Prospective cognition depends on the capacity of the hippocampus to construct, evaluate, and update internal models of future events rather than merely storing past experiences. Far from being a passive archive, hippocampal circuits dynamically recombine fragments of memory into novel scenarios that can guide behavior before those situations are encountered in reality. This generative function emerges from the interplay of place cells, time cells, and pattern completion mechanisms that together transform discrete episodes into flexible simulations of what might happen next.
A central feature of hippocampal contribution to prospection is its role in relational coding. Individual experiences are encoded not as isolated snapshots but as networks of relationships among people, places, times, and outcomes. When anticipating the future, the hippocampus can selectively reactivate and recombine these relational codes to infer likely consequences of potential actions. This relational flexibility allows a limited set of prior experiences to support an enormous space of imagined futures, enabling efficient generalization beyond direct experience.
Electrophysiological recordings in rodents navigating spatial environments provide a detailed window into these mechanisms. During active movement, place cells fire in sequences that reflect the animal’s current trajectory. During rest or pauses in behavior, the same network spontaneously generates rapid sequences that either replay recent paths or, critically, preplay paths the animal has not yet traversed. This preplay phenomenon illustrates how the hippocampus can internally simulate future trajectories along a maze or environment, effectively running through candidate paths prior to overt behavior. Such sequences often unfold within a single theta cycle, compressing potential futures into brief bursts of neural activity suitable for comparison and selection.
Sharp-wave ripple events in the hippocampus are especially important for prospective sampling. These high-frequency oscillations, typically occurring during rest or slow-wave sleep, are associated with rapid sequential activation of cell assemblies representing locations or states. While initially interpreted purely as replay of past experience, subsequent work has shown that ripple-associated sequences frequently depict paths that lead to distant goals or unexplored options. The content of these events can be biased by current motivational states, ongoing tasks, and expected rewards, indicating that these sequences reflect goal-directed prediction rather than a neutral recapitulation of memory.
Theta oscillations organize another layer of prospective processing by segmenting ongoing experience into discrete cycles that can carry successive “snapshots” of possible future states. Within a single theta cycle, the firing of place cells can sweep ahead of the animal’s current location, representing positions that lie further along a potential route. Across cycles, these sweeps can branch, representing alternative upcoming paths or decision points. This temporal structuring allows the hippocampus to multiplex information about present position, immediate future states, and more distal possibilities, forming a continuously updated internal trajectory of where the organism could go next.
At decision points in spatial tasks, hippocampal activity often exhibits sequences that “look ahead” down competing arms of a maze, with alternation between representations of one path and the other. These look-ahead sequences provide a neural substrate for prospective evaluation: the hippocampus effectively samples different potential futures before a commitment is made. Downstream structures, such as the prefrontal cortex and ventral striatum, can then use these sampled trajectories to estimate expected value, risk, and goal attainment likelihood, integrating hippocampal simulations into broader decision circuits.
The hippocampus also deploys prospective coding beyond spatial navigation, supporting simulations of social, semantic, and abstract futures. Human neuroimaging has shown that activity patterns in hippocampus during episodic memory retrieval closely resemble those during imagination of novel future events. The degree of overlap between past and future patterns predicts the richness and detail of imagined scenarios, consistent with the idea that prospection relies on strategic recombination of episodic memory traces. Damage to the hippocampus impairs not only recollection of past experiences but also the capacity to imagine detailed, coherent future events, underscoring a shared neural basis for remembering and forecasting.
Microcircuit mechanisms within the hippocampus support this generative capacity through a balance of pattern separation and pattern completion. In the dentate gyrus, sparse coding and strong inhibition facilitate pattern separation, allowing similar experiences to be encoded as distinct representations that reduce interference. In CA3, recurrent collaterals enable pattern completion, where partial cues trigger reactivation of full stored patterns. During prospective cognition, partial information about the current context can cue CA3 to reconstruct relevant episodes, which can then be flexibly recombined to generate hypothetical outcomes. This dynamic toggling between separating and completing patterns allows the system to avoid both rigid reuse of old episodes and unconstrained imagination.
Temporal coding adds another dimension to hippocampal prospection. Time cells fire at specific moments during structured intervals, providing a scaffold for ordering events in time. When constructing a possible future sequence, the hippocampus can recruit time cells to stitch together events into a coherent temporal narrative, not just a spatial path. This temporal scaffolding helps predict when outcomes are likely to occur, which is crucial for learning contingencies, estimating delays to reward, and planning sequences of actions.
Interactions with neuromodulatory systems shape how the hippocampus prioritizes particular futures. Dopamine projections from the midbrain convey reward prediction errors and motivational salience, biasing hippocampal plasticity toward events that signal potential gains or losses. Acetylcholine influences the balance between encoding new information and retrieving or simulating existing patterns, with higher cholinergic states favoring the incorporation of current sensory input and lower states facilitating retrieval and internal simulation. These modulatory influences enable the hippocampus to align its prospective computations with current goals, uncertainties, and environmental demands.
Hippocampal prospection is further supported by its extensive connectivity with the prefrontal cortex. Prefrontal areas can provide abstract goals, rules, and constraints that shape which memories are retrieved and how they are combined. In turn, hippocampal simulations inform prefrontal evaluation of strategies and contingencies. This bidirectional loop allows the hippocampus to do more than replay arbitrary segments of experience; it selectively generates simulations that are relevant to ongoing tasks, preferences, and long-term objectives, refining the space of futures explored.
Evidence from lesion and stimulation studies in humans highlights the behavioral consequences of disrupting these mechanisms. Individuals with hippocampal damage often show intact semantic knowledge about the future (for example, understanding that they might one day retire or move) but are unable to construct vivid, scene-based simulations of what those futures would actually feel like. They produce impoverished, fragmentary descriptions with diminished spatial and temporal structure. Noninvasive stimulation targeting hippocampal networks can, in contrast, enhance the detail and consistency of imagined future events, suggesting that modulating hippocampal excitability can directly influence prospective cognition.
A key computational implication of these findings is that the hippocampus functions as a generator of candidate futures that can be sampled, evaluated, and pruned by other brain systems. Rather than computing a single deterministic prediction, hippocampal circuits appear to produce a series of discrete, compressed trajectories or event sequences that represent different possibilities. These samples often incorporate variability and stochasticity, reflecting uncertainty about how the environment might unfold. In this sense, the hippocampus operates as a future sampler, transforming stored experience into a stream of imagined options that support flexible, adaptive behavior in dynamic environments.
Computational models of sampling future possibilities
Computational accounts of the hippocampus as a future sampler typically start from the premise that the brain approximates a Bayesian inference machine. Within this “Bayesian brain” framework, the hippocampus is not merely retrieving stored traces but drawing stochastic samples from a distribution over possible futures, constrained by past experience and current contextual cues. These samples reflect the organism’s learned priors—statistical regularities extracted from memory—and are combined with incoming sensory information to generate predictions about likely trajectories of events. Rather than encoding a single best-guess plan, the system maintains a distribution over options, and hippocampal dynamics implement a form of Monte Carlo exploration of that distribution.
One influential family of models formalizes hippocampal function as sampling-based reinforcement learning in a state space defined by relational structure. Here, each discrete episode provides a transition between states: moving from one place to another, changing from one social configuration to another, or shifting from one abstract rule to the next. These transitions can be represented as a graph or Markov decision process, where edges encode learned probabilities and rewards. The hippocampus, in this view, runs internal simulations by generating sequences that traverse the graph, effectively performing model-based planning. Preplay and replay sequences recorded during sharp-wave ripples correspond to sampled paths through this internal model, which can then be used to update value estimates and policies even in the absence of overt behavior.
Succession of states over time is captured particularly well by successor representation models, which have become central in connecting hippocampal coding with prospective computation. In these models, each state is represented not only by its immediate features but also by the expected discounted occupancy of future states reachable from it. The hippocampus is hypothesized to encode such predictive maps, with place cells and related ensembles reflecting how often, and how soon, one state leads to another under typical behavior. Sampling future possibilities then corresponds to activating sequences consistent with the successor representation, allowing the system to quickly estimate long-term consequences of actions without recomputing full trajectories from scratch. This predictive coding naturally supports fast generalization when goals or rewards change but the environment’s structure remains stable.
Another line of computational work emphasizes that hippocampal sequences during theta and ripple events resemble trajectories generated by algorithms for probabilistic planning, such as Monte Carlo tree search. In these models, the brain incrementally grows a search tree of possible futures, guided by both learned value estimates and uncertainty. The hippocampus provides candidate rollouts: simulated paths that extend forward from the current state or backward from desired goals. These rollouts can be biased toward high-value or high-uncertainty regions of the state space, improving the efficiency of exploration. Alternating forward sweeps down different maze arms in rodents is naturally interpreted as alternating Monte Carlo probes of competing options, with downstream circuitry integrating the sampled outcomes.
Sampling models also offer a principled explanation for variability and apparent “noise” in hippocampal activity. Instead of attributing irregularity to biological imperfection, these frameworks interpret variability as a functional reflection of uncertainty. When the environment is predictable and well learned, sampling converges on a narrow set of high-probability trajectories, yielding stable sequences. When ambiguity is high or the system encounters novelty, the sampled futures become more diverse and exploratory, broadening the distribution of imagined possibilities. In this way, stochastic generation of sequences supports an adaptive balance between exploiting known high-value paths and exploring alternatives that might reveal better options.
Computationally, the hippocampus can be cast as operating in different regimes depending on neuromodulatory state and behavioral demands. High acetylcholine levels during active exploration are modeled as biasing the system toward encoding and updating transition statistics, refining the internal model. Lower acetylcholine and increased sharp-wave ripple activity during rest or quiet wakefulness are modeled as shifting the system into a generative mode, where previously acquired transitions are recombined into simulated trajectories. Under this view, prospection emerges from periodic switches between online learning and offline sampling, with each mode supporting the other: better models yield more accurate predictions, and richer simulations guide more targeted data collection.
Generative models inspired by machine learning offer additional insights into how the hippocampus may construct novel futures. Variational autoencoder–like architectures, for example, separate a latent representation of abstract situation structure from concrete sensory details. The hippocampus can be seen as encoding compressed latent variables that capture relational and temporal structure across episodes, while cortical areas carry more detailed content. Sampling in latent space allows the system to generate new combinations of known elements, mirroring the recombinative nature of episodic prospection. This framework explains how the hippocampus can imagine plausible but never-experienced scenes by drawing from a learned generative model constrained by memory.
Graph-based and relational models further formalize how the hippocampus supports flexible sampling across different domains, not just physical space. In these models, nodes represent entities (locations, people, concepts) and edges represent learned relations or transitions. Hippocampal activity patterns correspond to walks on this graph, and preplay-like sequences are computationally analogous to biased random walks or path-finding algorithms. When goals are specified, value or reward signals can be overlaid on the graph, and hippocampal sampling preferentially traverses paths leading toward high-value nodes. This provides a domain-general account of how the same circuitry can support navigation in physical, social, and conceptual spaces.
These computational perspectives also highlight a crucial role for priors derived from long-term statistics of the environment. Over development and learning, the hippocampus and associated networks internalize regularities such as typical path structures, temporal contingencies, and social patterns. When engaging in prospection, the system does not sample futures uniformly but heavily weights those consistent with these priors. For example, trajectories that obey environmental constraints, like walls or social norms, are far more likely to appear in simulated sequences than physically or socially impossible ones. This constraint-based sampling explains why imagined futures are usually coherent and structured rather than arbitrary, even though they are assembled from fragmentary memory traces.
At the algorithmic level, models differ in whether they treat hippocampal sampling as primarily forward-looking, backward-looking, or bidirectional. Some frameworks emphasize forward sampling from the current state to predict what could happen next, useful for real-time decision-making. Others highlight backward sampling from salient outcomes, such as rewards or punishments, to infer which prior states could have led there, supporting credit assignment and counterfactual reasoning. Empirical evidence for both forward and backward sequences suggests that the hippocampus may implement a mixture of these strategies, with different directions dominating under different tasks and motivational states. Such bidirectional sampling allows the organism not only to predict upcoming events but also to rethink past choices in light of new information.
Computational models stress that the output of hippocampal sampling does not by itself determine behavior; instead, it provides a structured set of candidate narratives about the future that must be evaluated by other systems. Prefrontal and striatal circuits can be modeled as integrating the sampled trajectories to compute expected values, risks, and policy updates. In this hierarchy, the hippocampus specializes in constructing a temporally extended hypothesis space—what could happen given what has happened—while downstream evaluative circuits select among these possibilities according to current goals and constraints. Framing hippocampal activity as sampling from a generative model thus unifies diverse empirical phenomena—replay, preplay, predictive maps, and flexible imagination—under a common computational principle of probabilistic prospection grounded in memory.
Neural representations of imagined trajectories
Imagined trajectories are expressed in the hippocampus as structured patterns of population activity that unfold over time, rather than as static snapshots of single locations or states. In spatial tasks, this structure is clearest in sequences of place cell firing that represent ordered paths through an environment. During theta oscillations, these sequences sweep from positions just behind the animal’s current location to positions far ahead, forming a compressed internal “movie” of potential movement. Importantly, the start and end points of these sweeps do not merely mirror what the animal is currently doing; they often incorporate future-directed elements that indicate where it might go next, revealing a neural code that anticipates rather than just reflects ongoing behavior.
Within a single theta cycle, the phase of neuronal firing carries a fine-grained temporal code that enables such anticipation. As the animal runs, place cells fire earlier or later in the theta cycle depending on whether their corresponding location lies behind, at, or ahead of the current position. This phase precession produces a temporally ordered sequence of spikes that encodes an extended trajectory across space in a fraction of a second. The hippocampus thus multiplexes multiple points along a path into a brief time window, effectively compressing a potential future route into a single oscillatory frame. When the animal approaches a decision point, this phase-structured representation can abruptly reorganize, with sequences alternating between distinct paths, suggesting that the system is momentarily simulating divergent futures.
Sharp-wave ripple events reveal complementary forms of trajectory representation. During these brief high-frequency bursts, large ensembles of hippocampal neurons fire in tightly coordinated sequences that replay past journeys or preplay novel ones. The ordering of spikes during a ripple mirrors spatial progression across the environment, but the sequences are sped up by orders of magnitude relative to real movement. In many cases, ripples encode routes that lead from the current position toward a goal, even if that route has not just been traversed. Such events can jump across intermediate positions, stitching together discontinuous segments of experience into a coherent virtual journey. The resulting patterns embody imagined trajectories that can leap in space and time, connecting the present situation to distal outcomes through internally generated transitions.
Human neuroimaging studies offer convergent evidence for trajectory-like representations in more abstract tasks. When individuals imagine navigating familiar environments, hippocampal activity patterns shift systematically as they mentally move from one landmark to another, with multivoxel patterns discriminating both current and upcoming imagined locations. Similar dynamics appear when participants mentally simulate nonspatial sequences, such as steps in a recipe or stages of a social interaction. The hippocampus shows graded changes that track the progression from earlier to later events in the imagined sequence, even though no physical movement occurs. These findings indicate that the same circuitry that supports spatial navigation also encodes trajectories through conceptual and episodic spaces, suggesting a domain-general code for ordered progression.
One key property of these representations is that they are relational rather than purely sensory. The hippocampus binds together elements such as locations, objects, actions, and outcomes into structured sequences, and imagined trajectories consist of reactivated relational links between them. During prospection, partial cues from the current context can trigger the reinstatement of specific relational chains: for example, seeing a particular person in a certain place can cue a remembered social interaction and then an anticipated consequence. At the neural level, this corresponds to the sequential reactivation of assemblies that were coactive during past episodes, but now arranged to extrapolate into possible futures. The same network that encoded the original event is thus repurposed to generate a plausible extension of it.
Time cells provide an additional scaffold for imagined trajectories, particularly for sequences where temporal order matters more than spatial layout. These neurons fire at specific moments within a structured interval, such as during a delay or while an animal waits for a reward. When animals learn that a certain sequence of events unfolds over a fixed duration, time cell ensembles come to represent successive temporal segments of that sequence. During internal simulation, similar temporal sequences can be reactivated in the absence of external timing cues, reconstructing the expected order and spacing of upcoming events. Imagined trajectories therefore have both a spatial and temporal architecture, with place cells and time cells jointly encoding “where” and “when” along potential futures.
Representations of imagined trajectories are shaped by priors derived from accumulated memory. The hippocampus does not explore the entire space of logically possible sequences; instead, it privileges trajectories that are consistent with the learned statistics of the environment. For example, sequences rarely jump across walls or violate stable transition rules that have been internalized over repeated experience. When reward contingencies change but the spatial layout does not, sequences rapidly reorient toward new goals while preserving lawful transitions between intermediate states. This reflects an internal generative model where the structural backbone of the environment constrains which imagined paths are considered, and motivational signals modulate how strongly different branches of that structure are expressed.
There is growing evidence that these representations are goal-sensitive rather than neutral. When animals are trained to navigate to different reward locations within the same maze, hippocampal sequences during pauses or ripples selectively depict trajectories leading toward the currently relevant goal, even if an alternative path is physically closer. In humans, imagining a future scenario with a specific desired outcome biases hippocampal activity toward patterns that resemble successful episodes from the past. Neural decoding studies show that, before making a decision, hippocampal patterns transiently align with representations of the option that will ultimately be chosen, as if the system is briefly inhabiting that future. Imagined trajectories are therefore not merely probabilistic extrapolations but are shaped by current goals and values.
These neural codes are not uniformly precise; they reflect graded uncertainty about future states. In familiar, predictable settings, sequences are tightly ordered and repeatedly follow similar paths, indicating a confident internal model. Under conditions of ambiguity or novelty, hippocampal trajectories become more variable, branching more frequently and exploring alternative routes. Some ripple events represent “off-task” or apparently irrelevant paths, which from a computational perspective can be understood as low-probability samples that hedge against incomplete knowledge. Variability in trajectory content thus serves as a neural marker of uncertainty and exploration, aligning hippocampal dynamics with probabilistic prediction rather than deterministic planning.
At the population level, imagined trajectories can be viewed as transient trajectories through a high-dimensional neural state space. Each point in this space corresponds to a particular pattern of activity across the hippocampal ensemble, and sequences of points trace curves that correspond to meaningful paths in the external or conceptual world. Techniques such as dimensionality reduction and manifold learning reveal that these neural trajectories are structured: sequences representing similar external paths occupy neighboring regions of state space, and trajectories that lead to shared goals converge onto similar endpoints. During prospection, the hippocampus traverses these internal manifolds, with different imagined futures corresponding to different routes through the same underlying landscape of neural states.
Critically, imagined trajectories can be compositional, drawing on subpaths or motifs that have been learned separately. In spatial experiments, sequences sometimes include shortcuts that the animal has never physically taken but could infer from the layout of the environment. In more abstract tasks, individuals can mentally stitch together known action segments into novel plans, and hippocampal patterns reflect transitions between previously unpaired segments. This compositionality suggests that the hippocampus stores reusable building blocks of experience that can be flexibly recombined, allowing imagined trajectories to extend beyond a simple replay of past episodes. The neural code thus supports a form of generative prospection, in which new paths are assembled out of familiar parts.
The interaction between hippocampus and neocortex further enriches these representations. While the hippocampus provides a sparse, sequential scaffold, cortical regions contribute detailed sensory, semantic, and emotional content to imagined trajectories. During mental simulation, hippocampal sequences appear to cue reactivation of modality-specific cortical patterns—visual scenes in occipital cortex, sounds in auditory areas, and social or conceptual information in temporal and prefrontal regions. The timing of these cross-regional activations suggests a leading role for hippocampal sequences in orchestrating the unfolding of the imagined scenario, with cortical contributions filling in the rich experiential texture along the skeleton defined by the trajectory code.
The directionality of hippocampal sequences reveals how imagined trajectories can support both prediction and counterfactual reasoning. Forward sequences, which progress from current or recent states toward possible future outcomes, are well suited for anticipating consequences of actions. Backward sequences, which run from salient outcomes back through preceding states, can encode inferred precursors of those outcomes. Both types have been observed around reward delivery and decision points, with forward events dominating before choices and backward events more common after significant outcomes. This bidirectional coding allows the same circuitry to generate prospective paths and to reconstruct alternative routes that could have led to different results, embedding imagined trajectories at the core of flexible, experience-based reasoning about what might happen and what might have happened.
Interactions between memory, prediction, and decision-making
The translation of hippocampal simulations into concrete choices depends on their interaction with broader memory systems, predictive mechanisms, and decision-making circuits. Rather than operating in isolation, the hippocampus sits at the center of a loop in which stored episodes, current goals, and expected outcomes continuously shape one another. Memory provides the raw material—encoded experiences and relational structures—that prospection recombines into candidate futures. Prediction imposes constraints on which recombinations are plausible or likely. Decision mechanisms, in turn, select among sampled futures and feed back signals about success and failure, thereby updating which trajectories are prioritized in subsequent simulations.
Within this loop, episodic memory functions as a database of structured experiences that can be queried and reconfigured to support planning. When an organism faces a choice, partial cues from the current context trigger hippocampal retrieval of relevant episodes: places that look similar, past actions in comparable situations, or prior outcomes associated with similar goals. These recalled episodes are not retrieved verbatim; instead, their underlying relational scaffolds—who did what, where, and when—are extracted and used to construct hypothetical scenarios. In effect, memory supplies the building blocks for internal “what if” experiments, while the hippocampus orchestrates their assembly into temporally extended narratives.
Predictive processes impose additional structure on this reconstruction. The hippocampus and interconnected cortical areas do not treat all remembered episodes as equally informative; they weight them according to learned priors about how the world typically unfolds. These priors are distilled across many experiences and capture regularities such as stable spatial layouts, social hierarchies, or causal contingencies between actions and outcomes. When generating prospective sequences, the hippocampus preferentially samples transitions that align with these priors, leading to imagined futures that are coherent and statistically sensible rather than arbitrary recombinations of memory fragments. This alignment between memory-derived structure and prediction allows the system to extrapolate beyond direct experience while still respecting environmental constraints.
Interactions with prefrontal cortex are especially important for translating hippocampal simulations into goal-directed decisions. Prefrontal regions maintain representations of current objectives, rules, and constraints, and they use this information to bias which memories the hippocampus retrieves and how those memories are recombined. For example, when an animal or human shifts from seeking safety to seeking reward, prefrontal signals can alter which past episodes are deemed relevant and which imagined trajectories are explored. In turn, hippocampal output provides prefrontal cortex with temporally extended scenarios that include intermediate steps, potential obstacles, and anticipated outcomes, giving decision circuits a rich substrate over which to compute expected value, risk, and opportunity cost.
Striatal circuits, particularly in the ventral striatum, integrate hippocampal trajectories with reward-related learning. During sharp-wave ripple events that depict paths leading to a goal, striatal neurons often show activity patterns correlated with the expected value of those paths. Dopaminergic signals conveying reward prediction errors arrive concurrently, allowing the system to strengthen simulated trajectories that predict positive outcomes and weaken those associated with negative or unrewarded outcomes. This mechanism effectively trains the hippocampus to favor particular subsets of its generative repertoire in future deliberations, shaping not just which actions are chosen but also which futures are likely to be imagined at subsequent decision points.
An important implication of this architecture is that decision-making can proceed either through online deliberation or through policies distilled from prior simulations. In novel or high-stakes situations, hippocampal sequences around decision points tend to be more frequent and diverse, suggesting explicit mental exploration of alternatives before choice. The organism appears to pause, allowing multiple candidate routes to be simulated and evaluated. With repeated experience, however, the need for overt hippocampal deliberation diminishes as striatal and cortical circuits internalize a more habitual mapping from states to actions. Nonetheless, even in well-learned tasks, brief hippocampal events can still punctuate behavior, especially when contingencies change or uncertainty rises, providing a mechanism for rapid replanning when established policies no longer suffice.
The same machinery that supports planning also underlies counterfactual reasoning, where the system evaluates not only what could happen but also what could have happened under different choices. Backward hippocampal sequences that trace from outcomes to prior states provide a neural substrate for this kind of evaluation. When an outcome is worse than expected, these backward trajectories can highlight alternative paths that might have led to better results, enabling prefrontal and striatal circuits to adjust future choices. In this way, the interplay of memory and prediction supports both prospective decision-making and retrospective re-evaluation, ensuring that past experiences continue to inform future behavior in a flexible, context-sensitive manner.
Interactions with semantic and schematic memory systems add another layer of organization. While the hippocampus specializes in episodic detail and specific trajectories, cortical networks encode abstract knowledge about typical event structures—scripts for restaurant visits, medical appointments, or social gatherings. During prospection, hippocampal sequences are often embedded within these broader schemas, which provide expectations about the general order of events even when specific episodes are missing. This fusion of episodic and schematic components allows decisions to be guided by both concrete past experiences and generalized knowledge, especially in situations that are novel in detail but familiar in structure.
Emotional and motivational states further modulate this integrated system. Amygdala and orbitofrontal inputs can tag certain memories as more salient or affectively charged, biasing the hippocampus to retrieve and simulate trajectories that involve threat, reward, or social evaluation. Under anxiety, for example, negative episodes and worst-case scenarios may be preferentially sampled, skewing prediction toward adverse outcomes and influencing risk-averse choices. Conversely, optimistic or reward-focused states can tilt simulations toward positive trajectories. These biases reveal that decision-making is not only a function of objective memory content and learned priors, but also of momentary affective context that shapes which futures are most vividly inhabited in the imagination.
Evidence from behavioral experiments illustrates how these neural interactions manifest in everyday decisions. When people are asked to choose between long-term benefits and immediate gratification, the richness and vividness of their imagined future outcomes predict their tendency to wait for larger delayed rewards. Tasks that encourage detailed episodic simulation of future payoffs enhance patience, and neuroimaging shows increased coupling between hippocampus and prefrontal regions during such interventions. This suggests that strengthening the link between memory-based prospection and evaluative circuits can systematically shift preferences, highlighting a direct route from the quality of imagined trajectories to the quantitative parameters of choice, such as discounting of future rewards.
Across these examples, a common theme is that memory, prediction, and decision-making form a tightly coupled system rather than a serial pipeline. The hippocampus continuously transforms stored episodes into probabilistic forecasts; prefrontal and striatal regions continuously query, reshape, and evaluate those forecasts; and feedback from outcomes continuously sculpts which episodes and trajectories will be mobilized in the future. Viewing the hippocampus as a future sampler embedded within this broader network clarifies how remembering the past and imagining the future jointly determine the choices organisms make, and why disruptions in any component of this interactive system can reverberate across all three domains.
Implications for understanding planning and psychiatric disorders
Viewing the hippocampus as a sampler of possible futures reframes planning as a probabilistic process grounded in memory rather than as a fixed, stepwise algorithm. In this framework, effective planning emerges when internally generated trajectories are sufficiently rich, appropriately goal-directed, and accurately weighted by learned priors about the world. When these conditions hold, prospection provides a powerful basis for flexibly organizing action over extended time horizons. When they fail, planning can become impoverished, rigid, or pathologically biased toward particular kinds of imagined outcomes. This linkage between the quality of sampled futures and behavioral control has deep implications for understanding both normal goal pursuit and the wide spectrum of psychiatric disorders in which planning goes awry.
At the cognitive level, successful planning requires the ability to decompose long-range goals into intermediate steps, to foresee obstacles and detours, and to compare multiple strategies before acting. The hippocampus supports these operations by generating candidate event sequences that bridge the present state to various distal outcomes. Prefrontal and striatal circuits then evaluate these sequences according to expected value, cost, and risk. If hippocampal sampling is too narrow—failing to consider enough alternatives—plans may be short-sighted or brittle. If sampling is too broad or poorly constrained by environmental structure, deliberation may become inefficient or dominated by unrealistic scenarios. Psychiatric conditions can often be interpreted as extreme cases along these dimensions of sampling breadth, constraint, and valuation.
One way to organize these implications is to distinguish disorders of content, structure, and control in hippocampal prospection. Disorders of content involve biases in what kinds of futures are represented—for example, preferential sampling of threat-related trajectories in anxiety. Disorders of structure involve disruptions in how trajectories are organized in time and space, as in schizophrenia, where imagined sequences can become fragmented or loosely tethered to reality. Disorders of control reflect impairments in the coordination between hippocampal simulations and prefrontal evaluation, contributing to impulsive or inflexible behavior in conditions such as addiction and obsessive–compulsive disorder. In each case, aberrant planning can be traced to specific alterations in how the future sampler operates within the broader decision network.
Anxiety and related internalizing disorders provide a clear illustration of maladaptive biases in prospective sampling. Individuals with high trait anxiety often report vivid, repetitive anticipation of negative outcomes—worry and rumination that center on what might go wrong. Neuroimaging studies show heightened coupling between hippocampus, amygdala, and medial prefrontal cortex when anxious individuals imagine future events, with preferential activation for threat-related scenes. Within a sampling framework, this pattern suggests that priors about danger and loss have become overweighted, leading the hippocampus to generate a disproportionate number of aversive trajectories. Prefrontal regions may then evaluate options within a future space that is already skewed toward worst-case possibilities, promoting avoidance, hypervigilance, and difficulty disengaging from worry.
These biases can also distort risk evaluation. If the hippocampus frequently preplays low-probability catastrophes, decision circuits may treat them as more likely than they are, effectively miscalibrating subjective probabilities. Conversely, positive or neutral futures may be underrepresented, reducing the motivational pull of adaptive plans. Such imbalances can explain why anxious individuals may overprepare for unlikely threats while neglecting everyday goals, and why reassurance based solely on factual information often fails: the underlying generative model continues to sample catastrophic scenarios with undue frequency and salience. Interventions that aim to rebalance prospective content—such as guided positive imagery or cognitive restructuring—can be seen as attempts to reshape the sampling distribution, not merely to change explicit beliefs.
Depressive disorders are associated with a different pattern: a pervasive reduction in the richness, specificity, and motivational value of imagined futures. People with depression often generate fewer episodic details when asked to think about upcoming events and show diminished activation of hippocampus and ventral striatum during prospection. From the perspective of a future sampler, this suggests a low-entropy, low-vigor generative process that produces shallow or repetitive trajectories, many of which are colored by themes of failure or futility. Because positive future paths are either not generated or fail to elicit anticipatory reward responses, long-term planning can seem pointless, leading to apathy and reduced goal-directed behavior.
This impoverished prospection also affects temporal discounting and persistence. When the hippocampus cannot construct vivid, scene-based representations of delayed rewards, those rewards may carry little psychological weight compared to immediate relief or inaction. Behavioral findings that detailed episodic simulation can reduce impulsive choice in healthy participants imply the converse: in depression, blunted simulation may increase the subjective steepness of discounting, making future benefits seem abstract or unattainable. Therapeutic approaches that train individuals to elaborately imagine specific, positive future episodes—sometimes called episodic future thinking interventions—likely work in part by reengaging hippocampal networks and restoring a more robust stream of motivating trajectories.
Schizophrenia and related psychotic disorders highlight how disruptions in the structure and reliability of hippocampal representations can destabilize prediction and planning. Anatomical and physiological studies consistently report abnormalities in hippocampal volume, excitability, and oscillatory dynamics in schizophrenia, particularly excess baseline activity and altered sharp-wave ripple patterns. If the hippocampus is chronically hyperactive yet poorly synchronized, its internal simulations may become noisy, disorganized, or insufficiently constrained by real-world transition statistics. Trajectories could jump erratically between loosely related states, blending past, present, and imagined events into unstable narratives.
Such disordered sampling may contribute to formal thought disorder, disorganized behavior, and aberrant salience. When the brain’s Bayesian machinery attempts to reconcile unpredictable internal sequences with external inputs, it may assign excessive significance to coincidental events or idiosyncratic associations, fostering delusional interpretations. Planning in everyday life becomes difficult because the internal model does not provide reliable, coherent forecasts of consequences. Treatments that normalize hippocampal excitability or sharpen the coordination between hippocampus and prefrontal cortex could, in principle, reduce the volatility of internal simulations, making prediction more stable and behavior more organized.
Addiction offers another perspective on how the coupling between hippocampal prospection and reward evaluation can be hijacked. In substance use disorders, drug-related cues and contexts acquire exaggerated motivational salience, and the hippocampus forms strong associative memories linking these cues to intense reward states. When future-oriented simulation is engaged—either spontaneously or in response to triggers—the system may preferentially generate trajectories that culminate in drug use or related rituals. Ventral striatal circuits, tuned by repeated dopaminergic surges, then assign these trajectories outsized expected value, while alternative paths (such as engaging in sober social activities) may be underrepresented or undervalued.
Over time, this biased sampling can narrow the effective choice set, making relapse more likely even when explicit intentions support abstinence. Planning becomes canalized around familiar reward-seeking scripts, and the capacity to imagine detailed, rewarding, drug-free futures is diminished. Behavioral therapies that promote elaboration of alternative life narratives, along with context modification and cue extinction, can be interpreted as attempts to diversify the hippocampal generative repertoire and weaken the dominance of drug-centered trajectories. Pharmacological interventions that modulate dopaminergic and glutamatergic signaling may complement this process by recalibrating how simulated futures are weighted in downstream valuation systems.
Obsessive–compulsive disorder (OCD) illustrates a different failure mode: overconstrained, repetitive sampling that resists updating by new evidence. Individuals with OCD often experience intrusive thoughts about contamination, harm, or incompleteness, accompanied by compulsive rituals aimed at preventing feared outcomes. Within a future-sampling framework, the hippocampus may repeatedly generate highly similar trajectories in which a specific catastrophe occurs unless a ritual is performed. Even when rituals have been completed and external feedback suggests safety, these particular trajectories continue to be sampled and treated as credible possibilities, in part because prefrontal–striatal circuits assign them disproportionate weight.
This loop can produce a pathological coupling between prospection and action: simulated futures dictate rigid behavioral responses, and successful outcomes (the catastrophe does not occur) are misattributed to the ritual rather than to the low base rate of the event, reinforcing the underlying model. Exposure and response prevention therapy can be reconceived as a targeted effort to break this coupling: by blocking the ritual and allowing the feared outcome to repeatedly fail to materialize, the generative model is gradually forced to update, reducing the frequency and credibility of catastrophic preplay sequences. Neurobiologically, successful treatment may depend on restoring flexible communication between hippocampus, orbitofrontal cortex, and striatum so that new evidence can reshape both sampling and evaluation.
Autism spectrum conditions raise additional questions about how differences in relational memory and predictive coding affect planning, particularly in social domains. Many autistic individuals report difficulty imagining fluid, open-ended social interactions, and may rely more heavily on explicit rules or routines. If hippocampal–cortical systems represent social situations as high-dimensional state spaces, then alterations in how social episodes are encoded and generalized could limit the range or variability of simulated social futures. Planning conversations, group activities, or career trajectories may thus feel effortful or uncertain, promoting a preference for structured environments where the space of possible futures is narrower and more predictable.
These clinical patterns suggest that planning capacity cannot be evaluated solely in terms of executive functions like working memory or inhibition; it also depends on the integrity and bias of hippocampal prospection. Two individuals with similar abstract problem-solving skills may differ dramatically in their real-world planning effectiveness because one routinely samples vivid, diverse, and goal-congruent futures, while the other’s simulations are sparse, negatively biased, or poorly aligned with environmental structure. Assessing and training the quality of future-oriented imagination could therefore become a key component of clinical evaluation and intervention across diagnoses, complementing traditional tests of cognitive control.
The sampling perspective also clarifies why some psychiatric symptoms cut across diagnostic boundaries. Intrusive memories and flashbacks in post-traumatic stress disorder, for example, can be understood as hyperpotent episodes that dominate the pool of trajectories available for simulation. When partial cues trigger these memories, the hippocampus not only replays the past but also preplays similar catastrophic futures, sustaining hyperarousal and avoidance. Likewise, anhedonia, seen in both depression and schizophrenia, may involve a failure to generate future trajectories that carry sufficient predicted reward to motivate action, regardless of the specific upstream cause. Rather than treating each symptom as disorder-specific, it becomes possible to map them onto generic distortions in how the future sampler constructs and weights possible outcomes.
From a methodological standpoint, this view encourages the development of experimental paradigms that directly probe hippocampal sampling during planning tasks, both in health and in psychiatric populations. For example, maze-like decision problems, spatial or conceptual graphs, and narrative-based scenarios can be combined with electrophysiology or high-resolution imaging to assay how many trajectories are generated, how far they extend, and how strongly they are biased toward certain goals or valences. Computational modeling can then infer latent parameters—such as sampling temperature, prior strength, or goal weighting—that characterize individual differences in prospection. These parameters may provide more mechanistically meaningful biomarkers than broad behavioral scores, aiding both diagnosis and treatment personalization.
Therapeutically, framing the hippocampus as a future sampler suggests interventions that target not only synaptic plasticity or neurotransmitters but also the statistics of experience that shape priors and memory. Repeated exposure to diverse, mastery-building experiences can expand the library of episodes available for recombination, enriching the space of adaptive futures that can be imagined. Cognitive therapies that explicitly practice constructing detailed, alternative scenarios may recalibrate sampling strategies, encouraging exploration of neglected but realistic options. Pharmacological agents that modulate hippocampal oscillations—such as those influencing GABAergic or cholinergic function—could be tested for their ability to restore flexible, goal-sensitive sequence generation, potentially amplifying the impact of psychotherapeutic techniques.
Across these domains, the central implication is that psychiatric disorders can be fruitfully conceptualized as disturbances in a unified system that links memory, prediction, and choice through hippocampal prospection. Planning failures are not simply deficits in rational calculation; they often reflect deeper alterations in the generative machinery that constructs the very futures over which decisions are made. By examining how, when, and what the hippocampus simulates—how it samples, constrains, and evaluates trajectories—it becomes possible to connect neural circuit dynamics to the lived experience of indecision, dread, compulsive action, or loss of motivation, and to design interventions that directly target the construction of the future rather than only its outcomes.
