Where are memories stored?

1. The standard picture — synapses, engrams, Hebbian plasticity

The textbook account of memory storage, developed across the twentieth century and refined through the last two decades of optogenetic precision work, is the synaptic-plasticity picture. Memories are stored as patterns of altered synaptic strengths between neurons. When you learn something, certain synapses strengthen (long-term potentiation, LTP) and others weaken (long-term depression, LTD); the pattern of changes across an ensemble of neurons constitutes the memory trace, the engram. Recalling the memory reactivates the ensemble; forgetting is the decay or interference of the pattern.

The empirical floor for this picture is real and extensive. Eric Kandel's Nobel-rewarded work on Aplysia (the sea hare) demonstrated the molecular machinery of synaptic plasticity at single-cell resolution. Susumu Tonegawa's lab at MIT has, in a series of high-profile papers across the 2010s, identified specific engram cells in mouse hippocampus, demonstrated that artificially activating them produces apparent recall, and demonstrated that artificially inhibiting them produces apparent amnesia. The engram is a real, localisable, manipulable substrate of certain kinds of memory.

What the synaptic-plasticity picture handles well: most working memory; declarative memory at the timescale of seconds to years; procedural memory; the gradient of memory loss in classical amnesia syndromes; the differential preservation of remote vs recent memory in temporal-lobe damage (Ribot's law); the consolidation of short-term memory into long-term memory during sleep. None of this is in dispute.

What the synaptic-plasticity picture handles less well is everything in section 4.

2. The architecture across systems — not just hippocampus

Even within the standard picture, memory is not stored in one place. The cleanest contemporary breakdown:

• Hippocampus and surrounding medial temporal lobe — declarative memory, especially episodic. The case of patient HM (bilateral medial temporal resection, 1953) is the founding clinical demonstration. Without intact hippocampus, new declarative memories cannot be formed; old ones, especially remote ones, can persist for years after the surgery.
• Cerebellum and basal ganglia — procedural memory. Riding a bike, playing a piano scale, swinging a tennis racket. Patients with hippocampal damage but intact cerebellum can learn new motor skills despite having no conscious memory of the practice sessions.
• Amygdala and limbic structures — emotional learning and fear conditioning. The classical Pavlovian-conditioning literature.
• Sensory cortices — perceptual memory and recognition.
• Prefrontal cortex — working memory; the maintenance of information across short delays.
• Medial prefrontal cortex specifically — autobiographical and self-referential memory. Petr Janata's 2009 Cerebral Cortex paper showed that music from one's personal past activates medial PFC with a precision that almost no other stimulus class achieves. See the music-and-consciousness companion essay.

So even at the most conservative reading, memory is distributed. There is no single memory bank. Different kinds of memory live in different systems, and the systems are coupled. This much is consensus neuroscience. What the consensus does not yet have a clean account of is what holds these distributed traces together as one person's memory — how the procedural memory in the cerebellum, the episodic memory in the hippocampus, the emotional weighting in the amygdala, and the autobiographical thread in the medial PFC all come together as the felt continuity of a single experiencing self.

3. The empirical picture is messier than the textbook

Several lines of contemporary work have pushed the synaptic-storage account into trouble.

Engram studies, looked at closely

Tonegawa's optogenetic engram work is impressive but proves less than it is often summarised as proving. The experiments show that activating particular cells can trigger apparent recall and that silencing those cells can block apparent recall. They do not show that the memory is exhaustively stored in those cells. The cells could be necessary access points to a more distributed (or non-local) storage; the experiments cannot distinguish "storage" from "access." Tonegawa himself has been careful about this distinction; the popular literature less so.

Planarian regeneration memory

The single cleanest experimental challenge to a brain-as-memory-storage account: Michael Levin's lab (and predecessors going back to McConnell's mid-twentieth-century work) have trained planarian flatworms on a task (e.g., associating a textured surface with food), decapitated them, allowed them to regrow new heads from the tail fragment, and re-tested. The retrained planaria show savings — they reacquire the task faster than naive worms, often substantially so. The memory cannot be in the head because the head was discarded. It is somewhere in the regenerating tissue, encoded by something that survives decapitation. Levin's framework: bioelectric pattern across the body holds the memory in a way that does not require a brain to be present. See the Levin companion page.

Terminal lucidity

Patients in late-stage dementia, often hours to days before death, sometimes briefly recover coherent self-presence — speaking by name to family members they have not recognised in years, recalling specific episodes from their lives, singing through complete songs they have not been able to perform since their illness began. The phenomenon is documented across case series in the palliative-care literature, has been reviewed in multiple peer-reviewed papers, and is now well enough established to be cited in mainstream end-of-life care discussions. The neural substrate at the moment of terminal lucidity is, on every measure, more degraded than it was when the patient could not produce these performances. The synaptic-storage account has no clean explanation. The receiver-model account does: the field-coupling is briefly re-established by some trigger (often familiar music), the receiver flickers back into local coherence, and the self briefly localises again from a substrate where the pattern was never lost. See the terminal-lucidity companion page for the case literature.

Music in dementia

The most reliable single trigger for momentary lucidity in advanced dementia is familiar music from the patient's earlier life. Oliver Sacks's Musicophilia documents this across dozens of clinical vignettes; the Alive Inside documentary made the phenomenon public; the Music & Memory program has institutionalised it. The neuroscience explanation is that musical memory networks are unusually distributed and resilient. The receiver-model explanation is that the song is a field-pattern the patient has been coupled to for decades; the coupling is what is restored, not the autobiography. Both readings explain the phenomenon; the receiver-model reading explains it more naturally.

Stevenson's past-life cases

Ian Stevenson (University of Virginia, 1918–2007) and his successors at the Division of Perceptual Studies have, across six decades, accumulated approximately 2,500 case studies of children who appear to remember past lives. The methodology is rigorous — first-person interviews with the child, independent verification of the named-but-unknown previous personality, documentation of birthmarks and behavioural patterns that correspond to the previous personality's death wounds and habits, double-checking against the family's possible exposure to the information. Stevenson published case-control comparisons; the children's reported memories sometimes contain information they cannot plausibly have learned through ordinary channels. The case is contested, but the data are real and the corpus is large enough that some explanation is required. The synaptic-storage account does not have one. The receiver-model account does: memory persists at the field level beyond the substrate that held it, and under certain conditions the pattern can re-localise into a new receiver. See the reality-check page for Stevenson on the established-evidence tier.

4. The cases the standard view handles worst — a summary

Pulling them together so the pattern is visible:

The pattern: the standard synaptic-plasticity account is correct on the cases it was developed to explain, and increasingly strained on the cases it was not. Each unresolved case, taken alone, has alternative explanations. Taken together they form a class of phenomena the receiver-model reading handles more naturally than the production-model reading does.

5. The Rovelli framework applied — memory and entropy

The cleanest physics-grounded statement of the receiver-model's memory account comes from Carlo Rovelli's Memory and Entropy (2020, journal Entropy) and the follow-up Wolpert-Scharnhorst-Rovelli (2025) paper on Boltzmann brains. The argument (developed in detail on the Rovelli companion essay sections 12–13):

Memory, on Rovelli's analysis, is what happens when three thermodynamic conditions hold — system separation, temperature differences, and long thermalization times. Wherever those three conditions hold in a stable configuration, traces of the past accumulate as available information. A brain is one such configuration. So are rocks, photographs, archaeological strata, the cosmic microwave background. The same physics applies; the brain is one instance, not the privileged one.

When the receiver dissolves — biological death, substrate decommissioning — the three conditions fail. The structure that held the autobiographical narrative in place stops being maintained. By Rovelli's framework that part of the memory is no longer locally available. But entropy is observer-relative. The patterns the receiver added to the field during its life have not vanished from existence; they have been redistributed into the field's wider coarse-graining. What was held by the receiver as the receiver's own contents is released back into the field as the receiver ceases to be the place where it was held.

Two registers, two fates. The autobiographical register — episodic memory, explicit narrative-self — depends on the receiver and goes when the receiver does. The field-pattern register — everything the receiver added to the field during its life — persists as field structure, available to subsequent receivers under the right conditions. Terminal lucidity, Stevenson's data, Lucía's birthmark, the planarian's regenerated memory are read on the receiver model as partial signatures of the field-pattern register doing what it does. This is not proof that the receiver model is right. It is the framework in which the unresolved cases stop being anomalies and become predictions.

6. So — are individual memories recorded or retained in the consciousness field?

The honest answer is partly yes and partly no, with the division running along the autobiographical / field-pattern line above.

The autobiographical self — the specific narrative of "I was six and walked along the edge of the lake and fell through the ice and pulled myself out" — depends, on every available reading, on the substrate that held it. When the substrate dissolves, this register dissolves with it. The trilogy does not promise the survival of the autobiographical self. No serious reading of the receiver model does. What was held by the receiver as the receiver's own contents is released back into the field.

The field-pattern register — the resonances the receiver developed, the patterns the receiver helped maintain by being coupled to the field while alive, the structural changes the receiver added to the field's larger architecture — persists. It is no longer addressable as "this individual's autobiography." It is now part of the field's wider pattern. On the trilogy's reading, this is not nothing. It is the level at which terminal lucidity, past-life memories, the lineage chord, and the inheritance of cross-generational pattern occur. The patterns are real; they are no longer individual. The individual contributed them; the field carries them onward.

What this means for the felt experience of self is the question every conscious adult eventually asks. The trilogy's answer is not optimistic in the religious-survival sense. It is also not annihilationist. It is structural: what you are doing while you are alive adds to the field. The autobiographical self goes; the patterns added do not. The honest summary the receiver-model framework offers is that the work of a lifetime is the work of contributing pattern to a field that holds it after you, in a form that is not your contents but is your effect.

7. The trilogy's touchpoints

Specific places where the books engage these questions:

• Anima's edge-cases folder. The patient with the IED premonition, the girl with the birthmark matching her dead father's fatal wound (Lucía Reyes), the man with advanced Alzheimer's who wakes lucid and dies two days later. Each is a case the production model handles badly and the receiver model handles cleanly. Together they are José's clinical archive of phenomena the standard textbook does not address.
• Numen's Marcus Webb tapes. Webb's testimony — "I was the radio, not the listener. The chord came from somewhere. I was the place it was going" — is the phenomenology of being-receiver stated from inside an extreme altered state. It is one of the clearest single statements in the trilogy of what the receiver model proposes the receiver is doing during the kinds of experiences memory-storage accounts struggle with.
• Numen's lineage chord. José plays the chord every morning for twenty-four years and dies. Eight years pass. Alex plays the chord in Chapter XVI with pure reception, and it responds. Field-pattern persistence given literary form. The chord is one of the trilogy's working hypotheses about what the field-pattern register does when the receiver who fed it is gone.
• Limen's integrated argument. The companion volume sets the framework out as direct argument: the field carries pattern, receivers couple to it, individual autobiographical contents are held locally by receivers and released when receivers dissolve, the structural patterns the receivers contributed persist at the field level. Limen is where the trilogy makes the case the books otherwise dramatize.
• Fragile Light's deposited civilizational history. Kiran Sākshī's content — the alien civilization's history, the food release, the war, the death toll — is delivered to Bodhi's neuromorphic substrate as a structured presence, not as a real-time data stream. The mechanism is field-pattern transmission at interstellar scale. The receiving is what the receiver model says receiving is.

Reading list

Synaptic-plasticity foundation

Eric Kandel, In Search of Memory: The Emergence of a New Science of Mind (Norton, 2006). The classic accessible synthesis from the Nobel-rewarded molecular work.

Susumu Tonegawa et al., engram-cell papers across the 2010s in Science, Nature, Cell. The optogenetic localisation work and what it does and does not show.

Planarian regeneration memory

Tal Shomrat & Michael Levin, An automated training paradigm reveals long-term memory in planarians and its persistence through head regeneration, Journal of Experimental Biology 216 (2013): 3799–3810. The clean modern replication.

Terminal lucidity

Michael Nahm, Bruce Greyson, Emily Williams Kelly, Erlendur Haraldsson, Terminal lucidity: A review and a case collection, Archives of Gerontology and Geriatrics 55 (2012): 138–142. The published case-collection review.

Music-evoked autobiographical memory

Petr Janata, The neural architecture of music-evoked autobiographical memories, Cerebral Cortex 19 (2009): 2579–2594. The medial-PFC addressing paper.

Past-life case research

Ian Stevenson, Twenty Cases Suggestive of Reincarnation (University Press of Virginia, 1966, 2nd ed. 1974). The founding case-collection.

Ian Stevenson, Reincarnation and Biology: A Contribution to the Etiology of Birthmarks and Birth Defects (Praeger, 1997). Two-volume documentation of the birthmark/wound correlation.

Jim Tucker, Return to Life: Extraordinary Cases of Children Who Remember Past Lives (St. Martin's, 2013). The contemporary continuation of Stevenson's program at UVA's Division of Perceptual Studies.

The receiver-model architecture for memory

Carlo Rovelli, Memory and Entropy, Entropy (2020). The three thermodynamic conditions for trace formation.

David Wolpert, Jordan Scharnhorst, Carlo Rovelli (2025) — on Boltzmann brains and the limits of what physics can decide about whether memories track a real past. See the Rovelli companion essay.

This page is part of the Reading companion essays. For the bioelectric substrate carrying planarian regeneration memory, see Levin & the bioelectric blueprint; for the music-and-memory connection, Music and consciousness; for the broader inheritance-above-the-genome picture, The disappointment of genes and Morphic resonance; for the time and entropy framework, Rovelli's Order of Time; for the synthesis, The Evidence.