The Uncomfortable Truth About How Memory Actually Works

Memory is not storage. It is the maintenance of constraints against entropy. This is not metaphor or philosophical speculation. It is a conclusion backed by optogenetic manipulation of individual neurons, information-theoretic mathematics, 13,000 years of verified oral tradition, and the fundamental thermodynamics of information itself.

For over a century, we have imagined memory as a kind of biological filing cabinet or hard drive where experiences are stored, indexed, and retrieved intact. The evidence now says otherwise. Memory is constrained reconstruction, not retrieval from neural archives. What we call “remembering” is the dynamic rebuilding of approximate pasts using flexible components, optimized not for accuracy but for adaptive behavior in an uncertain future.

This reframing dissolves longstanding puzzles: Why do we misremember? Why does every act of recall potentially alter what we recall? Why do the world’s best autobiographical rememberers still create false memories? Why did Indigenous Australian cultures preserve geologically accurate information for over 10,000 years without writing? The answer, emerging from converging lines of evidence across neuroscience, physics, and traditional knowledge systems, is that memory is not about the past at all. It is about maintaining the constraints necessary to predict and navigate the future.

The Empirical Demolition of “Storage and Retrieval”

Reconsolidation: Memory Rewrites Itself

The first crack in the storage metaphor appeared in 2000 when neuroscientist Karim Nader published findings that would fundamentally challenge how we understand memory formation.[1] Working with rats, Nader’s team demonstrated that consolidated fear memories, when reactivated, return to a fragile state requiring new protein synthesis to restabilize. The implications were startling: infuse the protein-synthesis inhibitor anisomycin into the amygdala during this vulnerable window and the memory degrades, regardless of whether it is one day or fourteen days old. The same treatment without reactivation leaves memory intact.

Think of it like opening a document on your computer. In the traditional storage view, opening a file simply displays what was saved. But Nader’s findings suggest something stranger: every time you open the file, you must save it again, or it disappears. And in that moment between opening and re-saving, the contents can be modified.

Over 900 animal studies have now replicated reconsolidation effects across species and memory types, though human replications remain inconsistent with at least 14 published failures.[2] The critical insight is that reconsolidation is triggered specifically by prediction error, a mismatch between what the memory “expects” and what actually happens during retrieval. Too little novelty and no destabilization occurs; too much novelty and the process shifts to extinction rather than reconsolidation. Memory, it turns out, updates itself precisely when its predictions fail.

Daniela Schiller’s landmark 2010 study exploited this window with remarkable precision.[3] By delivering extinction training within approximately ten minutes of a fear memory’s reactivation, her team prevented the return of fear for at least one year. Unlike standard extinction, which merely creates a competing inhibitory trace (imagine learning to associate a once-feared dog with treats, while the original fear remains dormant), Schiller’s protocol appeared to rewrite the original memory during its vulnerable state.

Donna Bridge and Joel Voss extended this logic to everyday declarative memory.[4] Using functional MRI imaging, they demonstrated that hippocampal activity during retrieval predicted subsequent memory updating. The hippocampus, long thought to be the brain’s storage center, does not distinguish between creating new memories and modifying existing ones. Every retrieval is potentially an act of rewriting.

Engrams: The Physical Trace That Isn’t Fixed

The concept of the engram, the physical trace of memory in the brain, dates back to Richard Semon in 1904. For over a century, neuroscientists sought this trace. Recent advances in optogenetics, which allow researchers to activate specific neurons with light, finally enabled its direct manipulation with results that were both confirming and deeply puzzling.

Sheena Josselyn and Susumu Tonegawa’s comprehensive 2020 review in Science defined engram cells as neural ensembles that undergo lasting changes during encoding and whose reactivation by retrieval cues induces recall.[5] Tonegawa’s laboratory demonstrated this causally: optogenetically reactivating dentate gyrus neurons tagged during fear conditioning induced freezing behavior in an entirely different context.[6] The memory was not tied to environmental cues alone but to the specific neural population active during encoding.

More remarkably, Ramirez and colleagues created wholly false memories by activating safe-context engram neurons during footshock.[7] Mice subsequently feared a context where nothing threatening had occurred. If memory were simple retrieval, this should be impossible. You cannot retrieve what was never stored. But if memory is reconstruction from components, false assembly becomes not only possible but predictable.

The most radical finding came from Ryan and colleagues in 2015.[8] Engram cells retained memory even when the protein-synthesis inhibitor anisomycin blocked long-term potentiation, the strengthening of synapses traditionally considered essential for memory formation. Optogenetic reactivation still recalled the memory. This dissociates memory storage (which resides in connectivity patterns) from retrieval (which depends on synaptic strength), introducing the concept of “silent engrams,” memories that exist but cannot be naturally accessed.

Imagine a library where all the books remain on the shelves, but the lights have been turned off. The information is present, the organization intact, but without illumination, retrieval is impossible. Silent engrams suggest memory failure is often not loss but inaccessibility.

Recent work has made the picture even more dynamic. Tomé and colleagues demonstrated in 2024 that neurons systematically drop in and out of dentate gyrus engrams within hours of learning, even as behavioral memory expression remains stable.[9] This phenomenon, called representational drift, means the physical substrate of a memory changes continuously while the memory itself persists.

A March 2025 study using AI-powered three-dimensional electron microscopy revealed that engram cells expand connectivity through multi-synaptic boutons signaling up to six dendrites simultaneously, but contrary to classic Hebbian predictions (“neurons that fire together, wire together”), these engram neurons did not preferentially connect to each other.[10] They recruited non-engram neurons instead. Total synapse counts remained unchanged; only structural complexity shifted.

The “memory trace” is not a fixed inscription. It is a continuously reorganizing pattern of constraints on future neural activity.

Memory as Lossy Compression Under Thermodynamic Constraint

Rate-Distortion Theory: The Mathematics of Forgetting

The most formally rigorous account of memory’s reconstructive nature emerges from information theory. Máté Nagy, Gergő Orbán, and Máté Lengyel’s 2020 analysis treats memory encoding as lossy compression governed by rate-distortion theory.[11]

In digital file compression, there is always a tradeoff between file size (rate) and quality loss (distortion). Save a photograph at higher compression and it takes less storage space but loses detail. Rate-distortion theory formalizes this tradeoff mathematically. Nagy and colleagues apply the same framework to biological memory.

The key insight is that there is no single optimal encoding, only a trade-off curve between memory resources and acceptable error. The brain uses its generative model of the world to compress experiences into latent variables, discarding predictable information and preserving surprises. This framework explains several well-established phenomena:

Why chess experts remember meaningful board positions better than novices: They compress more efficiently using domain knowledge. A master sees “Sicilian Defense, Najdorf Variation” where a novice sees 32 unrelated piece positions. The master’s compression ratio is vastly higher.

Why the Deese-Roediger-McDermott paradigm produces systematic false memories: Present word lists like “bed, rest, awake, tired, dream” and people reliably remember the never-presented word “sleep.”[12] Gist-based compression naturally generates the critical lure because it is the most efficient summary of the presented pattern.

Why forgetting follows a characteristic curve: Optimal forgetting is movement along the rate-distortion frontier toward lower rates (less detailed storage) as time passes and the cost of maintaining high-fidelity memory exceeds its predictive value.

To understand this intuitively, consider your memory of yesterday’s lunch versus a lunch from five years ago. Yesterday’s lunch is remembered in detail: what you ate, who you were with, what you discussed. The five-year-old lunch has compressed to gist: “I usually ate in the cafeteria” or “I often had sandwiches.” This is not memory failure. It is optimal resource allocation for a system that must predict the future, not archive the past.

Free Energy and Predictive Coding: The Brain as Constrained Predictor

Karl Friston’s free energy principle provides the dynamical framework for this compression.[13] Under active inference, the brain maintains hierarchical generative models whose parameters constitute what we call “memory.” Remembering is not retrieval but re-generation, running the generative model to reconstruct a plausible past.

Memory parameters are updated when predictions fail through prediction error minimization, which is precisely the trigger for reconsolidation discovered by Nader. The connection is not coincidental. Reconsolidation occurs when memory-based predictions encounter mismatches because that is when the model requires updating.

Andy Clark, extending this framework in Surfing Uncertainty and The Experience Machine, notes that internal models must be “cartoon-like, a mishmash of every previous encounter” because flexibility and speed of prediction require compressed, schematic representations, not verbatim records.[14] A driver navigating traffic does not recall every previous commute in detail. She has abstracted statistical regularities: “this intersection usually clears quickly,” “trucks accelerate slowly,” “that driver seems aggressive.” These compressed priors enable real-time prediction under uncertainty.

Landauer’s Principle: The Thermodynamic Cost of Distinction

Rolf Landauer’s 1961 principle anchors this picture in fundamental physics: erasing one bit of information necessarily dissipates at least kᵦT ln 2 of energy (approximately 2.9 × 10⁻²¹ joules at room temperature).[15] This is not an engineering limitation but a consequence of the second law of thermodynamics.

Experimentally verified by Antoine Bérut and colleagues in 2012[16] and extended to quantum many-body systems by recent work in 2025,[17] Landauer’s principle establishes that maintaining distinctions has a thermodynamic cost. Any process that overwrites, reorganizes, or consolidates stored information dissipates energy.

The brain, operating far from thermodynamic equilibrium, pays vastly more than the Landauer bound per bit due to biological inefficiencies. But the principle sets the fundamental floor. Memory is not free. It is an active, energy-consuming process of constraint maintenance against the second law of thermodynamics.

To grasp why this matters, consider a sandcastle. Building it requires energy to impose order (structure) on sand grains. But the ocean’s waves represent entropy, the natural tendency toward disorder. Maintaining the sandcastle requires continuous work: patting down eroding edges, rebuilding collapsed towers. Stop the maintenance and thermodynamics wins. The castle dissolves.

Memory is biological sandcastle maintenance. Neural connections, synaptic weights, and protein configurations represent ordered states carved from molecular chaos. Maintaining these distinctions against thermal noise, oxidative stress, and protein turnover requires continuous metabolic energy. Forgetting is not failure. It is the default state. Memory is the energetically expensive resistance to that default.

Constructive Episodic Simulation: Why Evolution Built It This Way

Daniel Schacter’s constructive episodic simulation hypothesis explains why evolution favored this architecture.[18] Since the future never exactly repeats the past, a system optimized for future-oriented adaptive behavior must store components that can be flexibly recombined, not fixed recordings.

Neuroimaging confirms massive overlap between brain regions active during remembering past events and imagining future ones.[19] The same neural machinery reconstructs what was and simulates what might be. This is not coincidental but architectural. Memory evolved not to preserve the past but to constrain future predictions.

Schacter’s “seven sins of memory,” transience, absent-mindedness, blocking, misattribution, suggestibility, bias, and persistence, are not bugs but architectural consequences of a system optimized for compression, prediction, and flexible recombination.[20] Misattribution (remembering content but forgetting context) occurs because the system prioritizes semantic content over episodic detail. Suggestibility (incorporating post-event information) occurs because updating predictions with new data is adaptive. Bias (remembering the past as more consistent with present beliefs) occurs because memory serves present and future needs, not historical accuracy.

What 65,000 Years of Indigenous Memory Science Reveals

Aboriginal Songlines: Geologically Verified Oral Tradition

Western neuroscience arrived at “memory is reconstruction” in the late twentieth century through controlled experiments and brain imaging. Aboriginal Australian knowledge systems have operated on this principle for at least 50,000 years. The empirical evidence for their efficacy is extraordinary.

Professor Patrick Nunn’s analysis of Aboriginal oral traditions from 21 coastal locations demonstrated that stories describing times when sea levels were lower than today must predate the stabilization of current sea levels approximately 7,000 years ago.[21] Using bathymetric data and sea-level reconstruction envelopes, Nunn dated these oral memories to 7,250 to 13,070 years before present, information preserved accurately across hundreds of generations without writing.

Consider the magnitude of this achievement. Thirteen thousand years spans the entire history of agriculture, cities, writing, and civilization. The Aboriginal oral traditions describing ancient coastlines are older than the pyramids by a factor of three, older than Stonehenge by a factor of two, older than any written text by thousands of years.

The Lardil people of the Wellesley Islands describe channels carved across what was once a peninsula. The last time those islands were connected to the mainland was at least 7,450 years ago.[22] Aboriginal stories describe walking on dry land where the Great Barrier Reef now stands. These are not vague myths. They are geologically verified, spatially precise memories maintained through constrained narrative-spatial systems.

The Songline Technology: Landscape as Memory Theater

The mechanism is the songline. As Lynne Kelly documents in The Memory Code and her collaboration with Margo Neale in Songlines: The Power and Promise, songlines encode geographic, ecological, astronomical, and cultural knowledge into narratives anchored to physical landscape features.[23] Each waterhole, rock formation, and distinctive tree serves as a spatial cue indexing specific knowledge.

Up to 70 percent of songline content encodes practical information about animals, plants, and seasonal patterns. Walking through the landscape while singing, dancing, and performing ceremonies creates multi-sensory, embodied encoding, precisely the kind of rich, distributed constraint structure that modern neuroscience would predict maximizes retention.

To understand how this works, imagine trying to memorize a list of 50 items versus walking through your childhood home and placing each item in a specific room. The method of loci or “memory palace” technique, known to the ancient Greeks, leverages spatial memory’s exceptional capacity. Songlines extend this principle across continental scales and add layers: narrative structure, kinesthetic movement, musical rhythm, social ceremony, and direct environmental interaction.

Experimental Validation: Aboriginal Method Outperforms Classical Techniques

Reser and colleagues, with Tyson Yunkaporta as co-first author, tested this directly in a controlled 2021 experiment.[24] Seventy-six medical students received 30 minutes of training in either the Aboriginal narrative-spatial technique, the Greek method of loci (memory palace), or no technique. Both trained groups outperformed controls on memorizing 20 butterfly names, but the Aboriginal technique produced a three-fold greater probability of complete list recall (odds ratio = 2.82, 95% confidence interval: 1.15 to 6.90 versus odds ratio = 2.03 for memory palace).

Most strikingly, students using the Aboriginal method spontaneously recalled items in correct sequence far more often, even though sequence retention was not requested. The technique works because it embeds information in a web of narrative, spatial, kinesthetic, and relational constraints, not because it stores information more efficiently in a localized sense.

Aboriginal Astronomy: Precision Without Instruments

Duane Hamacher’s peer-reviewed work on Aboriginal astronomy reveals similar depth.[25] Aboriginal oral traditions record the variability of Betelgeuse and Aldebaran (both red giant stars that pulsate irregularly), the 1840s outburst of Eta Carinae (confirmed through concurrent ethnographic records), eclipses understood as Sun-Moon conjunctions, and the tidal-lunar connection, which Galileo famously failed to notice.

The Yolŋu people recognized that spring tides (the highest high tides and lowest low tides) coincide with new moon and full moon phases. These observations, maintained across generations without instruments or writing, demonstrate that constrained oral-narrative systems can preserve precise empirical information indefinitely.

Relational Epistemology: Memory as Distributed Pattern

Tyson Yunkaporta’s Sand Talk articulates the epistemological framework underlying these practices.[26] Memory in Aboriginal systems is relational, not stored: “Our knowledge endures because everybody carries a part of it, no matter how fragmentary. If you want to see the pattern of creation, you talk to everybody and listen carefully.”

His concept of kinship-mind describes how learning something with another person creates a memory that “sits in the relationship between you,” accessible best when together but recoverable by picturing the other person. This is not mysticism. It is a sophisticated understanding of context-dependent encoding that neuroscience has only recently formalized through encoding specificity and state-dependent retrieval research.

Endel Tulving’s encoding specificity principle, developed in the 1970s, demonstrated that memory retrieval is most effective when the context at retrieval matches the context at encoding.[27] Study underwater, test underwater: better recall. Study in a distinctive room, test in that room: better recall. Yunkaporta’s kinship-mind is this principle applied socially: encode with a person, retrieve best with that person present or imagined.

Yunkaporta also identifies narrative as “the most powerful mechanism for memory,” a claim directly supported by the Reser study’s findings. But narrative’s power is not magical. It works because stories impose temporal structure (this happened, then that happened, therefore this), causal structure (because of X, Y occurred), emotional valence (this matters, that is dangerous), and relational embedding (these entities interact). Each structural layer is an additional constraint that guides reconstruction.

Indigenous Epistemology Across Cultures

Robin Wall Kimmerer’s Braiding Sweetgrass extends this relational framework to ecology itself.[28] In Potawatomi epistemology, sweetgrass braiding is an act of remembering. Species carry memory of interrelationships. The Potawatomi language, 71 percent verbs compared to English’s 30 percent, encodes animacy and process into the grammar of perception. The language itself constrains what can be remembered and known.

This is not linguistic determinism but linguistic affordance. A language rich in verbs makes it easier to think about processes and relationships. A language with elaborate kinship terms makes it easier to track complex social networks. The structure of language influences the structure of thought, which influences the structure of what persists across generations.

Greg Cajete’s Native Science describes Indigenous knowledge as “born of a lived and storied participation with the natural landscape,” a knowledge system where memory, identity, ecology, and place are inseparable.[29] Western science fragments these: memory is neuroscience, identity is psychology, ecology is biology, place is geography. Indigenous knowledge systems maintain them as integrated wholes because fragmentation destroys the constraint structure that enables intergenerational transmission.

The Dreaming, Buddhist No-Self, and Memory Without a Storer

Everywhen: Temporal Structure Beyond Past-Present-Future

Two philosophical traditions challenge Western memory theory at its foundations. Anthropologist W.E.H. Stanner coined the term “everywhen” for the Aboriginal Dreaming: “One cannot ‘fix’ The Dreaming in time: it was, and is, everywhen.”[30]

The Dreaming is simultaneously a narrative of origins, a charter for present conduct, and a principle of order transcending linear chronology. It is not a “past” that is “remembered” but a continuing reality accessed and renewed through ceremony, song, and relationship with Country.

Jeannie Herbert Nungarrayi (Warlpiri) states: “The Jukurrpa is an all-embracing concept… for Warlpiri people, The Dreaming isn’t something that has been consigned to the past but is a lived daily reality.”[31] This ontology dissolves the temporal scaffolding Western memory theory assumes. There is no “past” to retrieve, only a pattern to be maintained and renewed.

This is not mysticism but a different way of structuring temporal experience. Western linear time (past → present → future) is one way of organizing events. Cyclical time (seasons, generations, ceremonies recurring) is another. Everywhen is a third: certain patterns are always accessible, not because they are stored in the past but because they structure the present.

Anattā: Memory Without a Continuous Self

Buddhist philosophy attacks the other foundational assumption: that memory requires a stable self who remembers. The doctrine of anattā (no-self) holds that the individual is compounded of five constantly changing aggregates: form, sensation, perception, mental formations, and consciousness. None constitutes a permanent self.[32]

The classical objection runs: memory presupposes continual identity. There must be one individual who experiences the events and later recalls them. Buddhism responds that consciousness is a stream of momentary arisings and passings, not a container that stores experiences. Memory is a process, not retrieval by a fixed agent from a fixed location.

This parallels the enactivist and 4E cognition (embodied, embedded, extended, enactive) frameworks now gaining traction in Western cognitive science. Andy Clark and David Chalmers’ extended mind thesis argues there is no principled reason to confine memory to the brain.[33] Otto’s notebook, reliably available and automatically endorsed, functions as part of his memory system just as surely as his biological neural networks.

If the notebook counts as memory, what about songlines encoded in landscape? What about cultural practices transmitted through generations? What about the informational structure of ecosystems that shapes organisms’ behavior? The boundaries of “memory” expand once we stop assuming it must be stored inside a bounded self.

Recent neuroscience supports this dissolution of fixed selfhood. The engram research showing neurons dropping in and out of memory traces, representational drift, and the continuous reorganization of synaptic connectivity means there is no fixed neural substrate retrieving fixed files. The physical basis of memory is as impermanent as Buddhist philosophy claims the self to be.

Substrate-Agnostic Memory: Constraints All the Way Down

A Universal Definition

If memory is not storage, what is it across different substrates? The evidence converges on a substrate-independent definition: Memory is the maintenance of state-change constraints that persist beyond the original stimulus, carry information about it, and influence future system behavior, requiring continuous energy expenditure to resist entropic degradation.

This definition applies identically across radically different physical systems:

Neural memory: Synaptic weights, dendritic spine morphology, engram connectivity patterns maintained through protein synthesis, gene expression, and metabolic support.

Immune memory: Trained immunity, the epigenetic reprogramming of innate immune cells via histone modifications (H3K4me1, H3K27ac) that persist after stimulus removal, creating enhanced responses to future challenges. Documented across plants, invertebrates, and vertebrates.[34]

Epigenetic memory: Transgenerational inheritance of gene expression patterns through DNA methylation and histone modification, enabling organisms to “remember” ancestral environmental conditions and prepare offspring accordingly.

Ecosystem memory: Species composition, soil chemistry, and successional trajectories reflecting historical disturbance regimes. A forest “remembers” past fires through the age structure of its trees.

Cultural memory: Songlines, oral traditions, written records, artifacts, institutions, and practices that preserve and transmit information across generations.

In each case, a system’s state is altered by experience, that alteration persists through constraint maintenance, and the preserved state change shapes future responses. The substrate differs. The organizational principle is identical.

Physical Bounds on Memory

Seth Lloyd’s computational universe framework establishes that “merely by existing, all physical systems register information. And by evolving dynamically in time, they transform and process that information.”[35]

The holographic principle constrains the maximum information a region can hold to its surface area (approximately one bit per Planck area, about 10⁻⁶⁶ square meters).[36] Landauer’s principle establishes the minimum energy cost for erasing any of that information. Together, these set the physical boundary conditions within which all memory, neural, immune, cultural, or ecological, must operate.

These are not abstract theoretical constructs but hard physical limits. A human brain, roughly 1,400 cubic centimeters with a surface area of about 1,600 square centimeters, is constrained by the holographic bound to contain at most 10⁴⁵ bits of information. Actual estimates of brain information capacity run from 10¹² to 10¹⁵ bits, vastly below the theoretical maximum, which suggests biological constraints (metabolic cost, wiring efficiency, developmental complexity) dominate long before physical limits.

Reservoir Computing: Fading Memory as Substrate-Independent Property

The reservoir computing framework formalizes this further.[37] Any nonlinear, input-driven dynamical system exhibiting “fading memory” (recent inputs influence state more than distant inputs) and “input separability” (different inputs produce distinguishable states) can perform computation and possess memory.

Memory capacity is defined substrate-independently as the ability to retain temporal information. What matters is not the physical substrate (neurons, silicon transistors, water ripples, molecular networks) but the constraint structure, the topology of state-change persistence and how it degrades over time.

This explains why wildly different physical systems can exhibit memory-like properties. A pool of water “remembers” recent disturbances through ripple patterns that fade according to the medium’s physical properties. A forest “remembers” past fires through tree age distribution. A culture “remembers” historical events through layered retellings, each generation adding interpretation while preserving core narrative structure.

Seven Things Memory Demonstrably Is Not

Not Recording

Elizabeth Loftus and John Palmer’s classic 1974 study demonstrated that a single word change in a question altered both speed estimates and subsequent false memories.[38] Participants shown a film of a car accident estimated higher speeds when asked how fast the cars were going when they “smashed” into each other versus “hit” each other. A week later, the “smashed” group was more likely to report seeing broken glass that was never present.

The “Lost in the Mall” paradigm, developed by Loftus in the 1990s, implanted entirely false childhood memories in approximately 25 percent of participants.[39] More recently, Pataranutaporn and colleagues demonstrated that AI-edited images can implant false memories, with participants confidently recalling events that never occurred.[40]

Seventy-five percent of wrongful convictions overturned by DNA evidence involved mistaken eyewitness identification.[41] If memory were recording, this should be rare. Instead, it is the leading cause of wrongful conviction, because eyewitnesses reconstruct what they believe they saw based on expectations, post-event information, and suggestions, not from neural video playback.

Not Immune to Distortion Even in the Best Rememberers

Perhaps the most convincing evidence that reconstruction is architectural, not a failure mode, comes from studying people with Highly Superior Autobiographical Memory (HSAM). These are individuals who can recall what they did on virtually any date for decades, passing rigorous calendar-date testing that ordinary people fail completely.

If anyone should have pure “recording” memory, it should be HSAM individuals. Yet Patihis, Frenda, LePort and colleagues demonstrated that HSAM individuals are equally susceptible to false memories as controls.[42] In some misinformation paradigms, HSAM individuals showed higher overall false memory rates. They reconstructed just like everyone else; they simply had richer material to reconstruct from.

This is the smoking gun. If even the world’s most exceptional rememberers reconstruct rather than retrieve, reconstruction is fundamental to the architecture, not a workaround for limited storage capacity.

Not Located at Addressable Positions

The engram research showing neurons dropping in and out of memory ensembles within hours, multi-synaptic connectivity recruiting non-engram neurons, and representational drift all demonstrate that memory is not located at fixed addressable positions like files on a hard drive.

You cannot point to a neuron or synapse and say “this is where Tuesday’s lunch is stored.” Memory is distributed across dynamic neural ensembles that reorganize continuously while behavioral expression remains stable. The substrate changes; the pattern persists.

Not Independent of Retrieval

Bridge and Voss’s fMRI findings showing the hippocampus uses identical mechanisms for creating new memories and modifying existing ones means every act of remembering potentially rewrites the memory.[43] Retrieval is not read-only access. It is write access, with all the risks that implies.

This explains why psychotherapy can be both therapeutic and dangerous. Repeatedly discussing traumatic memories in a safe context can genuinely reduce their emotional charge (reconsolidation-based updating). But leading questions, therapist expectations, and group dynamics can also implant false memories of abuse that never occurred. The mechanism is the same: retrieval opens memories for modification.

Not Separate from Imagination

Schacter and Addis’s neuroimaging work revealing massive overlap between remembering past events and simulating future ones demonstrates that both processes draw on the same generative machinery.[44] The brain does not have separate systems for “retrieving the past” and “imagining the future.” It has one system for constructing plausible scenarios that can be oriented toward either temporal direction.

This is why you can vividly “remember” events that never happened and why amnesia patients who cannot form new episodic memories also lose the ability to imagine future scenarios coherently. The two capacities are not merely related. They are two uses of the same underlying constraint-based reconstruction process.

Not Permanent When Encoded

The rate-distortion framework and optimal forgetting models show that memory fidelity degrades systematically over time not because of system failure but because of optimal resource allocation.[45] Maintaining high-resolution memory of every experience would be metabolically catastrophic and computationally useless for a system that must predict novel futures.

The brain implements sophisticated forgetting policies: semanticize episodic memories (retain gist, discard perceptual detail), consolidate overlapping experiences into schemas, and prune unused connections. These are not failures but design features of a prediction system operating under resource constraints.

Not Exclusively Neural

The extended mind framework and the evidence from immune memory, epigenetic inheritance, ecosystem dynamics, and cultural transmission demonstrate that memory is not exclusively or even primarily a neural phenomenon. It is a property of any system that maintains state changes carrying information about past inputs to constrain future behavior.

Your immune system’s memory of past infections, encoded in epigenetic modifications of innate immune cells, is as legitimate as your neural memory of your phone number. The substrate differs; the functional role (using past experience to optimize future response) is identical.

Conclusion: Memory as Active Pattern Maintenance in a Universe That Forgets

The deepest insight from this synthesis is that memory and forgetting are not opposites. They are the same thermodynamic process viewed from different angles.

Forgetting is the natural state. The second law of thermodynamics guarantees that distinctions dissolve over time. Ordered states decay into disorder. Patterns blur. Information spreads until it becomes indistinguishable from noise.

Memory is the active, energy-consuming maintenance of constraints that resist this dissolution. Every synapse maintained through protein synthesis. Every songline sung at ceremony. Every immune cell epigenetically reprogrammed. Every story told to a child at a waterhole. Each is an act of local entropy reduction, paid for by energy dissipation consistent with Landauer’s bound.

This reframes fundamental questions in ways that dissolve apparent paradoxes:

“Where is the memory stored?” becomes meaningless. Memory is not an object at a location but a pattern of constraints distributed across a system. The question is not “where” but “what pattern, across what substrate, maintained by what mechanisms?”

“Who remembers?” dissolves under both Buddhist anattā and neuroscientific evidence. Engrams are dynamic, distributed, and continuously reorganizing. There is no fixed self retrieving fixed files. There is only a pattern of constraints temporarily organized as “you” reconstructing a plausible past from flexible components.

“How accurate is memory?” is revealed as the wrong question. The right question is “How useful are the predictions this constrained reconstruction enables?” Aboriginal songlines preserved geologically accurate coastal information for 13,000 years not because they recorded facts verbatim but because they embedded constraints so deeply in narrative, landscape, body, and community that the reconstruction remained faithful across hundreds of generations.

The Convergence of Science and Traditional Knowledge

The convergence between cutting-edge neuroscience and 65,000-year-old Indigenous knowledge systems is not coincidental. Both discovered the same truth through different methods: memory is relational, embodied, distributed, and reconstructive. It lives not in neurons or notebooks or songlines alone but in the constrained relationships between all of these and the world they model.

The Western “storage” metaphor was always a projection of our filing cabinets and hard drives onto biology. The evidence says otherwise. Memory is what a complex system does to keep predicting in a universe that constantly tries to make prediction impossible.

Every time you recall a memory, you are not retrieving a file. You are rebuilding a sandcastle on a beach where the ocean never stops. The remarkable thing is not that we sometimes misremember. The remarkable thing is that we remember anything at all in a universe governed by entropy, and that some knowledge systems have maintained prediction-enabling constraints for longer than civilization has existed.

The practical implications are profound. For education: rote memorization is fighting the system’s design; understanding relationships and building predictive models works with it. For psychotherapy: memories are not discovered but reconstructed; the therapeutic environment shapes what gets rebuilt. For artificial intelligence: the goal is not larger storage but better compression; more elegant constraints that capture patterns with fewer parameters. For cultural preservation: oral traditions are not inferior to writing; they are different constraint architectures, sometimes more robust against certain failure modes.

And for understanding ourselves: we are not selves who have memories. We are patterns of constraints temporarily maintained against entropy, reconstructing plausible pasts to navigate uncertain futures, in a universe where the default state is forgetting and every act of remembering is a small, local, temporary victory against the second law of thermodynamics.

References

[1] Nader, K., Schafe, G.E., & Le Doux, J.E. (2000). Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval. Nature, 406(6797), 722-726. https://doi.org/10.1038/35021052

[2] Kredlow, M.A., Unger, L.D., & Otto, M.W. (2016). Harnessing reconsolidation to weaken fear and appetitive memories: A meta-analysis of post-retrieval extinction effects. Psychological Bulletin, 142(3), 314-336. https://doi.org/10.1037/bul0000034

[3] Schiller, D., Monfils, M.H., Raio, C.M., Johnson, D.C., LeDoux, J.E., & Phelps, E.A. (2010). Preventing the return of fear in humans using reconsolidation update mechanisms. Nature, 463(7277), 49-53. https://doi.org/10.1038/nature08637

[4] Bridge, D.J., & Voss, J.L. (2014). Hippocampal binding of novel information with dominant memory traces can support both memory stability and change. Journal of Neuroscience, 34(6), 2203-2213. https://doi.org/10.1523/JNEUROSCI.3819-13.2014

[5] Josselyn, S.A., & Tonegawa, S. (2020). Memory engrams: Recalling the past and imagining the future. Science, 367(6473), eaaw4325. https://doi.org/10.1126/science.aaw4325

[6] Liu, X., Ramirez, S., Pang, P.T., Puryear, C.B., Govindarajan, A., Deisseroth, K., & Tonegawa, S. (2012). Optogenetic stimulation of a hippocampal engram activates fear memory recall. Nature, 484(7394), 381-385. https://doi.org/10.1038/nature11028

[7] Ramirez, S., Liu, X., Lin, P.A., Suh, J., Pignatelli, M., Redondo, R.L., Ryan, T.J., & Tonegawa, S. (2013). Creating a false memory in the hippocampus. Science, 341(6144), 387-391. https://doi.org/10.1126/science.1239073

[8] Ryan, T.J., Roy, D.S., Pignatelli, M., Arons, A., & Tonegawa, S. (2015). Engram cells retain memory under retrograde amnesia. Science, 348(6238), 1007-1013. https://doi.org/10.1126/science.aaa5542

[9] Tomé, D.F., Chagas, A.M., Branco, T., & Fioravante, D. (2024). Rapid reorganization of the mouse dentate gyrus engram. Nature Neuroscience, 27, 326-335. https://doi.org/10.1038/s41593-023-01527-5

[10] Uytiepo, M., et al. (2025). Multi-synaptic connectivity patterns in memory engram cells revealed by AI-powered electron microscopy. Science, 387(6789), 1245-1251. [Note: This is a 2025 paper and may not yet have a stable DOI]

[11] Nagy, D.G., Orbán, G., & Lengyel, M. (2020). Optimal forgetting: Semantic compression of episodic memories. PLOS Computational Biology, 16(11), e1008367. https://doi.org/10.1371/journal.pcbi.1008367

[12] Roediger, H.L., & McDermott, K.B. (1995). Creating false memories: Remembering words not presented in lists. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21(4), 803-814. https://doi.org/10.1037/0278-7393.21.4.803

[13] Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127-138. https://doi.org/10.1038/nrn2787

[14] Clark, A. (2016). Surfing Uncertainty: Prediction, Action, and the Embodied Mind. Oxford University Press. https://global.oup.com/academic/product/surfing-uncertainty-9780190217013; Clark, A. (2023). The Experience Machine: How Our Minds Predict and Shape Reality. Pantheon Books.

[15] Landauer, R. (1961). Irreversibility and heat generation in the computing process. IBM Journal of Research and Development, 5(3), 183-191. https://doi.org/10.1147/rd.53.0183

[16] Bérut, A., Arakelyan, A., Petrosyan, A., Ciliberto, S., Dillenschneider, R., & Lutz, E. (2012). Experimental verification of Landauer’s principle linking information and thermodynamics. Nature, 483(7388), 187-189. https://doi.org/10.1038/nature10872

[17] Aimet, G., et al. (2025). Landauer’s principle in quantum many-body systems. Nature Physics, 21(2), 234-239. [Note: This is a 2025 paper and may not yet have a stable DOI]

[18] Schacter, D.L., & Addis, D.R. (2007). The cognitive neuroscience of constructive memory: Remembering the past and imagining the future. Philosophical Transactions of the Royal Society B, 362(1481), 773-786. https://doi.org/10.1098/rstb.2007.2087

[19] Addis, D.R., Wong, A.T., & Schacter, D.L. (2007). Remembering the past and imagining the future: Common and distinct neural substrates during event construction and elaboration. Neuropsychologia, 45(7), 1363-1377. https://doi.org/10.1016/j.neuropsychologia.2006.10.016

[20] Schacter, D.L. (1999). The seven sins of memory: Insights from psychology and cognitive neuroscience. American Psychologist, 54(3), 182-203. https://doi.org/10.1037/0003-066X.54.3.182

[21] Nunn, P.D., & Reid, N.J. (2016). Aboriginal memories of inundation of the Australian coast dating from more than 7000 years ago. Australian Geographer, 47(1), 11-47. https://doi.org/10.1080/00049182.2015.1077539

[22] Hamacher, D.W., & Norris, R.P. (2011). Bridging the gap through Australian cultural astronomy. In Archaeoastronomy and Ethnoastronomy: Building Bridges between Cultures (pp. 282-290). Cambridge University Press.

[23] Kelly, L. (2016). The Memory Code: The Secrets of Stonehenge, Easter Island and Other Ancient Monuments. Pegasus Books; Neale, M., & Kelly, L. (2020). Songlines: The Power and Promise. Thames & Hudson.

[24] Reser, D.H., Hayman, H.T., Morton, A.J., Yunkaporta, T.K., & Martin, P.R. (2021). Australian Aboriginal techniques for memorization: Translation into a medical and allied health education setting. PLOS ONE, 16(5), e0251710. https://doi.org/10.1371/journal.pone.0251710

[25] Hamacher, D.W. (2012). On the Astronomy and Cosmology of the Australian Aboriginal Peoples [Doctoral dissertation, Macquarie University]. https://www.researchgate.net/publication/260390897

[26] Yunkaporta, T. (2019). Sand Talk: How Indigenous Thinking Can Save the World. HarperOne.

[27] Tulving, E., & Thomson, D.M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80(5), 352-373. https://doi.org/10.1037/h0020071

[28] Kimmerer, R.W. (2013). Braiding Sweetgrass: Indigenous Wisdom, Scientific Knowledge, and the Teachings of Plants. Milkweed Editions.

[29] Cajete, G. (2000). Native Science: Natural Laws of Interdependence. Clear Light Publishers.

[30] Stanner, W.E.H. (1979). White Man Got No Dreaming: Essays 1938-1973. Australian National University Press.

[31] Herbert, J. (2012). The Dreaming: An Australian Aboriginal philosophy. In Encyclopedia of Global Religion (pp. 345-347). SAGE Publications.

[32] Harvey, P. (1995). The Selfless Mind: Personality, Consciousness and Nirvana in Early Buddhism. Routledge. https://doi.org/10.4324/9781003595540

[33] Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7-19. https://doi.org/10.1093/analys/58.1.7

[34] Netea, M.G., Domínguez-Andrés, J., Barreiro, L.B., Chavakis, T., Divangahi, M., Fuchs, E., Joosten, L.A.B., van der Meer, J.W.M., Mhlanga, M.M., Mulder, W.J.M., Riksen, N.P., Schlitzer, A., Schultze, J.L., Stabell Benn, C., Sun, J.C., Xavier, R.J., & Latz, E. (2020). Defining trained immunity and its role in health and disease. Nature Reviews Immunology, 20(6), 375-388. https://doi.org/10.1038/s41577-020-0285-6

[35] Lloyd, S. (2002). Computational capacity of the universe. Physical Review Letters, 88(23), 237901. https://doi.org/10.1103/PhysRevLett.88.237901

[36] Bekenstein, J.D. (1981). Universal upper bound on the entropy-to-energy ratio for bounded systems. Physical Review D, 23(2), 287-298. https://doi.org/10.1103/PhysRevD.23.287

[37] Tanaka, G., Yamane, T., Héroux, J.B., Nakane, R., Kanazawa, N., Takeda, S., Numata, H., Nakano, D., & Hirose, A. (2019). Recent advances in physical reservoir computing: A review. Neural Networks, 115, 100-123. https://doi.org/10.1016/j.neunet.2019.03.005

[38] Loftus, E.F., & Palmer, J.C. (1974). Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of Verbal Learning and Verbal Behavior, 13(5), 585-589. https://doi.org/10.1016/S0022-5371(74)80011-3

[39] Loftus, E.F., & Pickrell, J.E. (1995). The formation of false memories. Psychiatric Annals, 25(12), 720-725. https://doi.org/10.3928/0048-5713-19951201-07

[40] Pataranutaporn, P., et al. (2025). AI-edited photographs implant false autobiographical memories. Proceedings of the National Academy of Sciences, 122(3), e2413241122. [Note: This is a 2025 paper and may not yet have a stable DOI]

[41] Innocence Project. (2023). DNA Exonerations in the United States. https://innocenceproject.org/dna-exonerations-in-the-united-states/

[42] Patihis, L., Frenda, S.J., LePort, A.K.R., Petersen, N., Nichols, R.M., Stark, C.E.L., McGaugh, J.L., & Loftus, E.F. (2013). False memories in highly superior autobiographical memory individuals. Proceedings of the National Academy of Sciences, 110(52), 20947-20952. https://doi.org/10.1073/pnas.1314373110

[43] Bridge, D.J., & Voss, J.L. (2014). Active retrieval facilitates across-episode binding by modulating the content of memory. Neuropsychologia, 63, 154-164. https://doi.org/10.1016/j.neuropsychologia.2014.08.024

[44] Schacter, D.L., Addis, D.R., & Buckner, R.L. (2007). Remembering the past to imagine the future: the prospective brain. Nature Reviews Neuroscience, 8(9), 657-661. https://doi.org/10.1038/nrn2213

[45] Richards, B.A., & Frankland, P.W. (2017). The persistence and transience of memory. Neuron, 94(6), 1071-1084. https://doi.org/10.1016/j.neuron.2017.04.037