How an Ancient Map Organizes Time, Tasks, Concepts, and Social Relations
46.1 The fish that inferred without moving
Chapter 45 began with a mobile animal leaving a refuge and ended with a map that could run beyond the body’s present location. The hippocampal–entorhinal system combined self-motion, direction, boundaries, landmarks, routes, and goals so that an animal could act on structure that was not fully visible from where it stood.
The next step does not begin with human abstraction. It begins with a goldfish.
A goldfish learns four overlapping discriminations: A is rewarded over B, B over C, C over D, and D over E. Each pair can be learned separately. The critical test presents B and D together for the first time. Nothing in the immediate display marks the correct choice. To prefer B, the fish must use the order implied by separately learned pairs.
Intact fish make that choice. Goldfish with lesions of hippocampal pallium learn the original premise pairs but fail the novel B-versus-D test [@soteloparrilla2025transitive]. The spared learning matters. The lesion does not erase the stimuli, the rewards, or every association. It removes the flexible use of relations among those associations.
No route through the tank solves that problem. Yet the result does not turn a fish into a formal logician. Transitive inference names the structure of the test, not a conscious syllogism performed in words. The experiment identifies a simpler operation: elements encountered separately can be organized so that their relations guide a choice not specified by the present cue.
This finding changes the evolutionary question. Physical navigation remains the clearest basal problem because it supplies measurable states, movements, boundaries, and goals. It also gave an ancient vertebrate an obvious adaptive advantage. An animal that could estimate where it was and which route led to food or shelter wasted less time and energy than one restricted to visible cues or fixed response chains.
But navigation was relational from the beginning. A place is not a colored patch inside the head. It is a state defined by relations among boundaries, directions, routes, recent movement, and possible outcomes. A shortcut uses relations that have not been experienced in that exact combination. A detour requires the animal to preserve a destination while changing the route. The goldfish inference result exposes the same broad architecture outside literal location: separately learned elements can be placed into an organized structure and used beyond the current stimulus.
The conclusion is therefore neither “the hippocampus is a space module” nor “the hippocampus is a general relation faculty.” Both formulations turn a useful operation into a box.
Physical navigation supplied the basal ecological problem and remains the clearest evolutionary anchor. But an ancient hippocampal or medial-pallial system already organized states, contexts, events, and transitions in ways that could guide behavior beyond the present cue. Later brains expanded what could enter that architecture, which goals could orient it, and what actions or inferences could be guided by it.
The rest of this chapter asks how far that claim can be taken. Time, sound, task structure, concepts, and social relations can all be organized in map-like ways. They do not all become copies of a room. The scientific task is to identify what the states are, how they change, and what the learned structure allows the animal to do.
46.2 What deserves to be called a map?
The word map is useful only if it excludes something.
A neural population that responds differently to two stimuli has distinguished them. That is not yet a map. A list has an order. A category has members. A similarity analysis can place almost any collection of activity patterns into a two-dimensional picture. None of those facts alone shows that the nervous system represents a structure through which behavior can proceed.
The overview introduced two terms that make the claim more exact. A state is a situation the animal can occupy. It may be a physical location, a moment within a sequence, a position along a task variable, or a context defined by recent events. A transition specifies what tends to follow that state after an action or event. A map preserves enough of those states and transitions to guide behavior when the next step is not supplied by one immediate cue.
That definition does not require a miniature floor plan. Different problems have different structures. A route through an enclosure may be continuous and approximately metric. A delay is ordered through time. A task can be a discrete graph in which some states lead to several possible successors. A social relation may change along dimensions such as affiliation or power. What makes these representations map-like is not that they can be drawn on paper. It is that relations among states support prediction, traversal, recombination, or inference.
Four levels of evidence should be kept separate.
- State tuning shows that activity varies with a nonspatial variable. A neuron may fire at one tone frequency or one moment in a delay.
- Relational organization shows that neural similarity, distance, direction, or connectivity preserves relations among several states.
- Flexible use shows that the learned structure supports a novel prediction, route, categorization, or inference.
- Causal dependence shows that perturbing the circuit selectively impairs that flexible use while sparing simpler routes to performance.
The levels accumulate. A firing field along a frequency axis is stronger evidence than an undifferentiated response to sound, but it does not by itself prove that the animal can infer an untrained route through auditory space. A sixfold fMRI signal in a learned conceptual plane is evidence for structured population coding, but it is not a direct recording from grid cells. A selective lesion contrast can establish causal dependence, but only for the demands actually imposed by that task.
This hierarchy allows firm conclusions without turning every result into the same claim. The goldfish experiment reaches the fourth level for one form of flexible relational use. The experiments that follow reach different levels with different methods.
Not every relation is a map, and not every map is Euclidean. A representation earns the name when relations among states support prediction, traversal, recombination, or action beyond what one present stimulus specifies.
46.3 Time: when the route is a sequence
Physical navigation provides an obvious succession of states because the body moves. A temporal task reveals what happens when the body remains in roughly the same place but the relevant situation continues to change.
The comparative evidence again starts with goldfish. In delay conditioning, one event overlaps with the event it predicts. The first stimulus is still present when the second arrives. In trace conditioning, the first stimulus ends and a stimulus-free interval follows before the second event occurs. The animal must bridge a period in which neither event is present.
Lesions of goldfish hippocampal pallium severely impair trace conditioning while sparing delay conditioning [@gomez2022trace]. The contrast cannot be explained by a general inability to perceive the stimuli, produce the response, or learn that one event predicts another. The lesion becomes consequential when the relation must survive across the empty interval.
Mammalian recordings show one way a hippocampal population can organize such an interval. Christopher MacDonald and colleagues trained rats on a task in which events were separated by a delay. Different hippocampal neurons fired at successive moments within that delay. Across the population, the cells formed an ordered sequence spanning the gap. When the temporal structure of the task changed, many cells changed the time at which they fired: the sequence retimed rather than acting like a clock with one permanently assigned neuron for each absolute second [@macdonald2011time].
These neurons became known as time cells, but the label should not suggest a dedicated stopwatch. Their activity also reflected location, behavior, and the larger event sequence. That mixture is functional. Time matters because something happened before, something is expected later, and the current moment occupies a position within that organized episode.
The control problem is concrete. The earlier cue is no longer available. The appropriate response depends on how far the animal has progressed through the interval and on which sequence is underway. A succession of hippocampal states can preserve that progress until the later event arrives.
This is not simply space translated into seconds. Spatial routes and temporal sequences differ. The animal can reverse direction in a corridor; elapsed time does not run backward. A boundary in a room and the end of a delay are not the same sensory event. Yet both problems require the nervous system to distinguish a current state by its relation to states that are no longer present and states that have not yet occurred.
The body need not change location for the hippocampal system to organize progress through a structured episode. When behavior depends on an event that has disappeared, successive neural states can bridge the gap and place the present within a learned sequence.
Time is therefore the cleanest first step away from literal place. It does not require language, conceptual instruction, or a large human neocortex. The goldfish lesion result places temporal bridging within an ancient vertebrate system. Mammalian time cells reveal a population mechanism through which an interval can acquire internal structure.
46.4 Moving through a tone
Time changes whether the animal acts or not. Dmitriy Aronov, Rhino Nevers, and David Tank designed a different task in which the animal actively controlled a nonspatial state.
A rat moved a joystick. Joystick movement changed the frequency of a tone along a continuous auditory axis. The animal learned to move through that frequency range in order to complete the task and obtain reward. Recordings from hippocampus and medial entorhinal cortex revealed neurons with firing fields at particular sound frequencies. The population represented other parts of the behavioral task as well, and many neurons active in the auditory task overlapped with cell populations active during spatial navigation [@aronov2017nonspatial].
The result is striking because tone frequency is not physical location. A high tone is not north of a low tone. The rat cannot walk around 8 kilohertz or shelter behind 12 kilohertz. The experiment nevertheless gave frequency the structure of a controlled task variable.
At any moment, the rat occupied a state defined partly by the current tone. Its joystick movement was an action. That action produced a predictable transition along the frequency axis. Auditory feedback reported the consequence of the movement. A target region and reward supplied an outcome. The rat was therefore not passively hearing a succession of sounds. It was controlling progress through a variable whose states and transitions mattered for action.
This is why the firing fields matter. A neuron responding to one frequency in an ordinary listening task might simply be auditory. A population whose fields tile positions along a behaviorally controlled variable, and whose members overlap with populations used during navigation, supports a stronger conclusion. Common hippocampal–entorhinal circuit mechanisms can organize task states that are not locations in the room.
The result does not show that the circuit laid a hexagonal floor over sound. The investigators found fields at particular frequencies and broad task-related activity; they did not demonstrate a complete auditory copy of the spatial grid system. Nor does the experiment show by itself that rats can take novel shortcuts through frequency. Its contribution is more precise:
The hippocampal–entorhinal circuit can organize movement through a continuous, behaviorally relevant variable even when that variable is not physical location.
The control perspective explains why. A controller needs a state estimate sufficient to predict how action will change the situation. Physical position is one such variable. Tone frequency became another because the rat’s action moved it predictably toward or away from an outcome. What enters the map depends not only on the sensory modality but on the structure of the task.
46.5 Learning a graph that was never shown
A continuous axis is only one kind of map. Many real problems are better described as graphs.
A graph contains nodes and links. A node is a possible state. A link specifies a possible transition. A subway map is a familiar example: two stations can be physically close but poorly connected, while distant stations can be linked by an express line. What matters for travel is not Euclidean distance alone but reachability through the network.
Mona Garvert, Raymond Dolan, and Timothy Behrens asked whether people could acquire such a structure from experience without ever being shown the graph. Participants viewed sequences of objects generated by random walks through a hidden network. One object tended to be followed by another when the corresponding nodes were linked. The participants performed cover tasks; they were not instructed to learn or describe the underlying graph.
During later fMRI, responses in the hippocampal–entorhinal system reflected relations produced by the earlier transition structure. The most informative account was not simply whether two objects had appeared next to one another. A weighted estimate of the states likely to follow each object—the object’s expected future under the experienced random walk—captured important neural and behavioral effects [@garvert2017map].
That result connects directly to the successor representation introduced in the overview. A successor-like code does not merely store the present state. It represents the states likely to follow from it under a learned pattern of behavior, with near and probable successors weighted more strongly than remote or unlikely ones [@stachenfeld2017predictive]. Two nodes can therefore acquire similar representations because they predict similar futures even if they were never presented as a pair.
What the participants learned was not a hidden floor plan. It was which states led toward which others and how much of the future each state predicted.
This point matters because it breaks the false equation between map and two-dimensional metric surface. A discrete graph can contain branches, loops, bottlenecks, one-way transitions, and clusters. Rock–paper–scissors is a directed cycle and can be drawn without contradiction as three nodes connected by directed edges. A subway line can loop. A task can return the animal to a previous state. Space does not forbid cycles, and neither does a state-transition map.
The graph also reveals how larger units can emerge from local transitions. Anna Schapiro and colleagues exposed participants to sequences generated from a network containing communities of mutually predictive items. Individual transition probabilities were arranged so that obvious surprise at a boundary was not required. Nevertheless, behavior and distributed cortical activity came to group items belonging to the same transition community [@schapiro2013neural]. Repeated local predictions caused a larger event structure to emerge.
That complementary result did not identify a hippocampal map by itself. It shows why transition learning is computationally powerful. An organism does not need to receive a labeled category or event boundary from outside. If some states repeatedly lead to one another, their shared future can make them cohere.
For a controller, graph structure may be more useful than geometric distance. A nearby resource is irrelevant if a barrier makes it unreachable. A state that lies several steps away may be important if every route passes through it. A bottleneck can deserve disproportionate representation because it controls access to many futures. A cluster can define a context because transitions within it are common and transitions out of it are rare.
A useful map need not preserve physical distance. It may preserve reachability, transition probability, bottlenecks, communities, or expected future occupancy.
The map metaphor earns its keep here because the learned relations permit more than recognition. They allow the system to anticipate where a state leads, group states by shared futures, and answer questions about a structure that was never explicitly presented.
46.6 From feature space to concept space
Graphs show that a map can be discrete. Human experiments also ask whether a continuous nonspatial structure can recruit a signature associated with grid coding.
Alexandra Constantinescu, Jill O’Reilly, and Timothy Behrens created an artificial two-dimensional domain from cartoon birds. Each bird occupied a position defined by neck length and leg length. Participants learned the relations among the birds and later made judgments that required movement through the learned feature space. The investigators estimated the direction of those inferred trajectories.
During fMRI, activity in entorhinal and connected regions varied with trajectory direction in a sixfold pattern. Trajectories aligned with an estimated grid orientation produced a different population signal from misaligned trajectories [@constantinescu2016conceptual]. Sixfold directional modulation is the population-level signature expected when many grid-like units with a common orientation are sampled through BOLD imaging.
The result is not a direct recording from human grid cells. fMRI averages activity across large populations, and a sixfold model can be tested without observing the local firing fields that define a grid cell in a rodent. The result also does not show that all concepts are arranged hexagonally. The experiment deliberately created a continuous two-dimensional domain and asked participants to infer trajectories through it.
Within those limits, the conclusion is substantial:
When conceptual knowledge is learned as a continuous two-dimensional relational structure, the human entorhinal system can express a population-level directional signature analogous to grid coding in physical space.
The phrase conceptual space can still be too generous. Neck length and leg length are stimulus features, and an experimenter can assign coordinates to them whether or not the participant uses those dimensions to define a concept. Stephanie Theves, Guillén Fernández, and Christian Doeller designed a task that separated the complete feature space from the dimensions that mattered for categorization.
Participants learned objects varying along three continuous features. Only two dimensions were relevant to the learned concept; the third had to be perceived and remembered but did not determine category membership. Hippocampal activity reflected distances defined by the conceptually relevant dimensions more strongly than distances in the full feature space [@theves2020concept]. The map did not reproduce every available property of the stimuli. It emphasized the dimensions that organized behavior.
That selectivity is exactly what a control system requires. The world contains more variables than any controller can represent with equal precision. A useful state estimate preserves differences that predict consequences and collapses differences that do not matter for the current task. The rat in the auditory experiment did not need a complete model of sound. It needed the dimension its action could control. The person learning a concept did not need all features weighted equally. The relevant dimensions defined which items belonged together and how a novel item should be treated.
A 2026 developmental study extended this line of work. More than two hundred participants from eight to twenty-five years of age learned a deliberately two-dimensional knowledge map in which objects varied along two ranks. Entorhinal grid-like modulation became stronger with age and covaried with inferential performance; related distance codes in medial prefrontal cortex also changed across development [@qu2026development]. The study links the maturation of map-like population signals with increasingly effective use of a learned relational structure.
It does not show that intelligence is a grid code. The design was cross-sectional, the knowledge domain was constructed as a two-dimensional map, and the neural measures were correlational. Its importance is more specific: children and young adults differed not only in remembering trained pairs but in the neural organization that supported inference across a structured domain.
These experiments identify a progression of claims. The Constantinescu study found a grid-like directional signal in a learned two-dimensional feature space. The Theves study showed that hippocampal distance coding selected the dimensions that defined the concept rather than copying every feature. The developmental study linked similar map-like organization with flexible inference across age.
A conceptual map is not a warehouse containing every property of an object. It is a task-shaped relational structure that preserves the dimensions needed to categorize, predict, or infer what follows.
46.8 What evolution added without replacing the core
The expansion from location to time, task, concept, and social relation raises a historical question. Was the hippocampal system originally spatial and later borrowed for every nonspatial use, or was it relational from the beginning?
The evidence supports a more useful answer than either slogan.
Space has explanatory priority. A mobile animal must continually estimate its relation to a world that extends beyond current sensation. Vestibular, proprioceptive, motor, visual, and tactile signals provide direct evidence about self-motion and orientation. Boundaries and landmarks correct accumulating error. Goals such as shelter, food, water, or escape occupy actual locations. The ecological advantage is immediate and the variables are measurable.
Development also shows that directional organization is deeply prepared. Head-direction cells are present in rat pups before eye opening. They are not present before all experience: the pups have already received vestibular, proprioceptive, tactile, motor, and social input. Their preferred directions are initially unstable and drift together; visual experience rapidly anchors and stabilizes the directional system [@bjerknes2015coherence]. Place-related activity is present early, while stable adult-like grid organization develops later [@langston2010development; @wills2010development].
That sequence shows that spatial orientation has specialized bodily inputs and an early developmental scaffold. It does not show that the wider hippocampal system was ever incapable of organizing time or learned order. A relational architecture can receive evolutionarily prepared input for an especially important domain.
The comparative lesion evidence supplies the other half of the argument. Goldfish hippocampal pallium is required not only when location must be defined by relations among environmental cues, but when events must be associated across a temporal gap and when overlapping premise pairs must support an untrained choice [@rodriguez2002conservation; @gomez2022trace; @soteloparrilla2025transitive]. Those experiments do not recover the exact repertoire of the common ancestor of teleosts and tetrapods. They do make a pure space-only ancestor difficult to defend as the established starting point.
The basal operation can be stated concretely. The animal must distinguish states that should not control behavior in the same way. It must bind cues to context and sequence. It must learn which states tend to follow which others. It must preserve enough of that structure for an absent event or hidden goal to influence the next action.
Physical navigation is the most visible expression of that operation. Later pallial and cortical expansion changed its reach.
Association cortices supplied richer information about objects, scenes, sounds, and agents. Expanded hippocampal and entorhinal circuits differentiated more complex contexts and events. Amygdalar and orbitofrontal systems supplied more varied outcomes. Striatal systems supplied learned action values. Frontal systems maintained remote goals, rules, and alternatives that could select which state dimensions mattered. Language allowed another person to specify a relation the listener had never directly experienced. Cultural artifacts placed maps, diagrams, schedules, genealogies, and records outside the nervous system and fed their structure back into it.
Those additions did not turn a fish map into a detached human faculty called reasoning. They added layers of control over an inherited circuit. Later systems changed which cues defined a state, which transitions were relevant, which goals oriented sequence generation, and how the resulting structure was read out into action or inference.
Evolution did not replace the map when behavior extended beyond physical space. It changed what could enter the map, what goals could orient it, and what actions or inferences could be guided by it.
Stephen Jay Gould and Elisabeth Vrba introduced exaptation for a feature that is useful in a current role but was not built by natural selection for that role [@gouldvrba1982exaptation]. The concept separates present utility from historical origin.
Some modern uses of hippocampal–entorhinal machinery are plausible exaptations. The ancestral vertebrate system was not selected so that humans could learn fictional birds in a laboratory, follow an organizational chart, or construct a culturally defined genealogy. Expanded cortical systems can co-opt older state-and-transition machinery for problems that did not exist in the ancestral niche.
That does not make all nonspatial hippocampal function one late exaptation. Temporal bridging and flexible ordered relations are present in teleost experiments. Navigation itself already requires relations among cues, contexts, routes, and goals. Modern function may therefore combine several histories: conservation of ancient operations, elaboration of those operations within enlarged circuits, and exaptation for genuinely novel domains.
The evidence supports physical navigation as the basal ecological anchor. It does not establish one simple sequence in which an exclusively spatial organ was suddenly borrowed for every later form of structured thought.
46.9 Where the map analogy stops
The success of the map idea creates its own danger. Once any structured representation is called a map, the word explains nothing.
A neural pattern is not a cognitive map merely because an analyst can plot it. Spatial language is not enough. Faster responses to small numbers on one side of space would show a learned spatial association, not a hippocampal map. A category hierarchy is not automatically traversed as a state-transition structure. Familiar words and established semantic knowledge do not become hippocampal because they are related to other words and facts.
The strongest tests ask what the representation does.
Does neural organization preserve relations beyond sensory similarity? Does changing the transition structure reorganize the code? Does the representation predict a novel route, successor, category judgment, or inference? Does disrupting the circuit remove that flexible use while leaving simpler performance intact? These questions distinguish an explanatory map from an attractive diagram.
Different domains also need not share one geometry. The time-cell sequence is ordered but not a two-dimensional plane. The auditory task contains a continuous axis. The hidden task structure is a discrete graph. The bird experiment deliberately creates a two-dimensional metric space. The social task imposes two dimensions on an evolving relation. Shared participation of hippocampal–entorhinal circuitry does not require those representations to become identical.
Those boundaries cannot be drawn by claiming that maps must be continuous or that nearness is transitive. If A is near B and B is near C, A need not be near C. A directed cycle can be represented as a graph. Loops are fundamental to navigation: an animal can leave a refuge, traverse a circuit, and return. The head-direction system itself represents a circular variable in which a full rotation returns to the starting direction.
High dimensionality creates a real problem, but not because a controller must fit the entire world onto a literal sheet. A useful controller selects a state description sufficient for the present demand. Theves and colleagues found exactly this selectivity: hippocampal distances emphasized concept-relevant dimensions rather than every feature [@theves2020concept]. Compression inevitably discards information. That is not a failure if the discarded dimensions do not alter the predicted consequence or required action.
Some structures may depend heavily on other systems. Formal proof, grammar, recursive syntax, and deeply nested rules cannot be assigned to the hippocampus merely because their elements stand in relations. They involve specialized sensory, semantic, frontotemporal, and control operations. The relevant question is not whether a whiteboard can depict them. It is whether hippocampal state-and-transition organization makes a measurable and necessary contribution under a defined task demand.
The same boundary applies to routine reading and reasoning. Reading this sentence depends on visual analysis, language networks, established lexical and semantic knowledge, attention, and task control. The hippocampus may contribute when a new relation must be learned, when context disambiguates meaning, or when separated statements must be integrated into a novel inference. It is not the organ that reads, thinks, or understands.
Finally, a map does not determine why the animal should move through it. The hippocampus can preserve a route to both water and shelter. Thirst changes the value of one destination. Threat changes the value of the other. Frontal and action-selection systems determine which possibility governs behavior. The map supplies structured alternatives inside a distributed controller; it is not the controller as a whole.
Shared operations do not require identical representations. The same hippocampal–entorhinal system can participate in several domains without turning them all into copies of physical space or becoming a general-purpose organ of thought.
46.10 Coda: from structured domains to particular events
This chapter began with a fish expressing an untrained relation and ended with human concepts and social trajectories. Across those examples, the hippocampal system helped preserve structure beyond the currently available cue. A present state acquired meaning from what preceded it, what could follow it, and how it related to other learned states.
A structured domain is not yet a memory for one particular occurrence.
The same room can contain breakfast today, an argument tomorrow, and an uneventful hour next week. The same person can be encountered as a colleague in one context and a friend in another. The same sequence of actions can lead to a different outcome because of one object, one bodily state, or one decision. An episode must remain distinguishable from other events that overlap with it in place, time, people, and meaning.
That problem requires both separation and binding. Where did the event occur? When did it occur? Who was present? What happened first? Which action produced the outcome? What did the body need or feel? Later retrieval must reconstruct enough of that conjunction from partial cues without confusing it with every similar event.
The next chapter, The Diary on the Map, turns to that problem. It begins with the patient who made the hippocampus central to modern memory research, but it will not treat his case as the origin of the system’s function. H.M. enters after the evolutionary and physiological foundation has been built. His amnesia reveals what is lost when a brain can still perceive, act, speak, and use much of its established knowledge, yet cannot normally bind the changing stream of experience into new, retrievable events.
Established findings. Teleost hippocampal-pallium lesions impair association across a temporal gap and flexible use of overlapping learned relations while sparing simpler delay conditioning or trained premise discriminations. Mammalian hippocampal populations form sequential time fields and fields along a behaviorally controlled auditory variable. Human hippocampal–entorhinal activity reflects learned transition graphs, conceptual dimensions, and trajectories through an experimentally defined social space. Entorhinal fMRI signals can show sixfold directional modulation in deliberately constructed two-dimensional nonspatial domains.
The chapter’s organizing interpretation. Physical navigation is the basal ecological anchor of hippocampal function because it requires an embodied animal to estimate states and transitions beyond current sensation. That navigation system was relational from the beginning: places are defined by context, route, direction, boundaries, and possible outcomes. Ancient medial-pallial systems also contributed to temporal bridging and flexible ordered relations. Later pallial, cortical, and cultural systems expanded the variables, goals, and readouts available to the inherited state-and-transition architecture.
Open questions. Comparative evidence cannot recover the exact behavioral repertoire of the first vertebrate hippocampal system. Human fMRI signatures do not establish that every nonspatial domain is represented by local grid cells or by one shared geometry. The field has not settled how task goals select relevant dimensions, when successor-like coding best explains the representation, how several domain-specific inputs are combined, or which nonspatial inferences depend causally on hippocampal–entorhinal circuitry. These questions delimit the account without reducing the system to either a space-only module or an abstract relation faculty.
46.7 Social relations as changing states
Social animals inhabit a world in which other animals are among the most consequential and least stationary features.
A dominant individual may lose rank. An ally may defect. A parent may become unavailable. An unfamiliar animal may become a mate, competitor, predator, or source of information. The relevant state is not simply the other animal’s identity. It includes the current relation between that individual and the acting animal, the history that produced it, and the actions now possible.
Rita Tavares and colleagues created a role-playing task in which participants interacted with fictional characters while trying to obtain a home and a job. Each interaction altered the participant’s relation to a character along two experimenter-defined dimensions: power and affiliation. Choices therefore produced trajectories through a two-dimensional model of social relations. Hippocampal activity varied with aspects of movement through that social space, and the strength of some effects was related to individual differences in social skill [@tavares2015social].
The experiment shows that hippocampal activity can track an evolving relational trajectory when social interactions are organized along learned dimensions. It does not establish that the hippocampus contains a universal coordinate system for social life. Power and affiliation were useful experimental variables, not the complete geometry of real relationships. Friendship, kinship, obligation, reputation, trust, threat, attraction, and institutional role do not reduce to two axes.
Nor is social mapping uniquely human. Fishes, birds, and mammals distinguish individuals, remember prior encounters, track dominance, return to social partners, and alter behavior according to who is present. What human cortical and cultural systems add is an enormous expansion of the states that can matter. A person can occupy the role of colleague, chair, rival, parent, citizen, creditor, or witness. Language names these relations. Institutions preserve them. Gossip and records transmit relational information beyond direct encounter.
The hippocampal system participates inside that larger controller. It can help distinguish particular people, places, interactions, and trajectories. Amygdalar and orbitofrontal systems contribute affective significance and outcome identity. Striatal systems learn the value of possible social actions. Frontal systems maintain goals, rules, obligations, and alternatives. Language and cultural knowledge define roles that no isolated navigation circuit could invent.
The distributed architecture prevents a common overreach. A social trajectory is not valuable merely because it has coordinates. It matters because bodily state, attachment, threat, status, and current goals assign consequences to possible transitions. The map preserves relations; other systems determine which destination matters and which action should be taken.