chapters 14 and 15), but the coherence of each loop depends on the two conflicting truths. What might those truths do in the two stable loops? One can’t tell from the description, and a solution probably involves the Test for the Controlled Variable, which is hard to do when communication has been severed.

Your first paragraph needs a forensic dissection.

That’s not the usual way a protocol is initiated. By the way, protocols do not need to be honest or between friends. Military opponents use them for deception and to gain advantage, con-men use them to get you to do something to your disadvantage and to their advantage.

A protocol is initiated when X wants to control some variable and can do so if X’s perception of the effect on the environment of an action of Y is controlled to some reference value, for example if X can get Y to lay a plank across a brook, or get Y to accept X’s belief about the result of some argument. There are two possible problems for X, one is if Y explicitly controls for something that would conflict with what X wants, or if Y does not control a perception that would be disturbed by what X says or does. The other is if Y cannot interpret X’s overt actions in a way that disturbs a perception X wants Y to correct by using the desired action (saying Y understands, for example).

“Error” in protocol can mean a wide variety of things. I believe that a full-fledged protocol, as described in Appendix 8 (Volume 4), has 19 different controlled perceptions, any or all of which might have errors that would need fixing, such as whether the environment is too noisy to allow accurate communication at the moment using the channel being used.

Eighteen of the 19 controlled variables of a protocol are what I tentatively call “syntactic”, taking the place of function words in written text. You could call them “functional”, since they all deal with the level of belief in some perception of what the other believes, such as whether X (the initiator) perceives that Y believes that X believes Y speaks Armenian , or has enough background knowledge to understand what it is that X wants to know or wants Y to believe.

Error in any of these 18 controlled functional/syntactic perceptions is, as in any hierarchic control structure, by sending output to the reference level of one or more lower-level controlled perceptions. Since the error in this case always involves perception of something about the other, the correcting action usually involves initiating a lower-level protocol. Protocols form hierarchies just as do simple perceptions (I don’t think I put it this way in PPC, but maybe I should).

Your use of the word “Treason” suggests that in some way you disturbed some element of your colleague’s self-image, making him or her perceive that by admitting the effect of your argument, he or she would be diminished in some way unacceptable. Breaking off communication suggests that they perceived the strength of your argument, but that resulted in a conflict not only because of the self-image problem, but possibly also because of the requirement for the colleague’s truth in completing the homeostatic network of concepts. What else might your colleague believe that you don’t?

Too much speculation, and I imagine the same would be true of my suggestions for

Firstly, what are you perceiving, from what viewpoint when you “find yourself in” a reinforcing feedback loop? Secondly, the situation is different if the initiating positive feedback loop has had no further repercussions or if it has initiated a small or large avalanche or landslide (not something the Netherlands landscape ever has, I presume). The results of avalanches are much harder to clean up than would be the fallen rock that could have started an landslide but did not. In effect, the avalanche stopped only because of reorganization of the terrain (snowpack, perceptual control hierarchy).

Back to my starting point: “Firstly, what are you perceiving, from what viewpoint when you “find yourself in” a reinforcing feedback loop?” If it’s not what you want to perceive, you are controlling whatever it is that you are looking at from some higher level that allows you to perceive the positive feedback as it happens (in retrospect is different). I can’t tell what that is, except that because you talk about it, it must be conscious, and a problem to control.

In my recent thinking, that puts it into the domain of Friston’s Predictive Control, basically “suck it and see” what might work. Are you controlling for maintaining good relations with the colleague? Is your colleague controlling for perceiving him/herself to be better than you? Questions like those come to mind, and I’m sure that you can think of lots more that might suggest different approaches.

Sorry I can’t be more helpful. But I thank you for bringing up something I have not considered in respect of protocols.

Martin

Evadeh · December 5, 2021, 12:18pm

Warren, you shared this video and paper with me on the role of gain in bipolar disorder.
Try this at 37 minutes,

From the video: Gain is the error sensitivity. The higher the gain, the more effort you’d put in to reduce the error. If you increase the gain, the goals are met quicker with a tighter fit. This reflects the state of mania, when people are witty and intelligent and fast. The problem is that if you keep on ramping the gain, the system may lose control.

I don’t yet get exactly how the sudden loss of control occurs, from that video. But it may well be we are talking about the same subject?

Do I understand right, that when you increase the gain, you start to control for a very low level of error to exist? Like wanting to be very precise, making no mistakes and putting quite some effort in that. This seems like quite sensible behavior to me: if for example I know I’ve got only one chance of convincing someone, I’d prepare my arguments a lot better and go over my proposal many more times than if I know that this one try won’t matter that much.

Where does the overshoot come from then? Why is upping the gain a problem? Isn’t that the positive (reinforcing) feedback effect? If the manic person tries to increase his manic state, believing that this state of mind solves all his problems, that would mean a positive feedback loop. When he notices that he gets more energy, that’s increased control (he wants energy to increase), which in turn increases energy and positive mood. Right? So the problem is maybe not the gain, but his controlling the increase of positive mood, which leads to a positive feedback loop.

What do you think?

Evadeh · December 5, 2021, 12:29pm

@MartinT
Thanks again for pointing me to chapters in your book. I see that I can’t use your language (protocols) lightly, without mistaking them for something not exactly what you mean. I don’t want to speculate too openly about this particular case, but your ideas surely help me make sense of what happened.
We haven’t got avalanches over here. But floods happen, and that’s sort of the same problem :). Thanks for clarifying!

bnhpct · December 5, 2021, 4:35pm

Let me add more about the emotional dimension of this.

You’re identifying the experience of ‘panic’ as the error signal. What is panic? It has two components (1) perceptions of the situation that you are panicked about, especially controlled variables that you are controlling poorly, and (2) somatic sensations resulting from control through the somatic branch of the hierarchy preparing the body to produce behavioral outputs as called for by the behavioral branch. The somatic branch splits from the behavioral branch at lower levels, perhaps below the Event level.

This is Bill’s latest draft of a model of emotion.

The somatic components of emotions and other higher-level (cortical) perceptions are usually not amenable to conscious control. I say ‘usually’ in two senses. Direct, conscious control of somatic sensations can be learned (viz. the more advanced practices of meditation, yoga, etc.). And relatedly, somatic sensations can be controlled indirectly by controlling somatic actions that affect variables that are also controlled by corporeal homeostatic feedback loops, as illustrated here:

The excitatory signals to the brainstem that arouse the body for action are outputs of the limbic systems. The limbic systems control perceptual inputs much faster than higher-level systems can construct the perceptions upon we place our awareness, and signals from limbic systems bypass intervening levels, providing input at cortical levels well above the somatic/behavioral divide.

Those limbic ‘snap judgements’ about the situation are among the inputs from which those higher-level perceptions are constructed, along with signals that more slowly make their way up the hierarchy. ‘Panic’ is an example of a higher-level perception of the sort that we call emotions (as distinct from ‘feelings’, which are somatic sensations).

This version of the diagram shows how somatic or ‘body’ variables B and limbic ‘snap judgements’ J bypass intervening levels and are combined with environmental input E at higher levels. It suggests that limbic systems do not control their inputs from the environment directly by outputs to muscles, but only indirectly, by excitatory outputs to the brainstem that affect conditions in the body and by signals sent up for the behavioral branch to control. Not shown is how they can also control behavioral responses ‘quicker than thought’ e.g. to prevent something toppling.

Limbic systems also receive perceptual signals that are constructed at higher levels. This can create a positive feedback loop through the limbic systems, which is an important part of constructing the cortical perceptions that we call emotions. We ignore the ‘feelings’ until they become sufficiently distressing and positive feedback generates an emotion that cannot be ignored. I have referred many times to research showing that talking about feelings slows the positive feedback between cortical assessments and limbic judgements. Attention goes to the ‘word pictures’, the scaffolding provided by language gives a standpoint off to the side which is unaffected by gut feelings and other corporeal sensations, and instructions about e.g. breathing, moving feet against the floor, etc. can be remembered and acted on. I have also talked about how attention to sensations in the body (cultivated in a form of vipassana meditation) forestalls their escalation to distressing levels and the clouding of perception by prejudicial emotions.

Attention goes to perceptions that must be controlled (not low gain) but are not being controlled well. By “attention goes” I mean that mobile or dirigible input functions (eyes, hands, etc.) are turned to receive those inputs most strongly, in preference to other sources of input. Those are the inputs to the limbic systems as well as to the behavioral branch. That is why attention is associated with error. The limbic system perceives the vial rolling off the edge of the table or the toddler stumbling, the limbic signal goes to the behavioral branch, and the hand reaches before any conscious decision to intervene.

(There are also in-hierarchy changes that are analogous to directing the eyes, skin sensors, etc., which somehow change the strengths of perceptual input signals or shift from observation mode to control mode by means that I at least do not understand. The analogy of attention to a shifting field of vision is alluring, but really no more than the old homunculus fable with all its problems of implementation and infinite regress.)

The limbic systems may also receive somatic signals, and that would be a more difficult potential path for positive feedback. This is not shown either.

It is important to distinguish error signals from the effects of poor control. Distressing sensations in the body and undesired emotions are effects of poor control, perceptions that are constructed by pathways mediated by the limbic systems.

MartinT · December 6, 2021, 4:27pm

Continuing my thread with Eva on protocols, I have added a few paragraphs in the beginning of Section 25.1, as follows:
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
Nineteen control loops may seem to promise a very complicated motif, but like most motifs that appear many times in normal behaviour, its structure is not very complex at all. Eighteen of the control loops form six groups of three, three groups performed by the initiator of the protocol, three by the “continuer”. Each group of three is based on the levels of belief about one of three different propositions.

• P1: That the Continuer has understood the Primal Message.
• P2: The quality of the communication process is sufficient for an adequate interpretation.
• P3: That it is not worth continuing to transmit this message.

P1 mentions a “Primal Message”. Each instance of a protocol has a Primal Message, which is the communication analogue of the error in the perception whose control required the Continuer to do something. If the Continuer has understood the Primal Message, she perceives what the Initiator wanted to achieve, the difference between the way the Initiator believes the World is and the way he would like it to be. The Primal Message should not be confused with another term we will use: the “Primary Message”, which consists of the overt actions, including language, that the Initiator initially uses to describe the Primal Message to the Continuer.

If we call a generic proposition “P” then each group of three propositions in each partner has control loops for belief perceptions with these three recursive reference values: (1) Self believes P, (2) Self believes Other believes P, and (3) Self believes Other believes Self believes P. When all three beliefs match their reference values of “believed” in both partners for any particular proposition P, no further action is taken with respect to that proposition unless some external disturbance affects the relevant belief perception.

The 3x3x2 pattern of three propositions at three levels of recursion in two partners defines 18 of the 19 control loops. What about the last one? The Initiator started the protocol because there was some perception that for some reason the Initiator thought was more easily controlled if the Continuer helped either by a physically effective action or by displaying a change in their understanding of the Primal Message. The protocol is the means for the Initiator to control that perception. When the Initiator uses the protocol, the Continuer takes the place of an atenfel, a component of the environmental feedback pathway in the control loop of the to-be-controlled perception.

In normal everyday conversation, both partners already believe all three recursions: (1) A believes, (2) A believes B believes, and (3) A believes that B believes A believes. Since “believes” is the reference level for all of them, no action is required. But if a loud noise interrupts the conversation and continues, at least one of the partners is likely to perceive their own inability to understand the other, or at least to question whether the other can understand them. In this case, at least one will act to correct the error, perhaps by signalling that they might move to a quieter place. That signal initiates a new “supporting” protocol as the action to reduce the error in one or more of the P2 control errors.

The supporting protocol is to a protocol hierarchy just what a lower-level perceptual control loop is to the perceptual control hierarchy. In fact, since a protocol is simply one way of controlling a perception when it is different from its reference value and the hierarchy inside the individual has no atenfels for control of that perception, a protocol hierarchy can be seen as a perceptual control hierarchy spread across two individuals. The parallel is not exact, because although the perceptions controlled in the 19 control loops of a protocol all require parts of their feedback loops to pass through the shared environments of the two partners, the feedback loops of the protocol itself are within the partners with no reference to how they are physically implemented.

So long as both partners control for all three recursions of P1 to come to their reference values, they will maintain error in P3 and probably its recursions, continuing the conversation. But always there is a possibility that one of the partners comes to believe P3 and commits an action that has been called an “Abort”. One possible reason for an Abort action could be that some controlled variable that contributed to the protocol being initiated has taken a value near enough to its reference value that the protocol action (sending the Primal Message) is no longer necessary. Perhaps a higher-level protocol has had all its P1 controlled beliefs come to their reference values and has ended, no longer requiring this supporting protocol to complete its work in getting its own Primal Message understood. In effect, the Abort in this protocol is performed by the Continuer saying “I get it now”, though of course at different levels if would not be done that way. Simply continuing the conversation that is ongoing at a higher level is often enough.

Protocols can fail, meaning that they are aborted with P1 still in error. P1 will not be in error if the higher-level protocol does the equivalent of saying “I get it now”. In that case, the protocol, though aborted, did not fail. It failed if it was aborted by one of the partners because they perceived that it was not reducing the error in whatever perception they were controlling by using it. The same effect might be if a person quit using one method to control a perception such as driving a nail into a rock when the hammering had no apparent effect on the nail or the rock.

A protocol is a way of communicating with reasonably high confidence a meaning in the form of a message from one entity to another. Either entity could be a human, a non-human living control system such as a dog, or might even be electronically embodied in silicon. The protocol structure was initially developed for human-computer interface design, and was later found to be applicable with minor extensions to the symmetric human-human situation.
The form of the protocol is the same in every case, but we will develop it in a context of inter-human communication. It is here, rather than in the directly observable physical actions, that error reduction is the important point. The one with the information to communicate wants to perceive that the other has understood the information. The other’s physical actions either succeed or fail in bringing the perception-reference difference within tolerance bounds.
The communicator of that message may be the Initiator who asked for the information, or who provided it because they wanted to perceive the Continuer to understand it (a reference value) and perceived that the continuer did not know it. Either way, the message has been passed (in a cooperative protocol as opposed to a deceptive one) when both partners perceive that it has been understood.

Often, especially at lower protocol levels, only two of the 19 possible controls result in any observable action, while all the others are already at their reference values. The performer of the first of these two loops we call the Initiator, who I often personify by assigning a name starting with “I”, such as Isaac or Irene. Using a “Continuer” (who we might call Carla or Charles, for example) the Initiator controls some perception that is in error. The Initiator therefore acts in a way intended to reduce the error, by disturbing some perception controlled by the Continuer so that the continuer’s control action reduces the error in the Initiator’s perception. This pattern, between mother Cora and baby Ivan, is shown in Figure 22.2 (lower part).

Here is a classic example of a completed basic protocol, in which only one party uses language, but each party performs some action:

Irene (in a wheelchair): Please would you close the window.
Carlo closes the window.
Irene: Thanks.

When any mechanism works smoothly and well, it may be hard to see anything of how it works. In this case, we have a perfectly ordinary exchange, in which Irene asks Carlo to do something for her and he does it easily. How many times a day do we engage in such interactions? “Please pass the salt”, “Hold the door for me a moment”, or a thousand other things that pass without a second (or even a first) thought, and are done without discussion.

We will use this simple interaction to probe a little deeper from a PCT viewpoint into what might be happening in such everyday interactions. We ask what each participant might perceive of the other’s behaviour that disturbs a controlled perception in ways that result in acts that can be perceived by the first, creating feedback loops that correct different kinds of problems in the interaction. Our answers are certainly not definitive, but they may be suggestive.

Depending on the circumstances, the same “close the window” protocol might have been performed without either Irene or Carlo using language. Irene might, for example, have been able to pass her Primal Message by simply pointing at the window she wanted closed. The form of the protocol is independent of the way in which its functions are instantiated. As we shall see, all of its functions can be described as the control of one or more of the 19 controlled perceptions in the motif.

Before Irene asked Carlo to close the window, she had a reference value to perceive it to be closed, whereas she perceived it then to be open. That error was her “Primal Message”. When he closed the window, the error vanished, and the protocol was complete. Always, protocols are completed when the initiator’s Primal Message (the error) is corrected and both participants perceive that to be the case[1].

When Irene acted, Carlo perceived something that disturbed some perception he controlled, perhaps a perception of Irene’s state of happiness. Irene relied on his control of some such perception, in the same way that baby Ivan’s use of their private language allowed mother Cora to know whether to feed Ivan or to find the pin that was sticking into him(Section 22.3).

Of course, Carlo might have been ill-disposed toward Irene at that moment, in which case Irene displaying increased discomfort would reduce any error in his perception of her happiness. Both these hypothetical situations (Carlo well or ill-disposed toward Irene) illustrate a fundamental requirement of any protocol, the ability of each party to display to the other in a way that the partner can perceive accurately, perhaps categorically, what the other intends to display, which might possibly be deceitful. These requirements lead to the different complexities of the full 19-control loop form of the protocol. We shall deal with them gradually.

[1] Notice that as stated here, the participants need to perceive errors directly, which Powers does not allow, but the Seth-Friston equivalent circuit connection for the hierarchy (Figure 7.2 in Volume 1) does allow. Later. however, we will find that direct perception of one’s own error is not required. It is the partner who perceives those errors.

EetuP · December 7, 2021, 9:28am

Thank you Eva for interesting message! According to your advice I try to give some negative feedback.

I think that you – perhaps intentionally – over exaggerate. It is true that life is based on negative feedback (NF), but there are important places for positive feedback (PF) too. Biology dictionary gives five examples of the uses of PF in our biological life: blood clotting, the menstrual cycle, labor and childbirth, digestion, and nerve signaling (https://biologydictionary.net/positive-feedback/).

I believe also in social life there are places where PF is useful and needed. In principle the difference between these feedback alternatives is perhaps this: If you get something which you don’t want (or don’t get what you want), then you give negative feedback. For example the But if you get something which you want to get more, then you give positive feedback. Is that right?

In addition, because life is based on self-balancing NF and because PF is not-balancing but in principle unlimitedly accelerating the latter must be somehow restricted to healthy limits. Conflicts and addictions are perhaps examples of often unhealthily unlimited PFs. So it seems to me that NF can use PF as a tool. For example in childbirth there is a higher NF where the error is that the child is perceived inside while the reference is to perceive it outside and the output uses the PF of labor. When the error of the higher NF loop is corrected then the lower PF is stopped. Generally: The lower PF unit wants something more and more but the higher NF unit decides what is enough.

As for your example about reading students’ reports, doesn’t it often happen that the next version is in some relation better than the previous but not as good as it could be. Now if you give negative feedback only, the student will think that she must try some thing else and the next version is possible worse in that relation. Rather, you will give positive feedback and say that it is now better but she should change it still more to that same direction. Isn’t it often so that we must encourage the students to continue their efforts, to try harder? I think this kind of PF is useful and even necessary in education and in life generally – if it is only confined under the higher NF which gives the direction and limits for it.

Evadeh · December 7, 2021, 11:23am

Thank you Eetu for your feedback! Very enlightening, and it changes my view of PF into something more positive

I also noted, in the process of grading papers, that I often give comments such as “Well done, this is what you should do more often!”. These comments (positive feedback in the laymans terms) often work well, students feel encouraged and more assured that their work is going in the right direction. These comments usually never lead to exponential effects where students only do that one reinforced thing… but perhaps they are also kept in check by the boundaries of the assignment?

I just read this in Donella Meadows’s systems thinking, that fits with what you explain here.

Reinforcing Feedback Loops—The strength of the gain of driving loops

A balancing feedback loop is self-correcting; a reinforcing feedback loop is self-reinforcing. The more it works, the more it gains power to work some more, driving system behavior in one direction. The more people catch the flu, the more they infect other people. The more babies are born, the more people grow up to have babies. The more money you have in the bank, the more interest you earn, the more money you have in the bank. The more the soil erodes, the less vegetation it can support, the fewer roots and leaves to soften rain and runoff, the more soil erodes. The more high-energy neutrons in the critical mass, the more they knock into nuclei and generate more high-energy neutrons, leading to a nuclear explosion or meltdown.

Reinforcing feedback loops are sources of growth, explosion, erosion, and collapse in systems. A system with an unchecked reinforcing loop ultimately will destroy itself. That’s why there are so few of them. Usually a balancing loop will kick in sooner or later. The epidemic will run out of infectible people—or people will take increasingly stronger steps to avoid being infected. The death rate will rise to equal the birth rate—or people will see the consequences of unchecked population growth and have fewer babies. The soil will erode away to bedrock, and after a million years the bedrock will crumble into new soil—or people will stop overgrazing, put up check dams, plant trees, and stop the erosion.

In all those examples, the first outcome is what will happen if the reinforcing loop runs its course, the second is what will happen if there’s an intervention to reduce its self-multiplying power. Reducing the gain around a reinforcing loop—slowing the growth—is usually a more powerful leverage point in systems than strengthening balancing loops, and far more preferable than letting the reinforcing loop run.

Population and economic growth rates in the World model are leverage points, because slowing them gives the many balancing loops, through technology and markets and other forms of adaptation (all of which have limits and delays), time to function. It’s the same as slowing the car when you’re driving too fast, rather than calling for more responsive brakes or technical advances in steering.

There are many reinforcing feedback loops in society that reward the winners of a competition with the resources to win even bigger next time—the “success to the successful” trap. Rich people collect interest; poor people pay it. Rich people pay accountants and lean on politicians to reduce their taxes; poor people can’t. Rich people give their kids inheritances and good educations. Antipoverty programs are weak balancing loops that try to counter these strong reinforcing ones. It would be much more effective to weaken the reinforcing loops. That’s what progressive income tax, inheritance tax, and universal high-quality public education programs are meant to do. If the wealthy can influence government to weaken, rather than strengthen, those measures, then the government itself shifts from a balancing structure to one that reinforces success to the successful!

Look for leverage points around birth rates, interest rates, erosion rates, “success to the successful” loops, any place where the more you have of something, the more you have the possibility of having more.

So the next version of my blog post could be more nuanced on positive feedback, and how it can be used in a balanced way. I’ll work on that!
Eva

wmansell · December 7, 2021, 12:31pm

Hi Eva, you could try it out with one of Bill’s live block diagrams. Extreme amounts of gain lead to overshoot, like when you are using a rudder on a boat until you get used to the delay….

MartinT · December 7, 2021, 8:22pm

No, it is not right. Your first move reduced but did not eliminate the error. You continue the negative the negative feedback until your error is reduced to within tolerance limits.

Whatever helped the student to improve the first time was negative feedback, since whatever she did reduced the error in your perception of the work. Her controlled perception might be her level of understanding, or it might be her perception of your level of approval.

“Giving” negative feedback is not possible according to PCT, but in everyday speech, to “give negative feedback” is to indicate disapproval, thus increasing the error in her perception of your approval (presuming she wants your approval). To “say that it is now better but she should change it still more to that same direction” is to continue your beneficial negative feedback.