Something curious often happens in coaching. A client suddenly sees something very clearly.
“I realized that I try to control everything.”
“I realized that I’m afraid of disappointing others.”
“I realized that I immediately interpret my manager’s silence as dissatisfaction.”
“I realized that I don’t need to please everyone.”
There is a sense of relief, sometimes even a powerful emotional moment. It feels as though something important has happened. And something important may indeed have happened.
Yet a week later, the person returns to the same situation and once again controls, adapts, becomes anxious about the manager’s silence, or avoids conflict.
Why?
One possible reason is that understanding our model does not yet mean changing it.
Over recent decades, predictive processing, generative model theory, and active inference have offered an interesting way of looking at how human beings perceive the world, learn, and change. Karl Friston, Andy Clark, and Anil Seth are among the thinkers associated with these ideas, while Mark Miller extends them in interesting ways in relation to human well-being, emotion, and adaptation. Adjacent ideas from Mark Solms, John Vervaeke, and others allow us to broaden the picture further.
This is not one unified theory, and it would be inaccurate to attribute the same positions to all of these authors. More importantly, generative model theory does not explain the whole human being. It does not tell us what a good life is, what is worth striving for, or what wisdom, love, truth, or meaning are.
But it can help a coach understand one very important problem more clearly: how a person builds a model of the world from experience, how that model later shapes experience, and why new experience does not always change it.
We do not see only what is there
Let us begin with a simple example. A traffic light in front of us turns red. We do not doubt it: it is red. We are driving, so we stop.
Does this mean that generative model theory claims we simply “created” the colour red? No.
The external world places very strong constraints on our possible interpretations. Certain physical properties generate certain signals, our visual system processes them, and through prior learning we already know that this object is a traffic light and that its red light means “stop.” When the signal is clear and strong, the space for interpretation is relatively narrow.
Now imagine a very different kind of signal. After a presentation, a manager says: “Interesting.”
What does that mean?
- Did she like it?
- Did she dislike it?
- Was she surprised?
- Is she doubtful?
- Is she simply being polite?
- Does she want more time to think?
Here we enter a much more interesting part of the human world. We do not have direct access to another person’s intention. We have a word, a tone of voice, a facial expression, the history of the relationship, and the context of the situation. From all of this, our system has to make an inference.
And it does not do so in a vacuum. We bring into the situation our previous experience, learned regularities, expectations, relationship history, culture, and understanding of ourselves. All of this forms part of our generative model.
So one person will hear: “She really did find it interesting.” Another: “She didn’t like it at all.” A third: “She doesn’t trust me.” The signal is similar. The experienced reality is no longer quite the same.
What is a generative model, really?
Put very simply, a generative model can be understood as our system’s continuously learning “map of the world.” It has been shaped by countless previous experiences and helps the system anticipate what is happening, what is likely to happen next, what matters, and what it should do.
But the word “prediction” can be misleading here. It does not mean that a person spends the whole day consciously guessing what will happen in the future. Much of this processing happens without us noticing it at all.
The system continually draws on what it has already learned and predicts what signals it should receive if its current model of the world is sufficiently good. So a person does not merely react to what is happening. They enter a situation already expecting something.
- “If I disagree with my manager, he will get angry.”
- “If I show that I don’t know, I will lose authority.”
- “If I don’t control things, people will make mistakes.”
- “If I say what I want, I will be selfish.”
- “If someone is silent, something must be wrong.”
Such models do not necessarily exist as clearly formulated beliefs. Often they operate much more deeply. A person no longer thinks: “I have a hypothesis that conflict threatens a relationship.” They simply feel that conflict is dangerous.
This leads to an important idea for coaches: our strongest models are often the ones we no longer experience as models. We experience them as reality.
What do I notice – and what do I allow to become evidence?
Generative model theory allows us to ask another very interesting question. Not only, “How does the client interpret the situation?” but also, “Which data does their system give the greatest weight to?”
Suppose a client is convinced that their manager does not value them. During the week, the manager thanks them three times for good work, asks for their opinion in one meeting, and on one occasion reacts quite critically to their proposal.
Which signal will become the most important? The first four may barely register, while the fifth becomes: “There. I knew it.”
This is not simply a conscious logical choice. In predictive processing, the idea of precision weighting is important: the system does not merely encounter information; in a sense, it also evaluates how much that information can be trusted and how much significance to give it.
For coaching, this opens up a much subtler space than simply “changing beliefs.” We can ask:
- What did you notice in this situation?
- What seemed especially important to you?
- What else was happening that received less of your attention?
- What data support your understanding?
- And what does not fit it?
- What evidence does your current understanding readily accept, and what evidence does it explain away as something that need not be taken seriously?
This is no longer an attempt to prove to the client that they are “thinking incorrectly.” It is an exploration of how their world becomes the world they experience.
A mistake can be information for the model, not a verdict on the person
Here we arrive at one of the most valuable ideas this theory offers for learning: prediction error. More precisely, it is the mismatch between what the system expected and what it experienced.
Suppose a client expects: “If I challenge my manager, she will get angry.” Eventually, the client dares to disagree. The manager responds: “That’s a very good point. I hadn’t thought about it.”
Something important has happened. The experience did not match the prediction. In theory, this creates an opportunity for the model to learn. But the person will not necessarily learn from it. They may say, “She was just in a good mood this time,” “She only said that to be polite,” or “That was an exception.” The old model remains intact.
So one of the most important truths about learning is this: new information does not yet mean a new model.
The same applies to what we call mistakes in everyday life. A perfectionist has a very clear idea of how something should work. They implement a solution, but the outcome is different from what they expected. They may react: “I made a mistake. I should have anticipated this. How could I not have known?” Information about the model then becomes information about my worth.
But another response is possible: “Interesting. I expected X and got Y. That means something is missing from my model. What is it?”
This is a very different relationship with error. In coaching, it can therefore be useful to help a client distinguish “my model failed to anticipate something” from “there is something wrong with me.” A mistake can become a place where reality corrects our model – if we allow it to do so.
We do not only interpret the world. We help create it
This is where the ideas of generative models and active inference become particularly interesting for coaching.
Suppose a leader has a fairly strong model: “My people are not ready to make important decisions independently.” From this model comes a prediction: “If I don’t control things, there will be mistakes.” So the leader checks, asks people to seek approval, intervenes quickly, and makes the more important decisions personally.
The team learns that initiative is not particularly worthwhile. People ask the leader more and more often. They experiment less and take fewer risks in deciding independently. The leader observes this behaviour and says: “See? They really can’t make decisions on their own.”
A cycle emerges: model → prediction → attention → interpretation → action → other people’s response → new data → reinforcement of the model.
We do not merely interpret the world through our model. Our model changes our actions, our actions change how other people respond, and we later use their response as evidence about the world.
So a very powerful coaching question may not be, “Is your belief true?” but, “How might your expectation be contributing to the situation unfolding in exactly this way?”
Insight is not yet a model update
We can now return to the coaching paradox described at the beginning. The client says: “I understand. I don’t need to please everyone.” Excellent. A conceptual shift has occurred.
But the next day, someone important is disappointed in them and their whole body reacts as if the relationship were under threat. The person knows one thing, but their system still expects another.
Insight and transformation are therefore not the same thing. A person can understand their model without yet having updated it.
Deeper change may begin when a person not only understands that disagreement does not necessarily destroy a relationship, but starts to experience this. Once, they express a different opinion and the relationship remains intact. Then they do it again. Gradually, it is not only the thought that changes. The expectation changes. Later, the bodily response may change as well.
At a deeper level still, the person’s very understanding of relationship may change: difference and connection are no longer experienced as opposites. Eventually, before every difficult conversation, the person no longer needs to remind themselves: “I have the right to disagree.” They can simply disagree and remain in relationship.
We might say: insight changes what I understand. Learning changes what I can do with my experience. Transformation can change the very way I experience myself, other people, and the world.
That is why an experiment in coaching is more than an “action step”
This perspective also changes how we might understand action after a coaching session. We often end a session by asking: “So what will you do?”
But we can ask differently: “What could you try in order to learn something new about your current model?” Then action becomes not homework, but an experiment.
If a client thinks, “If I don’t give the team very clear instructions, everything will fall apart,” perhaps the experiment is not, “From now on, I will delegate more.” That is too large and too normative a leap.
The experiment could be much smaller: “In the next situation, I will clearly state the outcome that is needed, but I won’t tell them how to achieve it. Then I will observe what actually happens.”
This creates new data. And after the experiment comes another very important part of coaching:
- What did you notice?
- What matched your prediction?
- What did not match it?
- What did you take seriously?
- What did you feel tempted to explain away as an exception?
- Did anything change in your expectations?
In this way, coaching can help a person move from insight to actual learning.
The body also participates in creating our experience
The generative model perspective also reminds us that a human being is not a mind separated from a body. The system constantly receives information not only from the outside, but also from within: about heart rate, breathing, tension, pain, energy, arousal, and many other bodily states.
The same situation can therefore be experienced differently depending on our state. A manager says: “Could we talk after the meeting?” One day I am rested and calm. Another day I am exhausted, tense, and have already felt insecure for several weeks. The words are the same. The world I experience may be very different.
This does not mean that the body “distorts reality.” But it participates in how a situation becomes meaningful. So it can be useful for a coach to be interested not only in “What do you think about this?” but sometimes also in:
- What is happening in you as you talk about this?
- What do you notice in your body?
- What changes when you imagine another possibility?
But caution matters here. A bodily signal is not a ready-made interpretation either. Tension in the chest does not automatically mean “fear of rejection.” That would already be the coach’s model imposed on the client’s experience. The body is worth listening to, not translating on the client’s behalf.
And then there is the coach
So far, we have spoken mainly about the client’s generative model. But there is not only one generative model in a coaching conversation. There are at least two.
The coach also predicts, interprets, selects information, and assigns different weight to it.
The client falls silent. The coach thinks: “Something very deep is happening here.” Perhaps. Or perhaps the client simply cannot remember a name.
The client has mentioned control three times. The coach thinks: “This is their core issue.” Perhaps. Or perhaps not.
The client is talking about their manager and clenches a fist. The coach notices: “The body has shown where the real problem is.” Again, perhaps. But this is already a hypothesis.
One of the most important disciplines of a good coach may therefore be this: hold many hypotheses, but do not become attached to any of them too early.
Or, more simply: I can have a hypothesis without needing it to be true.
This is a very important epistemological stance in coaching.
A “powerful question” can also be dangerous
In the coaching world, we like to talk about powerful, transformative questions. But a question can be elegant and at the same time full of the coach’s generative model.
For example: “What would change if you trusted your team more?” It sounds like a great question. But it already contains a diagnosis: the problem is that the client does not trust the team enough. Perhaps that is true. But we do not yet know.
Sometimes much more open questions can leave greater space for what the coach does not yet understand:
- What is happening for you here?
- What do you notice?
In transformational coaching, the coach’s wisdom may be expressed not in how quickly they see the essence of the issue, but in how long they can remain curious even when they think they have already seen it.
The coaching relationship itself can change the model
There is another particularly important implication of this theory. A client enters coaching not only with models about work, managers, decisions, or goals. They also arrive with models about relationship.
- What happens when I do not know?
- What happens when I make a mistake?
- What happens when I disagree?
- What happens when I show vulnerability?
- Do I need to impress?
- Will I be judged?
- Will the other person take responsibility away from me?
And then the client may experience something that does not match their previous prediction. They say, “I don’t know,” and do not lose the coach’s respect. They disagree with the coach, and the relationship does not collapse. They are silent, and no one rushes them. They do not know what to do, and the coach does not take responsibility away from them by offering a solution.
These are new data too. Transformation in coaching may therefore happen not only because of what the coach asked, but also because of the kind of relationship the client experienced with them. Sometimes the relationship itself becomes an experimental space in which a person can, perhaps for the first time, experience a different way of being with another person.
But a more effective model does not yet mean a better human being
Here generative model theory reaches a limit that matters greatly for coaching.
Imagine a person who becomes very good at manipulating others. They experiment, receive feedback, refine their predictions, and become increasingly accurate at anticipating what they need to say in order to make another person do what they want.
Their model is learning extremely well. But is this the kind of transformation we want?
Here we need to distinguish at least three things: adaptation, development, and wisdom. More effective adaptation does not necessarily mean a more mature person. Learning does not necessarily move in a good direction. Effectiveness is not the same as wisdom.
So in coaching it is not enough to ask, “How can the client change?” We also need to leave room for another question: “What is worth changing towards?”
This is where we encounter what predictive processing alone cannot give us: values, meaning, ethics, responsibility, relationship, wisdom.
- We can help a client become more courageous. But what will that courage be used for?
- We can help them become more influential. But what will they do with that influence?
- We can help them adapt better. But is the environment they are adapting to worthy of that adaptation?
Transformational coaching therefore needs to keep two questions alive: “How can a person’s model change?” and “Towards what kind of person, and what kind of life, is that change leading?”
What might all of this mean for a coach in practice?
There is probably no need to begin coaching sessions by explaining predictive processing to clients. In most cases, that is unnecessary. But the theory can change how the coach themselves thinks about what is happening in the session.
When a client says, “My manager doesn’t trust me,” we can hear neither a fact that must simply be accepted nor a “limiting belief” that must be changed, but a model worth exploring.
When a client makes a mistake, we do not need to rush to reassure them or urge them to “learn from the mistake.” Instead, we can help them explore what they expected and what reality showed them differently.
When a client receives feedback, we can be interested not only in whether they understood it, but also in whether they allowed that information to become significant enough to matter to their model.
When a client has an insight, we need not deceive ourselves that change has already happened. We can ask what new experience would be needed for the new understanding to become a lived possibility.
When a client experiments, we can be interested not only in whether the experiment “worked,” but also in what new things the person learned about themselves, other people, and the world.
And throughout the process, we can remember that our own questions, interpretations, and even what we consider “transformation” also arise from our models.
Perhaps, then, one of the most important coaching competencies is not the ability to understand a client quickly. Perhaps it is the ability to remain open enough for the client to surprise our model too.
Beyond insight
Generative model theory does not provide a complete theory of human transformation. It does not answer the question of what a meaningful life is, which values deserve our loyalty, or what a human being should become.
But it helps us see something very important. We do not enter each situation empty-handed. We bring models shaped by our previous lives, which help us anticipate what is happening and what is likely to happen next. These models influence what we notice, what significance we give to information, how we feel, and how we act. Our behaviour, in turn, changes the environment and other people’s responses, and from those responses we receive new data that may either change our model or reinforce it further.
A person can therefore have a great many insights and still remain, in a fundamental sense, in the same world.
Deeper learning requires more. It requires the ability to notice our model, tolerate what does not fit it, avoid confusing error with a verdict on our worth, allow new information to become significant, experiment, and live in a new experience long enough for what began as merely a new thought to become a new expectation, a new action, a new relationship, and perhaps a new way of being.
And the coach carries one further responsibility: to help this happen without imposing their own model of how the client should change.
Perhaps the essence of transformational coaching is not to help a person find a more correct answer. Perhaps it is to help a person become increasingly able to learn from reality, notice the explanations they themselves have created for it, allow experience to correct them, dare to experiment, and choose with ever greater awareness who they want to become.
And then insight becomes not the end of change, but its beginning.