Changing your mind is not the same as changing your predictions. Real updating is rarer, more specific, and shows up in an entirely different place than we've been taught to look.
You see something small with legs. Your body jumps back before you've thought anything. Spider. Your heart is still going. You think: spiders aren't that dangerous. You know this. You tell yourself this. The thought changes. The prediction doesn't. The next time you see something small with legs, the jump happens again.
What changed was the story you told about it. What didn't change was what your body expects. You didn't update the prediction. You learned a second explanation. The model added a new concept sitting on top of the old one. The prior underneath — small thing, fast movement, danger — is intact. Learning is fast, flexible, happens entirely inside the model. It can feel exactly like change. It isn't.
Now imagine something different. You see something small moving on the ground. You lean in instead of away. It's a baby octopus — iridescent, alien, moving in a way nothing in your experience has moved before. You don't have a category for this. You feel curiosity and hesitation at the same time. Fascination and wrongness. You want to touch it and you don't want to touch it. The prediction machine is generating error and there is no ready-made concept to collapse it into. And you stay. You don't resolve it. You let the not-knowing sit.
Later — maybe not immediately, but later — your body doesn't respond the same way to small moving things. Not because you reasoned your way there. Because the prediction system encountered something it couldn't fit into the old prediction, and you didn't escape before the window closed. That is updating.
The brain is very good at generating new explanations. That is not the same as generating new predictions. You can experience change without changing anything.
The trap is that the simulation is convincing. When the story changes, it can feel like relief, clarity, a new perspective — and it registers as change because the felt experience of the model has changed. But that is still happening inside the model. The automatic response — the jump before thought, the body bracing in familiar contexts, the urgency that doesn't wait for permission — that is the prediction. And it updates through contact, not through explanation.
Understanding is not the mechanism of change. Contact is. Updating happens when a prediction is active, something incompatible is felt, and the system doesn't escape or override it before the window closes. That is the whole thing. Every effective approach that actually changes patterns is creating those conditions — through different routes, at different levels, with different tools. The mechanism is the same.
If you still have to manage it, it hasn't updated yet.
Every time a memory — or more precisely, a prior — is activated and brought into live processing, it becomes temporarily unstable. There is a brief window during which the prediction is malleable. If, during that window, the nervous system encounters something genuinely incompatible — not something that refutes the prior intellectually, but something the nervous system actually experiences as disconfirming — the prior can update. This is memory reconsolidation. The mechanism behind EMDR, IFS unburdening, prolonged exposure, Coherence Therapy, and many other effective approaches. The approaches look different. The mechanism is the same.
The prior must be activated. Safe insight about a prediction that isn't live does not update it. You have to be in the state. The signal must be firing. Working with a pattern in the abstract, when it's quiet, is working with the map. The territory has to be accessible.
The incompatible experience must be felt. A logical counter-argument does not create the condition for update. The nervous system must encounter something that doesn't match the prediction at the level the prediction was built. Not a better story. A different felt experience.
There must be enough capacity. A system in collapse cannot integrate. A system that ejects when the signal gets too loud cannot complete the update. This is why capacity comes first — not as preparation, but as the substrate that makes the window accessible.
And — critically — you cannot resolve it too fast. If you reinterpret, explain, or regulate the signal before the update window closes, you cancel the error before the prior can revise. The urgency to make sense of it, to feel better, to reach for a concept — that urgency is what most treatments accidentally reinforce. Staying in the not-knowing is the mechanism. Wonder, not resolution.
Updating has two faces. The first is active — what is happening while the prior is being revised. The second is integrated — what is different afterward. Both are unfamiliar, because neither looks like the dramatic insight or emotional breakthrough we associate with change.
Something feels new — "I've never experienced it quite like this"
Curiosity without urgency — "huh, what is this?" rather than "I need to figure this out"
Time slows, attention deepens, more sensory detail and less narrative
Destabilization without collapse — the model is loosening, not breaking
Ambiguity is tolerated — "I don't know what this is yet" and that's okay
Absence of the old automatic reaction — same trigger, different response, no effort
Reduced need to manage — less fixing, less analyzing, less urgency
Loss of stickiness — thoughts arise but don't hook the same way
Different baseline, not just better peak — it doesn't hit the same way anymore
Less mental real estate — the system no longer flags it as high priority
The paradox is that updating often shows up as less rather than more. Less reaction. Less urgency. Less meaning-making. We are trained to notice intensity and insight and big shifts. Integration often arrives quietly, as an absence. Wonder is the door opening. Absence is the room having changed.
If you know change is happening when something feels new, you know it has happened when the old reaction doesn't come back.
Not all priors are equally accessible. The ease or difficulty of update maps onto the level of the system where the prior lives.
DMN narrative — the story layer
Held in language and concept. Accessible through reflection, reframing, and cognitive work. Updates when contradicting evidence is genuinely processed. This is what insight-oriented therapy and CBT primarily reach. Real and useful — but downstream of the signal.
Identity priors — interpretive, narrative-adjacent
Active at the moment of meaning capture. Am I resilient or fragile? Do I trust the signal? Am I usually at fault? These are accessible to the reflective mind — they can be approached through deconstruction first, then incompatible relational experience. Require both the prior activated and a genuinely disconfirming felt experience.
Interoceptive and relational premises — preverbal, body-held
Built before language. Woven into the interoceptive signal itself — they shape what pressure feels like and how threatening it registers before the DMN is involved. I am in danger. I am alone. I am bad. Not beliefs the person holds consciously — the texture of the signal itself. Require body-level and relational approaches. Language cannot reach where language was not present when the learning happened.
The confusion between these levels is part of why so much therapy stalls. The map updates at the top. The territory doesn't move. Someone understands their pattern completely and still runs it — because the understanding landed at the narrative level and the prior lives three levels down.
The nervous system is not only updated in therapy. Update happens wherever the conditions are met: prior activated, incompatible experience felt, enough capacity, not resolved too fast. Many ordinary states create this accidentally or by design.
Processes prediction error without the executive need to resolve it. Emotional memories are replayed and their affective charge reduced. The brain consolidates learning and integrates incompatible information without the waking urgency to make sense of it.
Temporarily quiets the self-referential DMN, reducing identity prior precision. The small self dissolves slightly in the presence of something vast. Prediction error arrives without threat. The window opens.
Incongruity without threat — mild prediction error in a safe container. The setup loads a prediction; the punchline violates it; the release is the system discovering the violation was harmless. A small update, repeated.
Generative error in low-stakes conditions. The system generates predictions and they are violated without consequence. Novel states and concepts arrive without urgency to resolve. The prediction machinery gets to revise in conditions where the cost of being wrong is low.
Sustained exposure to contexts where existing priors don't quite fit. The identity prior loosens when the environment stops confirming it. New concepts arrive that weren't in the existing model. The octopus rather than the spider.
Activating the prediction while preventing the compulsion or avoidance that would close the window. The nervous system discovers that the predicted catastrophe doesn't arrive. Experiential disconfirmation, not intellectual counter-argument.
Directly reduce DMN precision weighting and increase neural plasticity. The identity prior temporarily loosens, creating a window for reorientation. Integration is required — the window without intentional contact can destabilize rather than update.
The client's bond prior predicts abandonment, rejection, or harm. The relationship — over time, through attunement and rupture-repair — delivers something that doesn't match. Not the healer replacing the wound. The disconfirming experience itself, accumulated.
This is the part that most frameworks miss — and getting it wrong adds harm.
Some nervous systems are not in conditions where prior update is the right target. Neurodivergent nervous systems, postpartum nervous systems, presentations with chronically high baseline prediction error, trauma presentations with minimal capacity — in these contexts the prior update model doesn't apply, or applies only after conditions have been substantially changed. The stakes are higher. The sensitivity to prediction error is greater. Depletion is constant. Every demand costs more. Every mismatch activates more. Every intervention that requires sustained activation without capacity to complete it adds to an already-stressed system.
The intervention in these contexts is not update. It is accommodation, capacity restoration, legitimation, and constraint support. Reduce the total demand on the system. Identify what constraints are being threatened and address those directly. Build the tank before reaching for prior update work. Legitimate the signal — recognize what it is caring about, name that what the nervous system is doing makes sense given the conditions.
Trying to update a chronically depleted or highly sensitive system before these foundations are in place is not just ineffective. It can deepen the conclusion that nothing works, that the problem is the client, that the pattern is fixed. Treatment resistance is almost never a property of the client. It is almost always a description of an intervention meeting the wrong level — or the right level under the wrong conditions.
Wonder is the door opening. Absence is the room having changed. If you know change is happening when something feels new — you know it has happened when the old reaction doesn't come back.
A client has carried a shame prior for thirty years — "I am fundamentally too much." They have done years of excellent therapy. They understand the origin. They can narrate it with precision and compassion. And then something happens in the room that they didn't predict: the therapist is genuinely moved. Not performing care. Actually affected. The client watches this happening and something doesn't fit the prior. The prior said: too much generates withdrawal. This is not withdrawal. The signal is live. The incompatible experience is landing at the relational level — where the prior was built. The client starts crying. They don't fully know why. Something is completing that insight never reached.
That is not the therapist being warm. That is the mechanism working. The prior was activated, the incompatible experience arrived at the right level, and there was enough capacity in the room for the window to close with a revision. Three conditions met. Update possible.
Ecker, B., Ticic, R., & Hulley, L. (2012). Unlocking the Emotional Brain. Routledge. — The clearest clinical account of reconsolidation conditions.
Nader, K. & Hardt, O. (2009). A single standard for memory: The case for reconsolidation. Nature Reviews Neuroscience, 10(3), 224–234.
Lane, R.D. et al. (2015). Memory reconsolidation, emotional arousal, and the process of change in psychotherapy. Perspectives on Psychological Science, 10(2), 162–180.
Walker, M. (2017). Why We Sleep. Scribner. — On REM sleep and prediction error integration.