Understand the moment
A place to begin
Why can a new understanding feel true in one place and disappear in another?
Remembering opens access to an established memory, but it does not guarantee that the memory becomes revisable or is rewritten. A contradictory experience may modify earlier learning or create a competing memory, and a change that works in one context may remain difficult to retrieve elsewhere.
A conceptual framing from Flow HijackedReactivation creates an opportunity for revision, not a guarantee. Change depends on lability, causal assignment, context, precision and whether the result can travel beyond the revision episode.
The Flow Hijacked lens
Flow Hijacked conceptual synthesis
The Attractor Revision Corridor (ARC)
A Flow Hijacked conceptual synthesis: a multidimensional region in which an established representation remains causally active, becomes sufficiently revisable, and discrepant evidence is still assigned to that representation rather than to a separate cause. ARC is a falsifiable explanatory model—not a place in the brain, biomarker, validated measure or treatment protocol.
The proposed Attractor Revision Corridor asks under which joint conditions an established representation may change, while Revision Portability asks where later learning continues to guide action. These are separate, testable conceptual proposals, not verified mechanisms inferred from a powerful experience or instructions for revisiting traumatic memories.
What this lecture adds to the picture
The lecture places cellular lability, associative competition, hidden-cause inference, predictive processing, active inference and transfer on one map without treating them as the same mechanism. It proposes two separable Flow Hijacked constructs: ARC asks when an established representation may be revised; the Revision Portability Gradient asks where the resulting learning continues to govern prediction and action.
Alongside someone
Understand more. Leave room for the person.
The return of an old response does not by itself prove that an earlier change was false. You can acknowledge the difference between situations without using intense emotion as proof that a memory is being rewritten.
You can ask whether this account fits their experience. Disagreement matters too, and your own needs and boundaries still have a place in the relationship.
From understanding to life
A question to take with you
What changes around you between a situation where a new response is available and one where it becomes harder to find?
Think of one recent moment if that feels useful. Nothing has to be solved all at once, and there is no answer to submit.
More ideas for reflection and conversation
- Ask which prediction was active and what exactly contradicted it.
- Test learning across place, bodily state, cue combinations and available actions—not only where it was acquired.
- Treat return as information about retrieval and control conditions, not as a total verdict on the person or the change.
These are educational prompts, not a treatment plan or personal medical instructions.
Personal testimony
Where the idea meets a life
I met memory revision first as the stubborn return of a world I no longer wanted to inhabit. A familiar cue could make the old future feel immediate even after a sincere decision. That experience does not prove a mechanism; it explains why the lecture refuses the easy sentence that remembering automatically rewrites memory.
Roughly a year and a half in a therapeutic community taught me that context can protect change and limit it at the same time. Structure made different actions reachable, but leaving changed the cue set, the body, the available actions and the hidden question of what kind of world this was now.
The personal note in the full edition is testimony, not evidence. Its role is to keep the scientific endpoint honest: a revision that appears in one room or state may be real and still remain fragile. What matters in life is whether another response can be retrieved when the conditions of success no longer surround it.
When there is room for more
Introduction + web edition
Now follow the argument in depth.
When a familiar prediction meets experience that does not fit, the old representation may remain substantially intact, become modified, or be joined by a competing memory. Lecture 49 follows the complete retrieval-to-restabilization chain while separating reconsolidation from extinction, new learning, latent-cause inference, integration, differentiation and context-gated return.
About 4 min · Estimated time for the deeper reading. Follow the sequence or choose a section.
Choose from 6 sections
Listen to the synopsis
Web edition
About 4 min01
Retrieval is an entrance, not an edit command
Research anchorsNader, Schafe & LeDoux (2000); Ben Mamou et al. (2006); Lee (2009)
Retrieval makes information accessible; reactivation means an established representation is contributing to current processing. Destabilization is a further transition into a state that requires active restabilization. These stages overlap in time, but they are experimentally separable. A vivid recollection, conditioned response or distressed state shows expression—not necessarily lability.
This distinction is one of the field's strongest red lines. Post-reminder timing alone does not identify reconsolidation, and memory expression cannot reveal which molecular or computational gate opened. Causal animal work provides the strongest evidence for lability mechanisms; human work usually infers them indirectly from later behavior.
A memory can be retrieved and remain substantially stable.
02
Prediction error is a gate, not a magic switch
Research anchorsPedreira, Pérez-Cuesta & Maldonado (2004); Sevenster, Beckers & Kindt (2012)
Novelty, surprise and prediction error are related but not identical. The mismatch relevant to revision must bear on the prediction generated by the reactivated representation. An unusual experience can be emotionally intense yet leave the old model untouched when it is attributed to a special person, protected room or temporary bodily state.
Prediction errors also differ in content: value, identity, timing, probability, bodily consequence and causal structure need not update the same component. The phrase 'large prediction error' is therefore scientifically incomplete unless the violated prediction, response system and causal interpretation are specified.
Difference becomes revision-relevant only when it counts as evidence about the active model.
03
Why maximum discrepancy may not yield maximum revision
Research anchorsGershman et al. (2017); Kennedy et al. (2024); Mirea et al. (2024)
Very weak mismatch may confirm the current model or permit restabilization with little change. Intermediate, model-relevant mismatch may sometimes support modification. A large or contextually distinct discrepancy may instead imply that the present event came from another hidden cause, encouraging separate learning. Evidence for this nonlinearity is strongest in particular conditioning and computational paradigms—not as a universal biological curve.
The responsible image is a conditional probability field, not three hard zones. At any joint state, several outcomes remain possible: little-change restabilization, modification of old learning, formation of a competitor, or a change in performance without representational revision. Memory strength, reminder structure, context, precision and prior learning move those probabilities.
Maximum discrepancy can maximize separation rather than revision.
04
Context decides which memory gets the present
Research anchorsBouton (2002); Gershman, Blei & Niv (2010); Hupbach et al. (2007)
Extinction usually adds context-sensitive learning rather than simply erasing acquisition. Renewal, reinstatement and spontaneous recovery reveal that the older relation can remain available and recover control. Latent-cause models formalize the assignment problem: does the discrepancy belong to the old cause, or does it signal that a new situation has begun?
Context is not one label. Place, time, social role, bodily arousal, cue configuration and available actions can agree or conflict. Corrective learning can therefore work in one environment and fail elsewhere without proving that the earlier change was false. The transfer test must deliberately move beyond the room of success.
What was learned and what controls behavior later are distinct questions.
05
The revised ARC: a hypothesis with moving boundaries
Research anchorsGershman et al. (2017); Milton, Das & Merlo (2023); Kennedy et al. (2024)
ARC survives the hostile review only in revised form. It is not a scalar 'sweet spot' of surprise. It is a proposed joint region requiring at least causal reactivation, evidence that the target is revisable, and continued assignment of the discrepancy to the old representation. The relevant coordinates and thresholds differ across species, tasks, memory components and people.
ARC should be abandoned or absorbed into established theory if its joint variables add no predictive value beyond simpler associative or latent-cause models. Calling every failed result 'outside the corridor' would make the construct immune to evidence and therefore scientifically empty. A useful ARC must commit in advance to competing predictions, boundary estimates and outcomes that could prove it unnecessary.
ARC earns its name only if joint measurement predicts more than its component theories.
06
Revision Portability: can change travel?
Research anchorsBouton (2002); Craske et al. (2014); Seth (2013); Friston (2010)
The Revision Portability Gradient is the lecture's second Flow Hijacked synthesis. It describes how reliably revised learning governs behavior as four dimensions move away from the learning episode: external context, bodily state, cue configuration and action affordances. It applies to a learning episode and later conditions—not to a fixed trait, strength score or moral quality of a person.
The construct is useful only if body and action dimensions add predictive information beyond established generalization, transfer, state-dependent learning and renewal. Equal immediate improvement should not be assumed to imply equal real-world reach. ARC asks whether an old representation changed; Portability asks where whatever changed can still govern action. Neither may be used as retrospective proof of the other.
Change becomes consequential when it remains reachable after movement through life.
TECHNICAL READINGThe mathematical model and its limits
Mathematical model
From first principles
The revised Attractor Revision Corridor
ARC is written as a proposed joint region, not a dose of surprise. An established representation must be causally active, sufficiently open to revision and still treated as the cause of the discrepant event. The thresholds are operational commitments for a particular experiment—not universal constants.
𝒜ARC = {x : r(x) ≥ r₀, ℓ(x) ≥ ℓ₀, aold(x) ≥ a₀}- x
- joint experimental and organismic state
- r(x)
- degree of causal reactivation of the target representation
- ℓ(x)
- evidence that the target is labile or destabilizable
- aold(x)
- assignment of the discrepant evidence to the established representation
- r₀, ℓ₀, a₀
- prespecified task-specific thresholds
Claim boundary: A Flow Hijacked hypothesis formalization. It is not an established neural mechanism, validated measure, biomarker or treatment protocol; it should be retired if simpler models predict equally well.
CONNECT THE IDEASThe conceptual map and lecture poster
A conceptual illustration of the lecture. The shapes are not a brain scan, a personal measurement or a proven causal sequence.
One visual model
The update-or-separate decision field
The same discrepancy can restabilize old learning, modify it, create a competitor or change performance without changing the representation. ARC organizes the conditional region; it does not prove which outcome occurred.
Lecture poster
FLOW HIJACKED · 49
How the Brain Rewrites Memory
Reactivation creates an opportunity for revision, not a guarantee. Change depends on lability, causal assignment, context, precision and whether the result can travel beyond the revision episode.
Research anchors
Lineage, evidence and limits
Sources this page is thinking with
This is a focused set of sources doing conceptual work on the page, not a systematic review or exhaustive bibliography. The Research Library adds summaries, themes and cross-lecture links.
- Friston, K. (2010).The free-energy principle: A unified brain theory?.Nature Reviews Neuroscience, 11, 127–138.DOI ↗
- Clark, A. (2013).Whatever next? Predictive brains, situated agents, and the future of cognitive science.Behavioral and Brain Sciences, 36(3), 181–204.DOI ↗
- Seth, A. K. (2013).Interoceptive inference, emotion, and the embodied self.Trends in Cognitive Sciences, 17(11), 565–573.DOI ↗
- Nader, K., Schafe, G. E. & LeDoux, J. E. (2000).Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval.Nature, 406, 722–726.DOI ↗
- Suzuki, A., Josselyn, S. A., Frankland, P. W. et al. (2004).Memory reconsolidation and extinction have distinct temporal and biochemical signatures.Journal of Neuroscience, 24, 4787–4795.DOI ↗
- Pedreira, M. E., Pérez-Cuesta, L. M. & Maldonado, H. (2004).Mismatch between what is expected and what actually occurs triggers memory reconsolidation or extinction.Learning & Memory, 11, 579–585.DOI ↗
- Hupbach, A., Gomez, R., Hardt, O. & Nadel, L. (2007).Reconsolidation of episodic memories: A subtle reminder triggers integration of new information.Learning & Memory, 14, 47–53.DOI ↗
- Monfils, M.-H., Cowansage, K. K., Klann, E. & LeDoux, J. E. (2009).Extinction-reconsolidation boundaries: Key to persistent attenuation of fear memories.Science, 324, 951–955.DOI ↗
Citation does not imply that a source validates the complete Flow Hijacked synthesis.
AVAILABLE EDITIONS
The full lecture, at your own pace.
Complete 88-page bilingual source editions: 59 scientific sections, a three-page personal witness, 13 formal models, eight scientific visualizations and a 94-publication research portfolio.
FOLLOW THE THREAD
If another question has opened
Learning something new about a memory does not settle how change travels into life. The next lecture follows the conditions in which another response can be practiced.