47 The Diary on the Map
How Particular Events Keep a Predictive Map Current
An animal exploring while no need is urgent appears, from a narrow view, to be wasting time. It leaves a safe location, spends energy, and exposes itself to danger without immediately eating, drinking, mating, or escaping. Tolman’s rats revealed a problem with that narrow accounting. Animals allowed to traverse a maze before food was introduced later improved abruptly when food gave one destination value. Learning had occurred before the rewarded performance revealed it [@tolmanhonzik1930reward; @tolman1948cognitive]. This does not tell us that the rats had no motivation during exploration, or uniquely establish the representation they acquired. It separates learning about an environment from the incentive to express that learning in a particular way.
That distinction between learning the world and valuing one possibility now is the beginning of this chapter. Exploration can reveal routes, barriers, refuges, resources, risks, and uncertain transitions before a particular deficit makes any one of them urgent. An animal may encounter water while hydrated, shelter while safe, or an escape route before a predator appears. Information acquired under one bodily condition can become useful under another.
This is why exploration can be an allostatic investment. Regulation cannot always wait for a deviation to become severe and then begin searching blindly for the means to correct it. Anticipatory control reduces that delay by preparing for future demands [@sterling2012allostasis]. Exploration purchases options in advance, although its costs and benefits depend on the animal’s condition and circumstances.
But a general map is not enough. Worlds change. The spring discovered last month may now be dry. The direct path may be blocked. A cache may have spoiled. A familiar shelter may have acquired a predator. The controller therefore needs more than established relations among possible states and transitions. It needs evidence from particular encounters that can confirm, qualify, or contradict those relations.
A predictive map gives allostasis a geography. The diary keeps that geography current.
The chapter’s question is how particular encounters keep general knowledge useful. We begin with exploration and bodily need, then turn to Clive Wearing, H.M., and the interpretation of human amnesia. Their cases sharpen a distinction already demanded by animal behavior: knowing what generally holds is different from remembering what happened on one occasion. That distinction also changes what can legitimately be concluded when old knowledge survives brain damage.
47.1 What exploration purchases
Tolman’s cognitive map challenged the idea that learning consists only of reinforced stimulus-response chains. A rat can acquire information about paths and destinations before an experimental reward makes one destination worth pursuing. O’Keefe and Nadel later placed the hippocampus at the center of an account in which relations among places support action beyond the currently sensed cue [@tolman1948cognitive; @okeefenadel1978map]. The preceding chapters developed the comparative and physiological basis of that argument. Here the concern is what makes those relations useful when the body’s demands change.
Exploration reveals what an environment permits. An opening may afford passage, a hollow protection from weather, and a spring drinking. James Gibson called such possibilities affordances [@gibson1979ecological]. An affordance is a relation between an organism and a situation. A gap permits passage only for a body that fits; a branch permits perching only for an animal able to grasp it. Water may remain drinkable when the animal is hydrated, while thirst changes the value of approaching it.
Several contributions must therefore remain distinguishable:
| Contribution | Question answered |
|---|---|
| Map structure | What states, contexts, and transitions are available or expected? |
| Affordance | What can this body do in this situation? |
| Event-specific evidence—the diary | What happened when a possibility was tested on one occasion? |
| Current value | How much does the possible outcome matter under present and anticipated conditions? |
| Policy | How are actions selected in the states the animal encounters? |
These are functional distinctions within an interacting system. They are not five isolated anatomical compartments. Need can change attention, sampling, and the relations that are represented, as well as the value assigned to an outcome. Conversely, a learned relation can alter anticipated need: knowing that water will be unavailable later may change drinking now. As Section 45.7 explains, even a prediction of future state occupancy can depend on the policy under which it was learned. A useful map need not be a complete, goal-free description of the world.
Nor does talk of affordances require a neural token labeled affordance. The learned relation must preserve enough information to recover a possibility: where it was encountered, which approach made it available, what action the body could perform, and what followed. Its use depends on systems that register bodily condition, evaluate outcomes, and organize action.
47.2 A map is not enough: it needs a diary
Suppose an animal has learned that a spring lies beyond a ridge. The route may remain useful across many visits, but the proposition water is there leaves out conditions that matter. The last successful visit may have followed heavy rain. A more recent visit may have revealed a collapsed path or a predator. The relevant evidence concerns a conjunction: in this context, after this approach, an action produced this outcome.
That encounter must remain distinguishable from others sharing the same place, objects, and actions. If every visit were immediately averaged into a single expectation, an important exception could disappear before the animal had a chance to use it. An event record allows the controller to ask whether the circumstances of an earlier success still apply.
The map–diary distinction is deliberately functional. A map can represent change, uncertainty, and conditional transitions; it need not predict a stationary world. The diary names the contribution of separately recoverable encounters to learning and revising that structure. It need not supply a calendar date, preserve every sensory detail, or operate like a private video archive.
Scrub-jay cache recovery makes the biological problem concrete. In Clayton and Dickinson’s experiment, birds searched for cached wax worms or peanuts according to the food, its location, the retention interval, and learned consequences of perishability. The birds’ experience of whether worms degraded mattered; elapsed time alone was not a universal instruction to avoid them [@clayton1998episodic]. The result supports integrated what–where–when performance. It does not demonstrate human autobiographical consciousness or directly identify the neural mechanism.
Ergorul and Eichenbaum asked rats to judge the order of odor-bearing events encountered at different locations. Their probes examined the contributions of odor and spatial information. Hippocampal lesions disrupted the combined use of those cues in judgments about event order, despite evidence that the animals could perceive and remember individual cue information [@ergoruleichenbaum2004whatwherewhen]. Here “when” concerned order within a sequence, not the elapsed-time judgment tested in the jays. Both experiments demand distinctions among occurrences; they do not establish one identical representation of time or the subjective experience of recollection.
The neural mechanisms developed in Drawing the Map give this problem a possible implementation. Population distinctions and remapping can help separate overlapping contexts; they do not imply that every new event requires a new global spatial map. Ordered activity can preserve relations among parts of an encounter. The theta organization in Section 45.8 and the ripple-associated activity in Section 45.10 provide opportunities for relating and reactivating such information. Observing an ordered sequence, however, does not by itself establish that the animal is recollecting, planning, or consolidating. Those claims require additional evidence about use and causal contribution.
A recording study in food-caching chickadees brings the distinction between location and occasion closer to the population level. Chettih and colleagues found sparse, event-specific hippocampal activity during caching, alongside more conventional place-related activity. Patterns associated with particular caches reappeared during retrieval of those caches [@chettih2024barcoding]. Nearby caches could have similar spatial codes while retaining distinct event patterns. This is a candidate means of distinguishing encounters within an organized environment. The recordings do not establish that a pattern contains the whole memory, that its reactivation causes retrieval, or that it selects a distant cache before the bird approaches it.
The general relation makes a resource predictable. The particular encounter preserves evidence about whether it was available under relevant conditions. Their interaction allows an old map to remain useful without requiring every exception to become a new permanent rule.
47.3 The same world under hunger and thirst
Pamela Kennedy and Matthew Shapiro devised an experiment that joins internal bodily state, memory retrieval, and goal selection particularly clearly. Rats learned to approach one of three distinctive goal boxes in the same testing environment. One predicted food, one water, and one no reward. The boxes changed locations. Food deprivation made the food-associated box correct; water deprivation made the water-associated box correct [@kennedyshapiro2004internal]. A fixed destination or an unconditional preference for one object could not solve both conditions.
The critical demand was not simply to be hungry or thirsty. The animal had to use its state to recover which learned alternative predicted the appropriate outcome. After training, separate groups received neurotoxic hippocampal lesions or fornix transections. Both interventions severely impaired that conditional choice. The rats nevertheless continued to avoid the unrewarded box. Control tests supported preserved cue discrimination, discrimination of deprivation state, and deprivation-appropriate reward preference [@kennedyshapiro2004internal]. Thus the deficit cannot be explained simply by loss of motivation or inability to distinguish the objects.
The controls constrain the explanation; they do not prove that every sensory or motivational operation remained normal. Nor are a hippocampal lesion and a fornix transection anatomically interchangeable. One damages tissue; the other interrupts a major fiber pathway. Their convergence supports a role for the hippocampal system in this learned, context-dependent choice. A lasting postoperative deficit does not isolate a millisecond of retrieval from every contribution of retention, interference control, or relearning.
The following distinctions keep the biological interpretation precise:
| Capacity | What would demonstrate it here? |
|---|---|
| Bodily motivation | Deprivation changes the animal’s preference for food or water. |
| Simple discrimination | The animal distinguishes the visible goal objects or avoids the consistently unrewarded one. |
| State-dependent memory use | Hunger and thirst select different learned object–outcome relations in the same external setting. |
| Memory-guided choice | The retrieved relation guides approach when box location changes. |
This is a conditional retrieval task, not merely the familiar state-dependent-memory effect in which matching the state at learning and testing improves recall. It also does not show that a rat learned all the associations while sated and first used them when deprived. Both motivational conditions were part of training. Its relevance to exploration is architectural: bodily need can govern the use of learned alternatives. The broader claim that exploration prepares future options draws on the argument developed above, rather than on a manipulation this experiment did not perform.
In a later recording study, Kennedy and Shapiro compared hippocampal activity during contextual retrieval with activity during random foraging. Hunger and thirst had much stronger effects when motivational state determined the remembered goal. On the common approach, prospective firing distinguished the deprivation-defined, nonspatial target rather than simply the subsequent allocentric trajectory [@kennedyshapiro2009motivation]. These are recording results from a separate experiment. They show task-dependent organization of activity; they do not supply a second lesion result or prove that the recorded cells caused the choice.
The control interpretation is nevertheless substantial. The body supplies a question that the current visual scene cannot answer alone. Learned relations help recover an answer. Valuation and action systems turn that answer into behavior, and the next encounter can revise it. The hippocampus participates in that interaction without being the source of hunger, the complete representation of value, or the final policy.
47.4 Human memory arrives last
We have arrived, at last, at ourselves.
For most of this unit, the explanation began with the moving animal and the relations needed to guide its body beyond the present cue. The comparative evidence supports an old medial-pallial contribution to that problem. It does not establish that every temporal or nonspatial capacity evolved after spatial navigation. Living fish have their own evolutionary histories; they are evidence about continuity and diversity, not unchanged ancestors.
Human episodic memory elaborates event organization through language, self-knowledge, social roles, causal explanation, calendars, and narrative. A scrub jay recovering a cache and a person recalling a childhood meal face some related binding problems. Their capacities do not differ only in the length of the interval remembered. Human recollection can locate an event within a life story and communicate it to someone else.
Chapter 46 showed why shared relational demands do not guarantee identical representations. Its evidential criteria in Section 46.2 apply here too: performance, recorded activity, and intervention answer different questions. Human neuropsychology becomes most informative after the biological problem has been established. It reveals what can disappear when an event-organizing system is damaged, while leaving the evolutionary origin of that system to comparative explanation.
47.5 The man in the present tense
In 1985, the musician and musicologist Clive Wearing developed herpes simplex encephalitis. His injuries extended well beyond the hippocampus. The anatomical assessment reported bilateral hippocampal and amygdalar abnormalities, temporal cortical injury more extensive on the left, and involvement of structures including the mammillary bodies and basal forebrain. There was also milder abnormality in other regions, including left medial frontal cortex. Formal testing found exceptionally severe episodic amnesia together with substantial semantic impairment [@wilsonetal1995wearing]. This was not a selective hippocampal manipulation.
The detail that makes the case unforgettable is his diary. Wearing repeatedly recorded that he had only just awakened, rejecting earlier entries as accounts of a state that was not truly conscious. The investigators treated this persistent conviction as a delusion accompanying his amnesia, rather than an inevitable consequence of losing episodic memory. The diary is evidence of his behavior and belief; it is not a transparent report of what isolated hippocampal damage feels like.
His preserved musical abilities provide a complementary observation. Familiar performance could unfold through deeply practiced organization and the support of an ongoing situation, despite his inability to retain ordinary daily encounters. Calling this simply “procedural memory spared” identifies a dissociation but explains little about the interacting perceptual, motor, affective, and learned musical relations that sustain performance.
Wearing’s case makes the cost of event loss concrete. The difficulty is not only recalling yesterday. Without reliable access to what has just occurred, the present repeatedly loses information that would place it within a continuing personal history. A visitor can be familiar while the visit is forgotten. An action can be sustained while its occurrence fails to become available later.
This is the chapter’s interpretation of the functional problem, not a diagnosis of the mechanism behind every entry in Wearing’s diary. His severe amnesia, semantic losses, and abnormal beliefs arose in a broadly injured brain. Keeping those features distinct makes the case more instructive: a memorable narrative cannot replace an anatomical and behavioral account.
47.6 The cabinet and the cost of mistaking labels for mechanisms
Memory taxonomy is useful because performances dissociate. Long-term memory is conventionally divided into declarative and nondeclarative forms. Declarative tasks concern facts and events that can support flexible recognition, recollection, and report. Verbal declaration is an expression of memory, not a prerequisite for demonstrating it in an animal or a nonverbal person.
Within this domain, Tulving distinguished episodic memory for particular events from semantic knowledge that does not require re-experiencing one occasion [@tulving1972episodic]. Other forms of learning appear in skills, habits, priming, and conditioning. They do not share a single mechanism merely because they fall on the nondeclarative side of a classification.
A patient may improve on a task while forgetting the training sessions, or retain personal facts while failing to reconstruct a particular event. Such contrasts establish that capacities can come apart. They do not establish a separate anatomical cabinet for every psychological label. The operation must still be explained: what information is combined, how it remains accessible, and what a damaged system contributed to its use.
Lesion evidence is especially powerful because damage changes performance. Its anatomical scope also matters. A broad injury cannot be made selective by describing it with a narrow psychological term. H.M. illustrates both the explanatory force of a lesion case and the danger of allowing its familiar interpretation to become a substitute for the underlying evidence.
47.7 H.M. and the remote memory that became a fact
In 1953, William Scoville performed a bilateral medial-temporal operation on Henry Molaison to treat severe epilepsy. The operation damaged medial temporal polar tissue, much of the amygdala and entorhinal cortex, and the anterior hippocampal formation. Substantial posterior hippocampal tissue and posterior parahippocampal cortex remained. Entorhinal damage severely compromised normal hippocampal communication, but spared tissue should not simply be declared functionless. Later examination also identified abnormalities beyond the intended resection [@scovillemilner1957; @annese2014hm]. H.M. was neither a hippocampus-only lesion nor a brain from which the entire hippocampus had been removed.
His profound anterograde amnesia established an essential point. Ordinary encounters no longer reliably produced durable new declarative memories, although language, much established knowledge, and immediate retention under sustained attention remained available. He could also improve on some practiced tasks without recollecting the training episodes [@milneretal1968hm]. The dissociation between skill acquisition and memory for acquiring the skill was subsequently demonstrated in other amnesic patients as well [@cohensquire1980skill]. Medial-temporal damage does not abolish every form of learning.
The more contentious conclusion concerned H.M.’s past. The original report described early memories as apparently normal, with retrograde loss concentrated nearer the operation [@scovillemilner1957]. Such observations helped support a temporary medial-temporal role in memory. They did not establish that a richly detailed remote autobiography remained intact. A childhood address, a family fact, and a familiar story are all information about the past. None, by itself, demonstrates reconstruction of one particular lived encounter.
Later testing made this distinction explicit. Steinvorth, Levine, and Corkin separated personal semantic knowledge and knowledge of public events from the internal details of autobiographical episodes. H.M. showed much better preservation of remote knowledge than of specific autobiographical recollection, without a simple gradient that spared his earliest episodes. Yet he was not devoid of remote episodic detail: he described an airplane flight at age fifteen with substantial specificity and another event, an uncle’s wedding, that he could not assign to the requested period [@steinvorthetal2005remote]. Severe impairment and complete absence are different claims.
This changes the evidential burden. H.M. cannot safely be used as the uncomplicated demonstration that old episodes survive intact after the hippocampal system becomes unnecessary. The postmortem paper likewise distinguishes retained premorbid semantic knowledge from his profound autobiographical difficulty [@annese2014hm]. A surviving biography is not equivalent to a surviving diary.
There is also a limit to the retrospective correction. Testing conducted decades after surgery cannot, on its own, establish the quality of every remote memory immediately afterward. Aging, additional pathology, residual tissue, and differences in testing all matter. Kirwan and colleagues later reported detailed remote recollection in other patients with hippocampal or larger medial-temporal lesions, using the same general interview approach. They also emphasized later changes in H.M.’s brain when interpreting his poor performance [@kirwan2008remote]. These findings prevent a single late assessment from settling the general necessity of the hippocampus for all remote episodes.
They do not rescue the stronger inference commonly attached to H.M. The absence of an adequate early distinction between an episode and personal knowledge cannot be repaired by assuming that whatever survived was episodic. Nor does identifying a possible contributor to his later deficit demonstrate that his earlier autobiography had been normal.
H.M.’s retained past was evidence to investigate, not proof that detailed episodes had become hippocampus-independent.
The historical criticism should be attached to that inference. The original clinical reports, later measurements, and theoretical uses of the case are not interchangeable. Consolidation theory also drew on other patients and animal studies; it was never logically dependent on H.M. alone. Nevertheless, allowing his broad remote-memory sparing to stand for intact remote episodic recollection gave a disputed conclusion the authority of a celebrated fact. That is a consequential error, not a minor omission of detail.
47.8 An elegant model and the inference it cannot establish
Complementary Learning Systems gave a temporary hippocampal contribution a computational rationale. McClelland, McNaughton, and O’Reilly proposed a fast hippocampal learner and a slower neocortical learner. Rapid learning protects the conjunctions of particular experiences; gradual, interleaved learning allows shared structure to accumulate without each new experience overwriting older knowledge [@mcclellandetal1995cls].
The 1995 paper explicitly treated temporally graded retrograde amnesia as central evidence. Its argument also extended cortical learning to episodic tasks: if reinstatement repeatedly included an event’s original context, that information could eventually support performance through neocortical changes. The claim was conditional on what was reinstated. It was not that every event automatically survives unchanged, or that cortex is initially inactive.
That distinction makes the criticism more exact. The original model offered a mechanism by which performance based on event information could become cortically supported. It did not demonstrate that the resulting representation retains the richness, organization, and experiential quality of a particular recollection. Its learning problem and its anatomical interpretation must be evaluated separately.
Connectionist networks can store many relations in shared weights. If those weights change rapidly to master a new set of associations, previously learned performance can deteriorate severely. McCloskey and Cohen demonstrated this problem of catastrophic interference [@mccloskeycohen1989catastrophic]. Interleaving old and new examples can reduce it, allowing learning to accommodate both.
CLS used this computational problem to motivate complementary learning rates and representations. The fast system preserves information that can train the slower system over repeated reinstatements. This is a productive account of how rapid acquisition and gradual integration might coexist.
But successful behavior in a model does not establish why a biological system evolved, which information natural reinstatement preserves, or whether a human episode remains available after a particular circuit is no longer involved. A computational solution supplies a candidate mechanism. It does not supply those missing observations.
Three claims should therefore remain separate:
- Rapid learning of particulars and gradual integration of regularities can be computationally complementary.
- Interactions among hippocampal and cortical systems change with learning and time.
- A richly detailed episode can eventually be reconstructed without hippocampal participation.
The first is a useful principle and the second has substantial empirical support. The third requires evidence about the content of remote retrieval and the causal role of the circuit. Neither the principle nor the existence of reorganization makes it follow. H.M.’s retained remote knowledge did not supply the decisive evidence that the stronger interpretation required.
In this respect, the familiar transfer story sent explanation in the wrong direction. Asking where an unchanged memory goes encourages students to treat persistence of a fact, a successful conditioned response, and reconstruction of a lived scene as the same achievement. Asking what must be reconstructed exposes the distinction. Replay can support learning without proving transfer of an intact episode; cortical participation can increase without establishing that hippocampal operations have become dispensable for every form of retrieval.
This criticism is directed at an inference, not at every model called CLS. Later formulations explicitly discuss rapid learning when new information fits existing knowledge and a broader role for hippocampal computation in inference and behavior [@kumaran2016cls]. Those developments should be judged on their own claims. They do not retroactively turn early reports of H.M.’s spared past into a controlled demonstration of preserved remote episodic recollection.
The value of the fast–slow distinction survives this criticism. What must be rejected is its promotion into a necessary biological history in which detailed memories finish their lives as hippocampus-independent copies. That is an additional claim, requiring an additional test.
47.9 An alternative centered on the retrieved representation
Nadel and Moscovitch’s Multiple Trace Theory placed the critical distinction in the kind of memory being retrieved. Rich episodic recollection continued to depend on a distributed trace involving the hippocampal complex; repeated reactivation could create additional traces, making some memories more resistant to partial damage. Semantic knowledge could become less dependent on that system [@nadelmoscovitch1997consolidation]. The account did not merely lengthen a single consolidation interval. It challenged the proposed endpoint for detailed episodes.
The later transformation account emphasized the coexistence of different representations. A context-rich episode and a more schematic version can emerge from the same experience. Gist can support some remote performance with less hippocampal involvement, while retrieval of the detailed event continues to recruit hippocampal operations [@winocurmoscovitch2011transformation]. Transformation need not mean that a generalized version inevitably replaces and destroys the detailed one.
These positions belong beside, but should not be collapsed into, the broader claim of systems consolidation:
| Account | Main commitment relevant here | Crucial question for remote memory |
|---|---|---|
| Systems consolidation with a time-limited hippocampal role | Reorganization can permit cortical support for memories that initially depended on hippocampal interaction. | Does genuinely detailed event retrieval survive sufficiently characterized hippocampal damage? |
| Original CLS | Fast acquisition and slower interleaved learning solve different computational problems; repeated reinstatement can build cortical support. | What event information is actually reinstated and retained, and what does independent cortical performance preserve? |
| Multiple Trace Theory | Episodic traces continue to include the hippocampal complex; reactivation can add traces. | Do lesion extent and the richness of recollection explain vulnerability better than age alone? |
| Transformation account | Detailed and schematic representations may coexist with different circuit dependencies. | Is apparently spared remote performance based on the same information as recent performance? |
These are related families of proposals, not four mutually exclusive boxes. CLS is a computational framework for a division of learning labor; systems consolidation is a broader account of changing circuit dependence. Multiple-trace and transformation accounts also allow substantial reorganization. Their dispute concerns what changes, what remains available, and which operations still require the hippocampal system.
The remote-memory result in H.M. fits the episode–knowledge distinction emphasized by multiple-trace accounts better than the simple assertion that his old episodes were intact. It does not uniquely prove that theory: anatomical breadth, residual episodic performance, and other patients remain relevant. The stronger position taken here is that time alone is an inadequate explanatory variable. Tests of remote memory must specify what the person actually retrieves. That requirement should alter the main teaching diagram, rather than appear as a qualification beneath an unchanged transfer cartoon.
47.10 A map without its particulars
The patient K.C. makes the distinction visible in another form. After a motorcycle accident, he had widespread brain injury including large bilateral hippocampal lesions. Severe anterograde amnesia accompanied a profound loss of autobiographical recollection across his earlier life, while substantial factual and personal semantic knowledge remained [@rosenbaumetal2005kc].
His remote spatial knowledge was especially revealing. Tests showed considerable preservation of major landmarks, routes, distances, and directions in a familiar environment, alongside difficulty with less salient landmarks and location details [@rosenbaumetal2000spatial]. A broad relational structure could remain available when the individual encounters associated with it could not be recalled. That does not imply that every route-finding demand was intact or that the residual representation contained all the detail it once had.
K.C.’s injury was broad, so it cannot assign every loss to the hippocampus alone. The limitation is anatomical, not an argument that lesions provide no causal evidence. His case constrains an account of memory because some capacities survived damage that devastated others.
Knowing the layout of a neighborhood is not the same as remembering one walk through it. Knowing that a family trip occurred is not the same as recovering one scene from the trip. A map and the events through which it was learned can have different fates. This is why the content of the test matters as much as the date of the experience.
47.11 What memories become with time
The map and diary are not independent stores. Each encounter supplies evidence from which stable regularities can be learned: this route usually remains open, this kind of place contains food, this person usually responds in a particular way. General knowledge makes control efficient because the organism need not reconstruct its full history whenever it acts.
Efficiency creates a risk. Averages can conceal change. A spring that usually contains water may have been dry on the most recent visit. Preserving that encounter separately allows it to govern behavior before the long-run expectation changes. Conversely, an established schema helps interpret a new event and determines which deviations are worth retaining. Regularities and particulars constrain one another.
Several processes gathered under consolidation must therefore be distinguished:
| Process | What changes? | What does not follow automatically? |
|---|---|---|
| Synaptic stabilization | Cellular and molecular changes make newly altered circuitry less fragile. | The memory has become hippocampus-independent. |
| Systems reorganization | Contributions and interactions among neural populations change. | An unchanged episode has moved between two locations. |
| Representational transformation | Retrieval relies on different content, detail, or structure, often emphasizing gist or schema. | Every detailed trace has been destroyed or can no longer be recovered. |
The ripple-associated activity discussed in Section 45.10 supplies a possible setting for some of this interaction. Reactivated relations may influence plasticity, comparison, or later accessibility. Interruption experiments establish that disrupting particular activity periods can affect subsequent memory performance; they do not by themselves identify which content was being taught to which population. A physiological sequence is not a complete explanation of consolidation, any more than a decoded path is a complete explanation of planning.
Kitamura and colleagues provided a more direct intervention account in mice trained with contextual fear conditioning. They labeled cells active during learning and manipulated selected ensembles and pathways. Prefrontal cells associated with the experience were established during initial training, before natural context cues reliably recruited them for remote memory expression. Their later maturation depended on hippocampal ensemble output. Inhibiting selected prefrontal ensembles or their output impaired later fear expression under the tested conditions [@kitamura2017engrams]. Cortex was involved from the beginning; its contribution changed.
The same study also distinguished natural recruitment from artificial accessibility. Dentate-gyrus cells tagged during conditioning were preferentially reactivated by the conditioning context at an early test but not at a later one. Yet direct optogenetic activation of the tagged population could still induce freezing at a later test. Loss of preferential recruitment by the natural cue therefore did not establish complete destruction of the relevant trace. Conversely, stimulation-evoked freezing did not show that ordinary context-driven retrieval still used that population in the same way.
These experiments are often described in terms of engrams, the enduring neural changes supporting a memory. Experimentally, however, activity-dependent tagging selects cells meeting a labeling criterion during a particular interval. The selected ensemble is an entry point for testing memory-related circuitry, not a complete inventory of everything constituting the experience.
The distinction among tests is essential. Recorded reactivation shows a relation between prior experience and later activity. Inhibition can establish a contribution to measured behavior under specified conditions. Artificial activation can show that stimulating a population is sufficient to recruit an output through the remaining circuit. None alone identifies the entire stored content or establishes subjective recollection. Freezing is a measurable defensive response; it is not a report of the details of an autobiographical scene.
Kitamura’s results support changing circuit dependence in this fear task and place useful constraints on models of reorganization. They should not be recruited as either proof or disproof of lifelong hippocampal involvement in richly detailed human recollection. The broader lesson is that persistence, accessibility, and ordinary use are distinct properties. A memory-related population can remain artificially accessible after its natural recruitment changes, and a behavioral response can persist while the contribution of a particular pathway declines.
47.12 Remembering one event and constructing another
Remembering an event and constructing a possible event both require information that is not supplied by the present scene. In the fMRI experiment of Addis and colleagues, participants first constructed a past or future event from a cue and then elaborated its details. The tasks recruited overlapping regions, particularly during elaboration, while also showing differences between phases and between past and future conditions [@addis2007remembering]. The common network included hippocampal and cortical regions involved in autobiographical retrieval.
Shared recruitment is consistent with shared demands such as retrieving elements, organizing context, and elaborating a situation. It does not prove that remembering and imagining perform one identical computation. Regional BOLD overlap cannot establish that the same neurons perform the same operations, and a plausible future scene faces constraints different from those on a recollection of an actual event.
Lesion evidence adds another kind of constraint. Hassabis and colleagues asked five amnesic patients with bilateral hippocampal damage to imagine novel commonplace experiences. Four produced markedly impoverished, spatially fragmented constructions; one performed relatively well [@hassabis2007imagine]. This implicates the damaged system in the task without establishing that every form of imagination requires it. Verbal planning, predictions based on familiar facts, and use of an established script need not demand the construction of a richly situated scene.
Developmental amnesia complicates the inference further. In the study by Maguire and colleagues, the patient Jon could construct detailed fictitious and future experiences despite severe episodic impairment [@maguire2010developmental]. Residual tissue, developmental adaptation, semantic support, and strategy are possible contributors. That experiment did not isolate which explanation accounted for his performance. A discrepancy between early-acquired and adult-acquired injury is a question about developmental history, not proof of a single compensatory mechanism.
The organizing interpretation remains useful: hippocampal–cortical interaction can contribute to assembling particular situations from elements that must be related rather than merely listed. The contribution depends on the demands of the task and the resources available to the individual. It is not equivalent to the claim that all future-directed thought is episodic or hippocampal.
For a controller, constructing a possible encounter can reveal consequences before an expensive action is taken. An animal or person might consider approaching water by an exposed route, arriving after dark, or finding the resource absent. This is an application of event construction to anticipatory control, not a demonstration that the cited imaging task measured allostatic regulation.
Construction also creates opportunities for error. Knowledge can fill missing details, current goals can bias selection, and elements from different occasions can be combined. The constructive episodic simulation account relates that flexibility to both future thought and memory distortion [@schacteraddis2007constructive]. It does not follow that every act of retrieval changes a memory or that accuracy is biologically unimportant. Confusing a dry spring with a full one is costly precisely because the particular encounter matters.
Distortion is a cost of flexible reconstruction, not evidence that fidelity is biologically irrelevant.
47.13 Development without an ordinary diary
Developmental amnesia offers a longer view of the relation between particulars and regularities. Vargha-Khadem and colleagues described three young people with early-onset bilateral hippocampal pathology. Despite severe difficulties remembering everyday episodes, they attended mainstream schools and acquired language, literacy, and substantial factual knowledge within the low-average to average range [@varghakhademetal1997developmental].
The cases demonstrate that substantial semantic knowledge can develop in the presence of profoundly impaired episodic memory. They do not demonstrate complete absence of hippocampal function. Tissue remained, the injuries were not controlled surgical isolations, and the nervous system developed under altered conditions. Repetition, recognition, educational support, and surviving circuitry could contribute to learning. Their relative contributions cannot be read directly from the eventual level of knowledge.
This distinction between developmental timing and causal dependency matters. An ability appearing despite early injury does not prove that the affected system makes no contribution in ordinary development. Nor does the earlier appearance of one ability make it the cause of a later one. Compensation and alternative routes of learning are biological possibilities that require their own evidence.
The functional difficulty is easy to underestimate when school achievement is emphasized. A person can know a bus route, multiplication facts, and familiar people’s names while failing to retain an encounter from earlier that day. Routines, notebooks, schedules, and help from other people can supply information that an unreliable internal event record fails to provide. General knowledge supports much of life, but a recent exception may still determine whether a familiar action is safe or appropriate.
The developmental cases therefore strengthen the chapter’s distinction without converting it into two independent stores. Repeated structure can support knowledge acquisition under conditions in which ordinary event memory is gravely impaired. The ability to learn what generally holds does not remove the need to retain what happened this time.
47.14 The diary inside the controller
The hippocampus is not the diary by itself, just as it is not the cognitive map by itself. Event use depends on interaction among systems that register bodily condition, preserve relations, evaluate outcomes, and implement action. Their contributions can be described without giving each psychological term an exclusive anatomical home:
| Interacting contribution | Role in the use of memory |
|---|---|
| Brainstem, hypothalamic, and interoceptive systems | Register bodily condition and contribute to regulatory priorities. |
| Hippocampal–entorhinal and distributed cortical systems | Learn and recover relations among contexts, entities, transitions, and particular encounters. |
| Amygdalar, orbitofrontal, and other valuation systems | Contribute expected outcome identity, affective significance, and current consequences. |
| Frontal and striatal systems | Contribute goals, rules, action values, selection, and control of ongoing behavior. |
| Motor and autonomic systems | Implement responses that alter the animal’s relation to its surroundings and its body. |
| Returning sensory and interoceptive signals | Supply evidence about what the response achieved. |
These are overlapping contributions, not a wiring diagram or a list of direct projections. Thirst does not specify a route. A remembered route does not determine how much water matters now. Value does not reconstruct the last encounter. A policy selects actions, but those actions depend on learned structure and generate the evidence through which it changes.
The Kennedy–Shapiro dissociation makes one part of this interaction unusually clear. Appropriate motivation and simple discrimination can remain while conditional memory-guided choice fails. The task does not require a diary of a unique training trial; repeated learning can establish the object–outcome relations. It therefore supplies evidence about need-dependent use of memory, not a direct test of episodic recollection. Event-specific evidence becomes essential when the established relation needs qualification: the spring was dry, the familiar route failed, or a normally safe encounter ended badly.
The map may still contain the destination and its routes. The latest event changes the predicted outcome or confidence in the relation; current need changes the importance of that prediction; the policy determines what the organism does next. Keeping those contributions distinct makes their interaction intelligible.
47.15 Coda: the past changes what can happen next
Return to the animal that found water while hydrated. Exploration established a possibility before thirst made it urgent. Later dehydration changed the value of approaching the spring. But a useful controller must also recover evidence that qualifies the general relation: water was present after rain, the direct path was blocked on a recent visit, or a predator was encountered at dusk.
The map preserves possibilities. The diary preserves evidence about when those possibilities can be trusted.
The human cases now have a biological context. Wearing shows the devastating loss of ordinary event continuity in a brain with extensive injury. H.M. shows that medial-temporal damage can profoundly impair incorporation of new experience while leaving much established knowledge, immediate retention, and some forms of learning available. K.C. demonstrates how broad spatial and semantic knowledge can survive alongside profound autobiographical loss.
H.M. also leaves a methodological warning. Preserved information about the remote past must not silently become preserved detailed episodes, and successful performance must not silently become proof of hippocampal independence. The original CLS framework gave a powerful account of complementary learning demands. That achievement does not settle the fate of rich recollection. The critical question is what representation survives and what operation its retrieval still requires.
Human beings add another consequence to event memory. Language, testimony, instruction, written records, photographs, and public histories allow one person’s encounter to alter another person’s knowledge. Such communication can preserve useful evidence, propagate error, and establish practices that outlive any individual nervous system. A private diary can become part of a shared environment.
That is the transition to Unit 8. Chapter 48 develops the reciprocal relation between organisms and the physical and social worlds they construct, including cultural inheritance. Experiences transmitted across people help shape the conditions in which later brains develop and act. The continuity is not that every memory becomes public, but that knowledge no longer has to be acquired independently by every body that uses it.
This unit began with an animal moving through a world that extends beyond its present senses. It ends with the particular encounters that keep learned relations responsive to change. Past events matter because they alter present and future control. That claim gives memory a biological job without reducing it to prediction or making accuracy optional.
A predictive map gives allostasis a geography. The diary keeps that geography honest.
Established findings. Latent learning separates acquisition from rewarded expression. Animal tasks demonstrate combined use of event content, location, order, or elapsed time under specified conditions. Kennedy and Shapiro’s lesions impair conditional memory use beyond the motivational and discrimination capacities tested in their controls; their recordings separately reveal task-dependent effects of internal state. Human amnesia can dissociate established knowledge and some learned performances from ordinary new event memory. H.M.’s later testing found severe remote autobiographical impairment alongside substantial remote knowledge and a few specific recollections. Tagged-cell experiments distinguish natural recruitment, intervention effects, and artificial accessibility.
The chapter’s organizing interpretation. Exploration prepares options for future needs. Particular encounters preserve evidence that qualifies general knowledge, while bodily condition and goals influence which information is used. Hippocampal–cortical interaction contributes to keeping overlapping events distinguishable and reconstructing relevant relations. The strong transfer story misdirects explanation when it treats survival of remote knowledge as survival of an unchanged episode. Complementary learning rates and systems reorganization do not make that inference valid.
Open questions. Which features of remote recollection continue to require hippocampal operations, under which anatomical and task conditions? How do need, schemas, and recent exceptions determine retrieval? Which reactivated content changes downstream learning, and when does artificial accessibility correspond to ordinary use? How do developmental adaptation and residual circuitry support knowledge and construction after early injury? The remaining questions concern the changing content and causal organization of memory, not a single destination to which every memory eventually moves.