Alexis O. Kaya - Theoretical neuroscience perspective
ABSTRACT
When a familiar name cannot be produced, an event cannot be voluntarily recalled, or a life history becomes fragmented by neurological disease, what has failed: the representation of the past, the mechanisms that retrieve it, or both? The phenomenology of forgetting does not discriminate these possibilities. Building on the distinction between memory persistence and functional accessibility, this article investigates retrieval as a dynamic, cue-dependent operation within distributed neural systems. It develops Neuronal Detours as a provisional, operational hypothesis: under some circumstances, access to a previously encoded representation can be restored through alternative cues, network configurations, or compensatory strategies even when an ordinarily effective retrieval configuration fails. The concept denotes reconfiguration of access, not a literal anatomical road to an intact, immutable memory. Evidence from encoding specificity, tip-of-the-tongue states, retrieval competition, and experimentally manipulated engrams supports the possibility of dissociation between memory expression and aspects of trace persistence. Yet this evidence does not establish universal recoverability, nor does improved performance prove that an identical original memory has been recovered. Alzheimer’s disease illustrates why impaired encoding, consolidation, retrieval, and representational integrity must be distinguished rather than arranged into a single necessary sequence. Cognitive reserve and rehabilitation suggest conditional resilience, not proof of preserved hidden memories. The article proposes five research tests, including cue diversification, longitudinal population recordings, causal circuit perturbation, fidelity-sensitive reinstatement, and clinical stratification. The cost of forgetting, on this account, is not merely information loss: it includes reduced access, altered reconstruction, and interruptions in the experienced continuity of personal history. The resulting framework preserves both a scientific opportunity and a strict limitation: an inaccessible memory is not necessarily erased, but its continued existence cannot be assumed.
Keywords: memory retrieval; functional accessibility; forgetting; engrams; cue-dependent recall; neural compensation; Neuronal Detours; Alzheimer’s disease; autobiographical memory; Neurotenacity.
1. THE MEMORY THAT CANNOT BE REACHED
A person recognizes a face and knows that its name is familiar, yet cannot produce the name. Later, without any obvious new learning, the word becomes available. Another person finds that a location or a scent evokes details of an event that had remained unavailable to voluntary recollection. These ordinary episodes invite a powerful intuition: the past can remain represented while its retrieval fails. That intuition is important, but it is not by itself an explanation. A return of apparently forgotten information demonstrates that some information was available to be recovered or reconstructed under the later conditions; it does not disclose the precise state of the representation during the interval, or prove that the recovered account is wholly accurate.
The original puzzle in The Cost of Forgetting is therefore not simply why memories decay. It is how a nervous system establishes, loses, and sometimes re-establishes access to previous experience. Contemporary research has long distinguished the acquisition and retention of information from its expression on a particular test. The principle of encoding specificity established that a cue’s effectiveness depends on the relation between its properties and the circumstances of encoding, rather than on its generic informativeness (Tulving and Thomson, 1973). Consequently, failure to recall under one set of cues cannot automatically be equated with the disappearance of every relevant trace.
This article follows the preceding Architecture of Forgetting in refusing to identify non-recall with erasure, but it addresses a different question. The earlier distinction is diagnostic: storage and accessibility may diverge. The present question is mechanistic and testable: under what conditions can accessibility be reconfigured, and how would a compensatory route differ from new learning, guessing, familiarity, or the construction of a false memory? The distinction matters most when a claim about the brain’s capacity to “find another road” is taken beyond metaphor.
There is an equally important clinical reason for caution. Failure to retrieve a childhood event, episodic impairment during aging, and difficulty forming new memories in Alzheimer’s disease do not necessarily share the same mechanism. A unified theory of forgetting must explain variation, rather than make all cases illustrations of one preferred account. It must also preserve a boundary between the measurable accessibility of a representation and the wider ethical and philosophical standing of the individual who remembers—or cannot remember—it.
2. FORGETTING IS AN OUTCOME, NOT A SINGLE PROCESS
Forgetting is usually measured as a change in performance: an item that was recalled is no longer produced, a recognition judgment becomes less accurate, or a learned association ceases to influence behavior. Performance is observable; the persistence of the underlying representation is not usually observed directly. The inferential gap is the first obstacle to a science of retrieval access. Several mechanisms may generate similar behavioral outcomes.
First, a representation may be degraded or lost. Synapses, neurons, and coordinated circuit activity can be disrupted by disease or injury, while newly encoded associations may fail to consolidate. Second, information may remain represented but insufficiently activated by available cues. Third, competing representations may gain an advantage, making the target difficult to express despite residual information. Retrieval-induced forgetting experiments show that practiced retrieval can disadvantage related, unpracticed material; the precise roles of competition and inhibition vary by paradigm and remain debated (Anderson, Bjork, and Bjork, 1994). Fourth, a representation may be accessible in one form but unavailable in another: familiarity can survive when source recollection fails, while semantic information can be retained despite failures of phonological word retrieval.
A further possibility is reconstruction with reduced fidelity. An answer may reappear but contain details acquired after the original event. The constructive nature of human episodic memory makes recovered recollection different from inspection of a permanently sealed archive (Schacter, 2022). A rise in confidence can accompany an increase in error as well as accuracy. Thus the phrases “the memory survived,” “the memory returned,” and “the original event was reconstructed correctly” name different propositions, each requiring separate evidence.
The distinctions can be made explicit. Retention concerns whether learning has left enduring, potentially measurable consequences. Accessibility concerns whether those consequences can be mobilized under specified conditions. Expression concerns the behavioral or neural response generated during a test. Fidelity concerns whether the resulting content corresponds to the target experience or independently established facts. These dimensions are related but not interchangeable. Their separation prevents a common circular inference: one cannot establish that a hidden memory persisted merely by defining every subsequent recovery as access to an intact original.
3. CUES, COMPETITION, AND THE ARCHITECTURE OF RETRIEVAL
The clearest evidence for conditional access comes from experimental manipulations of retrieval cues. Encoding specificity predicts that an otherwise weak cue can be effective if it reinstates a distinctive relation formed during learning. Conversely, a generally familiar cue may fail when it poorly matches the encoded episode (Tulving and Thomson, 1973). The critical quantity is not a cue’s strength in isolation but its relationship to the encoded representation and competing associations.
The tip-of-the-tongue phenomenon provides a second case. People may retrieve semantic features, syllable count, or fragments of phonology without successfully naming a familiar object or individual. Studies suggest that lexical access and phonological retrieval can be differentially compromised, with frequency of use and aging contributing to word-finding failures (Burke et al., 1991). Such observations support incomplete activation or temporarily insufficient retrieval, rather than proof that an entire autobiographical memory exists unchanged behind a blocked neural pathway.
Competition also matters. Retrieval is selective because the same cue can activate several related responses. Reinforcing one target may suppress, weaken, or reduce expression of another under some conditions. In laboratory paradigms, deliberate attempts to prevent retrieval can decrease later recall (Anderson and Green, 2001). These results complicate any simple image of a passive gate in front of stored information. Retrieval conditions can modify what becomes accessible, and the act of remembering may itself change the future retrieval landscape.
A useful architectural language must therefore accommodate distributed activation, representational overlap, contextual matching, inhibitory control, and changing goals. A cue is not necessarily a signpost pointing toward a fixed location; it alters the state of a network so that particular patterns become more likely. Several cue configurations may produce a similar behavioral memory, and a single cue may evoke divergent recollections as attention, goals, and knowledge change. The presence of alternative effective cues is the psychological starting point for the detour hypothesis; evidence for a distinct neural mechanism requires considerably more.
4. WHAT ENGRAM EXPERIMENTS DO—AND DO NOT—DEMONSTRATE
Research on experimentally identified memory engrams offers a more direct way to investigate the distinction between expression and latent information. In mice, Ryan and colleagues (2015) identified neuronal populations recruited during learning and reported that direct optogenetic activation could evoke memory-related behavior after a manipulation that prevented ordinary recall. This demonstrated, within that specific animal model and experimental procedure, that an inability to retrieve a memory under natural conditions need not imply the disappearance of all information-bearing cellular organization.
Such findings have transformed the conceptual landscape, but they are easy to overgeneralize. Experimental engram activation does not establish that a fully detailed, veridical human autobiographical episode is preserved during prolonged forgetting. The behavior produced by stimulation is a proxy for memory expression; its correspondence to a complex subjective recollection cannot be read directly from the animal’s behavior. Nor does artificial excitation establish that the circuit would eventually recover naturally or that all component representations remain intact. The finding is a proof of dissociation within defined conditions, not a general demonstration of recoverability.
A 2025 theoretical synthesis proposed that competition among engrams can favor the expression of one memory over another and may contribute to adaptive forgetting (Autore, Drew, and Ryan, 2025). This is a serious framework for further experimentation, not a settled universal explanation. Engrams are dynamic populations embedded in changing networks; they may be redistributed, transformed, and influenced by later learning. Some failures likely reflect altered activation; others reflect damage to the representation itself. The point is not to replace the concept of erasure with an equally absolute doctrine of survival.
The empirical opportunity is instead to compare multiple interventions within the same system. If a trace can be identified independently of behavioral recall, investigators can ask whether a different cue or stimulation pattern reinstates its expression. They can then determine whether reinstatement resembles the original encoding pattern, whether it predicts performance on independently tested components, and whether new learning could explain the outcome. The engram approach thereby supplies methods for testing a conditional accessibility hypothesis while retaining strict limits on what those methods can establish about lived memory.
5. NEURONAL DETOURS: FROM METAPHOR TO AN OPERATIONAL HYPOTHESIS
The source manuscript introduces Neuronal Detours to describe the recovery of access when a preferred route no longer works. Its imagery is memorable: a city may remain even when a road leading to it is closed. Yet the analogy must not be interpreted literally. Most memories do not reside at isolated physical destinations, and the brain does not ordinarily traverse a single anatomical cable to recover a particular event. Episodic retrieval is a coordinated activity of interacting hippocampal, cortical, and control systems, and its network realization can change over time.
A more precise proposal is possible. A neuronal detour is a demonstrable change in the cue-dependent neural or cognitive configuration supporting access to previously acquired information, such that successful expression occurs under a new configuration despite impaired expression under the original or ordinarily effective configuration. This definition has four requirements. There must be evidence of earlier learning; demonstrable impairment under one retrieval condition; restored performance under a distinguishable condition; and controls strong enough to distinguish recovered access from relearning, response bias, or guessing. Neural evidence of altered recruitment would strengthen the claim, but increased activation alone would not establish compensation.
The term route can then refer to a causal configuration—effective cue–representation relations, interacting neural populations, control signals, and state-dependent patterns—not a single anatomical tract. A detour may be semantic, as when a person reaches a name through occupation and contextual associations; sensory, as when a particular environmental cue enhances autobiographical recall; strategic, as when a new retrieval procedure circumvents a failing one; or network-based, when experimentally altered circuit dynamics restore expression. These cases should not be assumed to share one mechanism merely because they fit the same analogy.
This definition also makes the proposal more demanding than “memory is cue dependent.” If a novel cue simply provides extra information and permits the participant to infer the correct answer, that is not necessarily retrieval of a surviving representation. If repeated training re-encodes an association, improved performance does not uniquely reveal an old memory. If neural recruitment increases while accuracy deteriorates, additional activity may reflect effort, inefficiency, or pathology rather than compensation. A meaningful detour requires evidence that a different configuration supplies effective access to previously retained information.
The hypothesis is thus nested within existing theories of retrieval rather than presented as a discovery of an entirely new brain structure. Its contribution lies in a falsifiable causal question: under which disruptions can a network change its retrieval configuration while maintaining informative continuity with a prior representation? The answer is expected to vary by memory type, lesion location, age, severity of pathology, and the extent of surviving organization. A detour is possible only if enough relevant information remains to be mobilized; no theory of access can recover contents that were never encoded or whose supporting organization has been irreversibly destroyed.
6. THE PROBLEM OF PROVING THAT THE DESTINATION SURVIVED
A conceptual danger follows immediately from the detour metaphor: the presumed destination is normally inferred from the successful journey. This makes the claim vulnerable to circularity. If successful recall is the sole evidence that a trace existed, one cannot independently distinguish survival from reconstruction or relearning. The research challenge is to measure information retention without relying exclusively on the retrieval outcome the hypothesis attempts to explain.
One strategy is to establish a controlled encoding event with independently known target features. Later tests can combine free recall, cued recall, forced-choice recognition, source memory, and confidence calibration. The profile of performance across tests may suggest different impairment patterns, although none is by itself a perfect measure of storage. A second strategy is longitudinal measurement of representational activity, examining whether patterns related to the target remain discriminable before access is restored. Even here, decoding success does not prove conscious accessibility, and failure to decode does not prove erasure because measurements sample only part of the system.
A third strategy uses causal perturbation. If one neural component is suppressed and recall fails, but another experimentally controlled condition restores target-specific behavior, the investigator can test whether the second manipulation depends on surviving portions of the initial memory circuit. Such designs must control nonspecific arousal, motivation, stimulation-induced biases, and inadvertent retraining. Human autobiographical memories add an additional obstacle: the truth of many details is independently unknowable. Recovery cannot be validated solely by the vividness or conviction of the report.
The goal is therefore not a binary verdict about whether the “whole memory” exists. It is a graded assessment of retained information, accessibility under stated conditions, and the accuracy and generalizability of expression. This is a stronger basis for a neuroscience of continuity than the assumption of a permanently intact internal archive.
7. ALZHEIMER’S DISEASE: ACCESS FAILURE AND REPRESENTATIONAL INJURY
Alzheimer’s disease is a central test case precisely because it resists simple classification. It affects the biological conditions of encoding, consolidation, network communication, and retrieval, often together. Synaptic degeneration and neuronal pathology can compromise the ability to form new episodic memories, while additional pathology disrupts access to previously acquired knowledge. The classic finding that synaptic loss correlates strongly with cognitive impairment illustrates that disease can damage the representations and circuits on which subsequent access depends (Terry et al., 1991).
It is therefore inaccurate to portray early Alzheimer’s disease as necessarily a stage of pure retrieval failure followed only later by storage failure. In typical amnestic presentations, deficits in learning and retention may appear early; altered encoding and consolidation cannot be reduced to blocked roads leading to otherwise intact destinations. Clinical performance is heterogeneous, however, and individual tests differ in their ability to separate poor acquisition, inefficient retrieval, and recognition deficits. Some semantic, procedural, affective, and personally salient abilities may remain relatively available while episodic recollection is severely affected. Such dissociations identify uneven vulnerability, not proof of comprehensive mnemonic preservation.
Modern biological criteria also emphasize that Alzheimer’s disease is defined by pathological processes that may precede clinical symptoms, and that cognitive syndromes alone do not precisely identify the underlying disease stage (Jack et al., 2024). A retrieval-centered perspective is clinically useful only if it is integrated with neuropathological mechanisms, standardized neuropsychological assessment, and longitudinal measures rather than substituted for them.
For the detour hypothesis, the most informative finding would be target-specific restoration of previously learned material after a change of retrieval conditions, accompanied by evidence that the restored performance depends on surviving representations. Even then, the successful retrieval of selected words, facts, or experiences cannot be generalized to an entire lost life history. Treatment claims must also distinguish improving learning of new information, compensating for a deficit with external support, and restoring old autobiographical memory. All three can have substantial clinical value; only the last directly tests the recovery of a particular pre-existing trace.
8. COGNITIVE RESERVE, REORGANIZATION, AND THE MISLEADING IDEA OF REDUNDANT ROADS
The observation that people with similar burdens of brain pathology may display different levels of cognitive performance has motivated research on brain reserve, cognitive reserve, and brain maintenance. These concepts should not be collapsed. Brain maintenance concerns relatively reduced age-related change; brain reserve often refers to structural resources; cognitive reserve concerns differential susceptibility of cognition to age- or disease-related change under specified conditions (Stern et al., 2020). None is equivalent to a simple count of spare connections available for rerouting.
Likewise, altered activation in an older or neurologically affected person does not automatically indicate adaptive compensation. An increase in recruitment may assist performance, reflect greater task difficulty, or indicate less efficient processing. The literature on maintenance, reserve, and compensation calls for designs that compare neural change with behavioral benefit and underlying disease burden (Cabeza et al., 2018). It is the conditional relation between reorganization and preserved function—not the mere presence of activation—that matters.
In the context of memory, reserve provides a plausible framework for predicting who may benefit from alternate cues, retrieval strategies, or environmental scaffolding. But an alternate strategy can support task performance without retrieving the original episodic trace. For instance, a reminder system may allow a person to attend an appointment even when the memory of arranging it is unavailable. Such practical compensation should be valued in its own right, without being inaccurately described as restored storage. Distinguishing adaptive support from reinstated recollection protects both mechanistic precision and clinical expectations.
9. WHY FORGETTING CAN BE ADAPTIVE
The source manuscript correctly resists the assumption that more accessible information is always better. Cognitive performance depends on selecting among competing representations, avoiding irrelevant interference, updating obsolete associations, and preventing old responses from dominating new circumstances. Experimental work on retrieval-induced forgetting and voluntary suppression indicates that the availability of memory can be modified by the act of retrieval or intentional control (Anderson, Bjork, and Bjork, 1994; Anderson and Green, 2001). These mechanisms can be useful in some contexts, though they are not automatically beneficial or uniformly effective.
A system that preserved every association with equal priority could struggle to select the response most appropriate to present goals. On the other hand, incomplete, distorted, or inaccessible memories can impose profound harms. The value of forgetting is therefore context dependent. An altered memory may facilitate flexible decisions while also depriving a person of information important for safety or personal history. Selection and loss should not be celebrated indiscriminately simply because some forms of forgetting have adaptive functions.
The broader lesson is that stability and access are not identical optimization targets. Neural systems must sustain enough historical information for learning and coherent behavior while allowing the relevance and availability of information to change. This balance offers a productive interface between the detour hypothesis and the larger Neurotenacity programme: informational continuity does not require every representation to remain continuously accessible or influential, and excessive activation of the past is not the same as healthy persistence.
10. WHEN HISTORY FRAGMENTS: IDENTITY, RELATIONSHIP, AND RECONSTRUCTION
Autobiographical memory supports the ability to locate current experience within a longer personal history, but that history is not stored as a single, perfectly ordered narrative. Conway and Pleydell-Pearce (2000) describe autobiographical remembering as construction shaped by knowledge structures and current goals. A person may retain semantic knowledge about their life while losing vivid recollection of particular episodes, or maintain attachment and recognizable preferences despite severe episodic impairment. Thus an alteration in autobiographical accessibility is not equivalent to the disappearance of personhood.
The emotional significance of interrupted access is nevertheless substantial. Individuals may lose the capacity to revisit valued experiences, narrate personal changes, or recognize the origins of present commitments. Family members may struggle to sustain a shared history. Retrieval failure can therefore produce a form of historical disconnection that is clinically and existentially meaningful even when some information remains biologically represented. Yet preserved nonverbal responses, habits, relationships, and externally supported narratives can continue to sustain identity and social belonging.
A further complication is fidelity. Recollection is reconstructive. A cue can restore useful information while also increasing the likelihood of misattribution; a leading question may elicit a confident but inaccurate narrative. Research on the constructive properties of memory cautions against treating vivid recovered details as independently verified facts (Schacter, 2022). This consideration is especially important when evaluating therapeutic claims about recovering previously inaccessible autobiographical events. Controlled laboratory reinstatement and clinical interviewing are not interchangeable, and attempts to recover memories must avoid suggestive techniques that create or reinforce false beliefs.
From a neuroethical perspective, preservation of personal continuity is a legitimate therapeutic aim, but the dignity of a person with amnesia or neurodegeneration is not conditional on mnemonic performance. There is no evidence-based threshold at which difficulty retrieving life events causes a person to cease being a person. The proper clinical question is how to support memory, communication, autonomy, relationships, and quality of life while accurately describing the limits of restoration.
11. FIVE EMPIRICAL TESTS OF THE DETOUR HYPOTHESIS
Prediction 1 — Cue diversity should reveal conditional access rather than simple guessing. Following controlled encoding, participants should complete tests that vary cue overlap, semantic relationship, sensory modality, and contextual reinstatement. The detour hypothesis predicts that a subset of targets missed under a habitual cue configuration will be retrieved accurately under a different, pre-specified configuration, with effects exceeding those explained by response bias and additional information. Alternative explanations include unequal cue strength, inferential reconstruction, and the inadvertent provision of the answer. Pre-registration should specify which tests count as genuine target recovery.
Prediction 2 — Some unavailable representations should remain neurally discriminable. Longitudinal recordings or imaging analyses may identify activity patterns related to experimentally learned targets even when their overt recall fails. Recovery under alternate cues should covary with target-specific reinstatement, not merely global arousal or mean activation. Crucially, positive decoding is not identical to recovered subjective experience, and the absence of a measurable pattern may reflect instrumental limitations. Claims of representation persistence should be calibrated to the sensitivity and sampling of the method.
Prediction 3 — Causal intervention should distinguish restored access from new encoding. In suitable animal preparations, investigators could first establish a tagged learning-related population, then selectively disrupt an ordinary retrieval configuration and test whether a distinct circuit or cue configuration restores target-specific behavior. The strongest evidence would demonstrate that restoration depends causally on preserved components of the original trace. Parallel control conditions must evaluate stimulation-driven responding, context generalization, and repeated-exposure relearning. A failure to recover under any alternative condition, despite strong evidence of trace loss, would limit the hypothesis rather than be explained away.
Prediction 4 — Accuracy and confidence should diverge under some apparent recoveries. In human episodic tasks with independently known details, alternate cues may increase successful recall, but they may also increase recombination and false recognition. The theory predicts that true detours can be isolated only through fidelity-sensitive analyses of target features, source attribution, and calibration, not simply increased recall volume or subjective vividness. If purported detours routinely raise confidence without improving accuracy, they should not count as successful restoration of the original memory.
Prediction 5 — Clinical benefits should depend on the integrity of surviving representations. In longitudinal cohorts including healthy older adults and patients with different forms of cognitive impairment, benefits from cue diversification and compensatory strategies should be modeled jointly with initial learning, delayed retention, recognition, biomarker or imaging profiles when appropriate, and subsequent progression. If the detour account is useful, interventions aimed at access should benefit some profiles more than others, and the benefit should diminish when relevant representation integrity is severely compromised. Broad, indiscriminate claims of recoverability would be disconfirmed by persistent deficits despite adequately tested, diverse access conditions.
These predictions define a research programme rather than a validated clinical model. They also make explicit the alternatives against which it must compete: weak encoding, trace degradation, inhibitory competition, executive dysfunction, re-encoding, compensatory inference, and false reconstruction. Without these competitors, the term Neuronal Detours would be an evocative metaphor rather than a scientific hypothesis.
12. RESTORATION, NEUROMODULATION, AND THE BOUNDARIES OF THERAPY
A neuroscience of accessibility could ultimately improve interventions, but restoration must be described with care. Rehabilitation already uses external cues, error-reducing supports, spaced retrieval in selected contexts, and individualized strategies to improve everyday function. Such interventions can be meaningful whether they recover an old memory, strengthen new learning, or allow a person to function without recall. Their benefits do not require the assumption that every forgotten episode remains present beneath the surface.
Experimental brain stimulation adds a different possibility. In particular settings involving patients with implanted electrodes, closed-loop stimulation of lateral temporal regions has been associated with improved performance on laboratory memory tasks (Ezzyat et al., 2018). These results do not demonstrate routine clinical restoration of lost autobiographical memories, nor prove that stimulation discovers alternative neuronal roads to an intact personal archive. Effects are task-, state-, site-, and population-dependent. Translation to patients with degenerative memory loss requires independent clinical trials, meaningful outcomes, and rigorous assessment of risks.
Future work may combine individualized cognitive assessment, neural measurements, external assistive technologies, and carefully evaluated interventions. It should distinguish four endpoints: improved retrieval of previously learned laboratory material, improved acquisition of new information, functional compensation without recovery of the old trace, and accurate reinstatement of personally meaningful autobiographical content. Each is valuable; they are not interchangeable. Scientific ambition is best served by maintaining these distinctions, especially when people living with neurological disorders may reasonably interpret the language of “memory restoration” as a promise.
CONCLUSION — THE MEMORY THAT HIDES
The cost of forgetting cannot be measured solely by the amount of information no longer recalled. Sometimes forgetting reflects degradation of the representation; sometimes it reflects competition or the failure of cues to mobilize information that remains partly available; sometimes it produces an altered, incomplete, or misleading reconstruction. These mechanisms may coexist. The subjective experience of absence cannot, by itself, disclose which process has occurred.
The central proposal of this article is deliberately conditional. Neuronal Detours names the hypothesis that access to previously learned information can sometimes be re-established through a different configuration of cues, neural activity, or strategies when an ordinary retrieval configuration fails. The hypothesis does not entail that memories occupy isolated destinations, that every trace survives, or that reconstructed recollections are literal reproductions of the past. It is scientifically useful only insofar as it predicts independently verifiable recovery, specifies competing explanations, and identifies cases in which detours should fail.
If the larger concern of The Architecture of Mind is how a changing nervous system preserves the consequences of experience, forgetting reveals a necessary complement: the past can shape a system even when it is not available to report, and the recovery of the past is an active process constrained by biology, cues, goals, and present circumstances. Yet history is not guaranteed to remain intact behind every failed recollection. What matters is not the comforting certainty that every lost memory still exists, but the disciplined possibility that some failures of remembering are failures of access—and that the distinction can be tested.
REFERENCES
Anderson, M. C., Bjork, R. A., & Bjork, E. L. (1994). Remembering can cause forgetting: Retrieval dynamics in long-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(5), 1063–1087. https://doi.org/10.1037/0278-7393.20.5.1063
Anderson, M. C., & Green, C. (2001). Suppressing unwanted memories by executive control. Nature, 410, 366–369. https://doi.org/10.1038/35066572
Autore, L., Drew, M. R., & Ryan, T. J. (2025). The cost of remembering: Engram competition as a flexible mechanism of forgetting. Trends in Neurosciences, 48(10), 728–738. https://doi.org/10.1016/j.tins.2025.07.011
Burke, D. M., MacKay, D. G., Worthley, J. S., & Wade, E. (1991). On the tip of the tongue: What causes word finding failures in young and older adults? Journal of Memory and Language, 30(5), 542–579. https://doi.org/10.1016/0749-596X(91)90026-G
Cabeza, R., Albert, M., Belleville, S., et al. (2018). Maintenance, reserve and compensation: The cognitive neuroscience of healthy ageing. Nature Reviews Neuroscience, 19, 701–710. https://doi.org/10.1038/s41583-018-0068-2
Conway, M. A., & Pleydell-Pearce, C. W. (2000). The construction of autobiographical memories in the self-memory system. Psychological Review, 107(2), 261–288. https://doi.org/10.1037/0033-295X.107.2.261
Ezzyat, Y., Wanda, P. A., Levy, D. F., et al. (2018). Closed-loop stimulation of temporal cortex rescues functional networks and improves memory. Nature Communications, 9, 365. https://doi.org/10.1038/s41467-017-02753-0
Jack, C. R., Jr., Andrews, J. S., Beach, T. G., et al. (2024). Revised criteria for diagnosis and staging of Alzheimer’s disease: Alzheimer’s Association Workgroup. Alzheimer’s & Dementia, 20, 5143–5169. https://doi.org/10.1002/alz.13859
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
Schacter, D. L. (2022). The seven sins of memory: An update. Memory, 30(1), 37–42. https://doi.org/10.1080/09658211.2021.1873391
Stern, Y., Arenaza-Urquijo, E. M., Bartrés-Faz, D., et al. (2020). Whitepaper: Defining and investigating cognitive reserve, brain reserve, and brain maintenance. Alzheimer’s & Dementia, 16, 1305–1311. https://doi.org/10.1016/j.jalz.2018.07.219
Terry, R. D., Masliah, E., Salmon, D. P., et al. (1991). Physical basis of cognitive alterations in Alzheimer’s disease: Synapse loss is the major correlate of cognitive impairment. Annals of Neurology, 30, 572–580. https://doi.org/10.1002/ana.410300410
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



