Presented at The Developing Group 4 June 2005
1. Contents
2. What is a Feedback Loop?
- ‘Feedback’ has two meanings
- Diagram 1 – A simple feedback loop
- Diagram 2 – Ways feedback loops are not established or maintained
- Diagram 3 (Part 1 and 2) – The interior/exterior perspective
3. Ten Principles — Feedback Loops:
- Are what make complex-adaptive systems adaptive.
- Involve “circular or more complex chains of causation”.
- Are a pattern of organisation.
- Involve the ‘originator’ adapting to the effects of its own behaviour.
- Involve attempting to satisfy a value or achieve a purpose.
- Cross physical boundaries.
- Involve cycle time and delay.
- Are of two types.
- Involve thresholds.
- Operate at multiple levels simultaneously.
4. What difference does it make to think in terms of feedback loops?
- 14 answers
- Diagram 4: Oscillation between spoon feeding and giving responsibility
5. Appendix
- Associations with other models
- References
2. What is a Feedback Loop?
‘Feedback’ has two meanings
In everyday speech ‘feedback’ has come to mean: A piece of information received from the environment or another person (as in “we responded to feedback”). While this kind of feedback may be a sensory observation, mostly it is what people like or dislike about our behaviour.
In cybernetics, feedback means: A series of interacting processes which together result in a system adapting to the effect of its previous behaviour. Technically feedback is a pattern of organisation that operates only as long as the part ‘originating’ action is influenced such that its behaviour changes.
It is clear that these two definitions relate to different levels — the everyday linear definition being a subset of the systemic meaning. In other words, feedback is either a part of a process or the whole circular process itself.
In order to distinguish between these two meanings, we will use the metaphor of a ‘feedback loop’ when we want to refer to the systemic definition.
From an individual’s viewpoint, a feedback loop exists when my system notices how the world responds to my behaviour and I adjust my behaviour in response that response, and so on. At any point my options are to do more or less of what I am already doing, or change to a different kind of behaviour. Either way, my aim will be keep things as they are or to change things to get more of what I want and less of what I don’t want. However, you cannot get a systemic perspective from an individual’s point of view. In a healthy system I take into account that other people are doing the same, that we are all part of a wider system that no one part can control.
It is interesting to note that the use of the word ‘feedback’ in everyday language has been accompanied by a shift of attention. In a feedback loop the emphasis is on all parts of the system changing to meet a value or goal. Each part is considered to be a contributor to the overall effect and therefore can influence whether this effect continues or not. In the everyday sense of feedback, the attention has shifted to the information supplied by an outside person. We do not think this shift from internal dynamic to external thing is accidental.
Let’s take another example to illustrate our point:
Originally zero-tolerance meant: we will not tolerate our failing to meet some standard (e.g. “We will not let this subway car leave the depot unless it is clean and free of graffiti.”). Zero-tolerance has come to mean we will not tolerate other people’s rule-breaking behaviour (e.g. “We will punish them every time they do not wear school uniform.”). The key to understanding the difference between the original and the everyday meaning is to ask “zero-tolerance of what?” The focus of attention in the former is on us meeting some criteria, in the latter it is on others failing to meet a criteria.
We do not believe that the shift in emphasis in the way feedback and zero-tolerance is defined —from more to less complex, from us to them, and in its most distorted form from personal responsibility to blame — is a coincidence. We believe it reflects the level of development of people’s emotional intelligence.
Diagram 1: A simple feedback loop
How to read a feedback loop diagram:
You can start anywhere because a feedback loop exists as a whole and a ‘start’ or ‘end’ is just wherever we choose to punctuate the loop for convenience. However we have described the loop as if it starts from an action.
Each arrow involves an interaction and therefore an influence.
Each process involves a re-encoding or transformation of information.
What is cause and what is effect depends on how you punctuate the loop. Systemically speaking every event is both a cause and an effect. A feedback loop is, minimally, the effect of the effect of the effect.
Wherever you start, you need to cycle round a feedback loop a couple of times to get a sense of the story behind the diagram.
Another name for the ‘external reaction’ of a feedback loop is ‘evidence’.
Whenever complex adaptive systems are involved, linear cause-effect explanations only account for “an arc” of a feedback loop, e.g. “We will be tough on the causes of crime” does not acknowledge that ‘crime’ is equally influencing the government’s behaviour.
Diagram 2: Ways feedback loops are not established or maintained (ie. no learning)
Given that the world always reacts — even if only at the level of physics and chemistry — there will always be an interaction between 1 and 2. (Hence the NLP maxim “You cannot not influence.”) It also follows that something has to happen to prevent the chain of influence from returning to its source. As the diagram shows, this can happen in three places:
a Between 2 & 3 Failure to notice the reaction to own behaviour. b Between 3 & 4 Failure to notice what was noticed
or
Failure to adequately process what was noticed.c Between 4 & new 1 Failure to act on what was processed.
See our Self-Deception, Self-Delusion, Self-Denial article for more on how individuals keep certain feedback loops from operating.
Systemically speaking, the system will maintain ‘no change’ through a dampening feedback loop. (See Principle 8 below.)
Diagram 3 – Part 1: The interior/exterior perspective
The line down the middle of Diagram 3 bisects process 1, ‘Action’ and process 3, ‘Notice Reaction’. This line represents a physical boundary between the internal behaviour of system X on the left-hand side, and external behaviour of it and another system, Y, on the right-hand side.
Note:
- A true observer can only see the external behaviour of system X and it’s interaction with system Y.
- However, the diagram is drawn from the viewpoint of an external perceiver with the ability to see ‘inside’ system X (left hand half of the diagram).
- When the observer is X, this is known as 3rd Position in NLP (Robert Dilts).
- In Symbolic Modelling ‘self-modelling’ mostly involves the client modelling the left-hand half of the diagram except where the modelling involves feedback from an external source:
– a verbatim written or spoken description
– a measurement from a machine e.g. weight as measured by scales.
An interior cannot be observed directly from the exterior. But, as Humberto Maturana and Fransisco Varela say in The Tree of knowledge, “all behaviour is the external dance of internal relations”. Therefore, the pattern of actions and reactions represents the relationship between the maps of X and Y. A relationship will continue to exist as long as the two systems are compatible, i.e. the people involved dance together. There is no value judgement here — compatible simply means the two systems have ongoing responses to each other. In the same way, Eric Burne, the originaotor of Transactional Analysis, noted that “complementary” transaction continue a conversation until a “crossed” transaction stops it.
Diagram 3 – Part 2: The interior/exterior perspective
Diagram’s 1-3 are drawn from the perspective of a perceiver perceiving the system taking action (called X). If we add in the internal process of the system that reacts (called Y) the resulting action by X is available as a ‘reaction’ to Y’s behaviour. Now there are two interacting feedback loops. We can map the information pathway across the two physical boundaries (marked by the vertical lines) thus:
Note that a large part of any ‘reaction’ may have little to do with the ‘action’ because it will be specified by the organisation of the system reacting. However as long as the two systems maintain a relationship, then the action-reaction cycle must be complementary. Given this, every reaction in an ongoing relationship tells us something about our self.
This means that a key part of learning from a reaction is to decide what is to be noticed, considered and acted upon, and what is to be ignored. This is not always a simple decision because we are all liable to “cognitive bias.” And, how do we know whether self-deception is involved when we ignore a reaction? One way is to look for what we call the “boulder of truth” in all feedback (everyday meaning). Knowing the desired outcomes of both systems help to decide what to pay attention to.
3. Ten Principles of Feedback Loops
1. Feedback loops are what make complex-adaptive systems adaptive.
Because a complex adaptive system is continually changing, so must all the other systems that interact with it — if the relationship is to continue to exit. This ‘need’ to maintain communion in order to preserve the self is the wellspring of co-evolution.
2. Feedback loops involve “circular or more complex chains of causation”.
Gregory Bateson’s now famous saying sums up the systemic nature of feedback loops.
As a result, a pattern of organisation that exhibits “operational closure” emerges. That is it forms a unity which has a measure of autonomy and the means to activate self-preserving changes (Maturana and Varela).
3. Feedback loops are a pattern of organisation.
A feedback loop is not a thing, it is not a process, and it is not information. It is a pattern of organisation that exists only as long as the interactions between the parts continue to form an unbroken circuit. A feedback loop emerges when the effects of the behaviour of one part of a system are available to modify that part’s own behaviour. i.e. it is self-adaptive. And because they are circular, feedback loops can be considered ‘iterative information pathways’.
The basic ‘unit of operation’ of a feedback loop is change, that is difference. And difference is a non-substantial phenomenon. It cannot be located in space or time, and it cannot be easily measured unless the salient attributes of the two events can be counted.
4. Feedback loops involve the ‘originator’ adapting to the effects of its own behaviour.
| Intended meaning | = X’s 4 |
| Action to convey meaning | = X’s 1 |
| Received meaning | = Y’s 3 & 4 |
| Action to convey meaning | = Y’s 1 = X’s 2 |
| Compare intended & received meaning | = X’s 3 & 4 |
| … and so on. |
5. Feedback loops involve attempting to satisfy a value or achieve a purpose.
The above definition begs a question: Adaptive for what? For the purpose of achieving or maintaining a state, value, criterion or preference. The effects of our action are compared with the goal or value: Does my behaviour achieve or move me towards or away from my purpose? Purpose can be as specific as ‘an internal body temperature of close to 98.4 degrees F’ or as general as ‘my survival’.
It is important to recognise that most goals and values are implicit, even when explicit goals are stated. The ‘mission statement’ may eb the publicly stated goal of an organisation, but the fact that corporation spend so much money trying to get ‘buy-in’ to the mission shows there are plenty of other goals being pursued.
At process 4 the comparison with a desired outcome or value helps the system decide whether to:
- increase or decrease the previous action
- change to a different but equivalent behaviour
- change to a different kind of behaviour
- change the reference point
This is often not a simple calculation because as Edward DeBono says “Sometimes you have to journey south for a while in order to journey north,” and the purpose or value itself can change. In other words, it is what we call ‘a dynamic reference point‘.
Gregory Bateson repeatedly stressed that if the goal is to maximise the criteria of a part of a system without consideration of the effects on other parts the result will eventually become pathological for the system as a whole.
6. Feedback loops cross physical boundaries.
Information pathways are no respecter of physical boundaries. Gregory Bateson pointed out that feedback loops are both interesting and a challenge to work with is because they cross what we traditionally consider to be external boundaries.
Thinking systemically means considering the unity of the information pathways as the principal unit of analysis and not to be restricted by physical boundaries. From this perspective, intelligence and mind are properties of the system; not of an individual. One could say that ‘intellectual property is theft’!
7. Feedback loops involve cycle time and delay.
The cycle time (also known as ‘delay’) is how long it takes for all the parts of the loop to interact and the ‘message’ to complete its journey round the loop.
The cycle time when riding a bike is not how long you stay in the saddle but how long before you know the effect of tilting — a fraction of a second. Whereas the cycle time in GCSE school examinations is several months. In the former learning can take place several times a second, in the later the delay is so great that learning rarely happens.
As a general principle, the longer the cycle time and the more transforms the information undergoes, the more difficult it is to see the systemic relationships.
8. Two types of Feedback Loop.
Feedback loops can be divided into two types by examining what happens to a system over time as a result of the loop operating. Either a change in the system is reinforced (i.e. more of more, or less of less) or a change in the system is dampened (i.e. less of more, or more of less). These two types are technically known as positive and negative feedback respectively. However this metaphor is open to confusion due the the value-laden meaning of those words in other contexts. There are plenty of other metaphors which we believe better distinguish between the two types of feedback:
Positive: self-reinforcing, self-amplifying, escalating, runaway, compounding, snowballing
Negative: self-sustaining, self-maintaining, dampening, stabilising, balancing
“Negative feedback, then, is a way of reaching an equilibrium point despite unpredictable — and changing — external conditions. The ‘negativity’ keeps the system in check, just as ‘positive feedback’ propels [the] system onward. [They are] a way of indirectly pushing a fluid, changeable system toward a goal. Negative feedback comes in many shapes and sizes. You can build it into ballistic missiles or circuit boards, neurones, or blood vessels. It is, in today’s terms, ‘platform agnostic’. At its most schematic, negative feedback entails comparing the current state of a system to the desired state, and pushing the system in a direction that minimises the difference between the two states.” Steven Johnson, Emergence (2001) pp. 138-140.
The continued survival of a system requires that the interplay between reinforcing and stabilising feedback loops creates a dynamic equilibrium (see Developing Group notes of 1 December, 2002).
What is a ‘dampening feedback loop’ dampening?
The simple answer is ‘change’. At the next level up, it is balancing a reinforcing feedback loop — or the possibility of one developing.
What is a ‘reinforcing feedback loop’ reinforcing?
The answer is again ‘change’. At the next level up, it is countering a stabilising feedback loop.
One reason why feedback loops require some thinking about is because they operate at two levels above what we usually attend to (i.e. what happens):
Feedback Loop
(a consistent pattern of change)^^^^^^^^^^
Change
(a difference between two events)^^^^^^^^^^
Events
9. Feedback loops involve thresholds.
Reinforcing and balancing feedback loops have to be understood in terms of limits and thresholds. Reinforcing feedback loops tend to move the system toward and eventually over a threshold. Balancing feedback loops keep the system within bounds and close to midpoints.
For example:
Take heroin —> experience a high and an increased tolerance to the drug —> take more heroin … until a threshold is crossed (the amount of heroin the body can accommodate) and. an overdose occurs.
Note the reinforcing feedback loop works the other way around: when the intake of heroin is reduced, the tolerance threshold is reduced at the same time. This is why it is common for addicts to overdose immediately after a detox process because they haven’t allowed for how their body has adapted to the lack of heroin — it now can’t accommodate the amount of heroin they took before they started detoxing.
The effect of an escalating feedback loop depends on:
The steepness of the rate of change
(the steeper the less time/opportunity for balancing feedback to operate).
The degree of tolerance or width of limits
(the less tolerance the less time/opportunity for balancing feedback to operate).
The closeness of the system to a threshold
(the closer to a threshold, the less time/opportunity for balancing feedback to operate).
10. Feedback loops operate at multiple levels simultaneously.
In a real sense, our personalities are partially the sum of all these invisible feedback mechanisms; but to begin to understand those mechanisms, you need additional levels of feedback.” Steven Johnson in Emergence (2001), p. 142.
Feedback in Relation to Gregory Bateson’s Levels of Learning:
III Learning to
learn to learnA feedback loop using the effects of learning II is established II Learning to
learnA feedback loop using the effects of Learning I is established I Rote
learningA feedback loop is established until the learning is complete 0 No learning No feedback loop is established
What difference does it make to think in terms of feedback loops?
14 Answers:
Simply remembering that any pattern manifested by a complex adaptive system will involve at least one feedback loop helps to maintain a systemic perspective.
Asking our self “What’s the pattern?” and then asking “What feedback loops are involved in maintaining the pattern?”.
By modelling a sequence determines whether it ‘completes a loop’ or not. Tracing the pathway around a couple of times both checks the consistency of the pattern and gives the client an opportunity to see the pattern of the feedback loop for themselves.
It helps to recognise that no part of the loop controls the system and that every part is playing its part — otherwise the pattern could not be maintained.
We appreciate that any change in the behaviour of the system will be maintained only if there’s a change in an appropriate feedback loop and/or the establishment of a new feedback loop.
It helps to not take sides and to remember that blame and guilt are the result of seeing arcs of a feedback loop rather than the whole pattern.
We let every response from the client influence our next action — but not too much!
We assume that characterological adjectives (e.g. I’m lazy) and nominalisations (e.g. an addiction) are names for a system of interlocking feedback loops.
We remember that every long-term change involves a change to the organisation of a feedback loop, and because they are self-maintaining, neither we nor the client can know the full extent of what will happen.
Modelling linear cause-effect ways of thinking is enhanced by maintaining a wider and more systemic perspective.
We remember that the client really wants to establish new self-stabilising feedback loops with more appropriate criteria. However to get there they may need an escalating feedback loop to operate for a while — and that’s scary.
We consider when, where and how a potential feedback loop is interrupted (both physically and informationally) and this guides our questions.
It keeps us in touch with the dynamic and multilevel nature of self-organising systems. They are continually changing over time to maintain their identity.
Development is about sustaining a change until it becomes the norm; and that eventually requires learning about the feedback process itself.
Diagram 4: Oscillation between spoon feeding and giving responsibility
Below is a diagram of a feedback loop depicting James’ reaction to working with pupils at a High School in the North East of England:
This diagram helped James to see that until the top half of the diagram became a reinforcing feedback loop he would likely continue to feel frustrated and disappointed; and the pupils would continue to learn through spoon feeding. He concluded that to add in the ‘missing link’ he would need to consistently focus on those pupils (or those aspects of an individual pupil) who showed the slightest interest in taking responsibility for any part of their learning — and to make this explicit.
5. Appendix
Associations with other models
The feedback loop model can be mapped on to the Kolb learning cycle:
| FEEDBACK LOOP | KOLB LEARNING CYCLE | |
| 1. Action | = | Active Experimentation |
| 3. Notice Reaction | = | Concrete Experience |
| 4. Internal Change | = | Reflective Observation & Abstract Conceptualisation |
And mapped on to Robert Dilts’ Neurological Levels:
| FEEDBACK LOOP | DILTS’ LOGICAL LEVELS | |
| 1. Action | = | Behaviour |
| 2. Reaction | = | Environment |
| 3. Notice Reaction | = | Behaviour |
| 4. Internal Change | = | All the higher levels |
And mapped on to Ken Wilber’s Four Quadrants:
| FEEDBACK LOOP | WILBER’S QUADRANT’S | |||
| Left-hand half | = | Interior individual | I | Upper left |
| Right-hand 1 & 3 | = | Exterior individual | It | Upper right |
| Right-hand half | = | Exterior system | Its | Lower right |
| Background of 4* | = | Interior system | We | Lower left |
References
Peter Senge, The Fifth Discipline (in particular Chapter 5)
Gregory Bateson, Steps to an Ecology of Mind (and any of his later writing)
Steven Johnson, Emergence
Malcolm Gladwell, The Tipping Point
Humberto Maturana and Fransisco Varela, The Tree of knowledge,
