Functional neurological disorder (FND) has historically been defined more by what it is not—namely, not explained by structural lesions or classic neurological disease—than by what it is. Predictive processing offers a way of reversing that situation by providing a positive, mechanistic account of how functional symptoms can arise from otherwise intact neural systems. In this framework, the brain is understood as a prediction engine that is constantly generating hypotheses about the causes of sensory inputs and then updating those hypotheses based on the difference between its predictions and what is actually sensed. Motor commands, bodily sensations, and even experiences of agency are all seen as emerging from this continuous process of prediction and error correction.
Within this account, functional symptoms are not arbitrary or “imagined”; they are the product of maladaptive predictions that become overly dominant in shaping perception and action. When prior expectations about movement, sensation, or bodily integrity are assigned excessive precision relative to incoming sensory signals, the system can come to “believe” in a symptom even in the absence of corresponding pathology. For example, if a strong prior expectation of leg weakness is held with high confidence, and if the system down-weights or misinterprets contradictory sensory evidence, the resulting motor output may be consistent with true weakness, despite intact corticospinal pathways. This reframes FND as a disorder of inference rather than a failure of will or an act of simulation.
Predictive processing also integrates cognitive, affective, and sensorimotor aspects of FND into a single explanatory scheme. Instead of treating motor symptoms, sensory disturbances, dissociative experiences, and affective states as separate phenomena, this framework views them as different expressions of the same core inferential process operating across multiple hierarchical levels of the nervous system. Higher-level predictions about the body, self, and environment constrain and shape lower-level predictions about specific movements or sensations, and vice versa. Maladaptive priors at higher levels—for example, expectations of being physically vulnerable, damaged, or out of control—can cascade down to influence specific symptom patterns, such as non-epileptic seizures or functional weakness.
A central notion in this framework is precision weighting, which describes how the brain determines the relative influence of prior expectations versus new sensory data in forming perceptual and motor inferences. Precision can be thought of as a context-sensitive estimate of reliability or confidence. In FND, certain priors related to symptoms may be given overly high precision, while prediction errors signaling mismatch between those priors and actual bodily states are given insufficient precision. This imbalance allows symptom-related predictions to persist and dominate experience, even when they are not supported by the current state of the body. In effect, the brain treats these priors as more trustworthy than the sensory input that could correct them.
Attention plays a crucial role in modulating precision. Within predictive processing, attention is not simply an added cognitive operation but a mechanism that increases the precision of selected prediction errors. Directing attention toward particular body parts or sensations can selectively amplify the influence of certain inputs in the inferential process. In FND, heightened self-focused attention, especially in contexts of anxiety or threat, may disproportionately increase the precision of symptom-relevant signals (or even noise) and of symptom-related priors. This can reinforce maladaptive inferences, making functional symptoms more salient, more convincing, and more resistant to change once established.
The framework naturally accommodates the influence of interoception—the brain’s representation and prediction of internal bodily states—on functional symptoms. Interoceptive signals from the cardiovascular, respiratory, gastrointestinal, and other autonomic systems are constantly being predicted and explained away by higher-level models of bodily state. If interoceptive predictions related to arousal, pain, or bodily threat are chronically biased, and if these predictions receive high precision, the resulting experiences can include palpitations, breathlessness, dizziness, or pain that do not correspond to peripheral pathology. These interoceptive inferences can, in turn, alter motor and sensory predictions, helping to explain how emotional states and bodily sensations become tightly coupled with functional motor phenomena such as tremor or gait disturbance.
By focusing on generative models—internal models the brain uses to simulate and predict the consequences of actions and environmental events—predictive processing offers a principled way to understand the sense of agency and its disruption in FND. In healthy movement, the brain generates predictions about the sensory consequences of intended actions and then compares them to actual feedback. A close match underwrites the feeling of voluntary control. In functional motor symptoms, distorted priors about movement or bodily control can lead to a mismatch between what is predicted at higher levels (for example, “my leg cannot move”) and what the lower motor system is capable of generating. The motor system may implement the “cannot move” prediction, and because the mismatch is effectively minimized by enacting the symptom, the experience is one of genuine involuntariness, even though the motor apparatus is structurally intact.
This same logic extends to functional sensory symptoms, such as visual loss, numbness, or altered sensation, in which high-level priors about absent or distorted sensation dominate over intact peripheral and central sensory pathways. The generative model predicts that a region is anesthetic or that vision is absent or blurred, and the system suppresses or reinterprets incoming sensory evidence that conflicts with that model. From the perspective of predictive processing, these symptoms are not failures of basic perception but successful implementations of top-down predictions that happen to be maladaptive in real-world contexts and incompatible with the person’s goals.
The unifying power of predictive processing lies in its ability to link psychosocial factors, such as life experiences, beliefs about illness, and cultural narratives, to concrete symptom expression via changes in priors and precision. Chronic experiences of threat, trauma, or illness can shape high-level beliefs about the body and self, which then become embedded as strong priors within the generative model. Illness beliefs derived from family, media, or healthcare encounters can similarly provide templates for symptom patterns. These prior structures bias the inference process, making certain symptom configurations more likely to be “selected” when the system attempts to explain ambiguous bodily or environmental signals, especially under stress or uncertainty.
At the same time, this framework is compatible with known neurobiology and mechanisms of brain function. Hierarchical predictive coding is thought to be implemented through recurrent cortical and subcortical networks in which higher areas send predictions to lower areas and receive prediction errors in return. Neuromodulatory systems, including those involving dopamine, serotonin, and noradrenaline, are believed to regulate precision weighting across these hierarchies. This allows the model to connect phenomenological aspects of FND—such as fluctuating symptoms, stress sensitivity, or response to certain medications and psychotherapies—to plausible neurobiological substrates without reducing the disorder to a simple chemical imbalance or structural lesion.
Because the same inferential principles apply across perception, action, emotion, and cognition, predictive processing brings together previously disparate accounts of FND—ranging from psychodynamic, cognitive-behavioral, and neurophysiological theories—into a common language. Concepts such as conversion, dissociation, learned associations, reinforcement, suggestion, and attentional bias can all be reinterpreted as processes that shape the priors and precision structure of the generative model. This allows heterogeneous findings from imaging, neurophysiology, neuropsychology, and clinical observation to be integrated rather than left as competing explanations.
Crucially, the framework offers a way of understanding variability within and between individuals with FND. Different symptom profiles can be conceptualized as different configurations of priors, prediction errors, and precision across the sensory, motor, and interoceptive systems. For one person, strongly weighted priors about abnormal movement may dominate, producing tremor or dystonia; for another, priors about loss of consciousness or seizure-like activity may be central; for yet another, priors related to pain, fatigue, or sensory loss may be most prominent. All of these patterns can be explained using the same underlying inferential architecture, without requiring entirely separate disease entities.
This approach provides a coherent narrative that can be communicated to patients and clinicians without resorting to dualistic or stigmatizing distinctions between “organic” and “psychogenic.” By framing FND as a disorder of prediction and inference in the brain—where symptoms are real, embodied experiences produced by over-precise expectations and misweighted sensory evidence—predictive processing bridges the gap between subjective experience and objective neurobiological function. It offers a conceptual scaffolding on which more specific clinical, therapeutic, and research strategies can be built, while maintaining a unified view of the disorder across its many presentations.
Neurobiological mechanisms underlying functional neurological symptoms
Understanding how functional neurological symptoms emerge from brain function requires mapping the abstract ideas of predictive processing onto concrete neural circuits and physiological processes. Contemporary neuroimaging, neurophysiological, and lesion studies suggest that FND involves altered connectivity and communication within and between networks that implement prediction, error signaling, and precision weighting. Rather than showing frank structural damage, these networks often display changes in functional organization, especially in regions involved in motor control, salience detection, emotional processing, and the sense of agency.
A consistent finding concerns abnormal interactions between prefrontal, premotor, and parietal areas and primary motor or sensory cortices. In functional motor symptoms, such as weakness or tremor, studies frequently show relatively preserved activity in primary motor cortex alongside altered activation and connectivity in higher-order regions like the supplementary motor area, dorsal lateral prefrontal cortex, anterior cingulate cortex, and inferior parietal lobule. These higher regions are central to generating motor predictions, monitoring action outcomes, and attributing agency. The pattern suggests that top-down motor plans and expectations about movement are modified or constrained at intermediate levels, shaping the final output in a way that matches symptom-related predictions without disrupting the basic capacity of motor neurons to fire.
The supplementary motor area and adjacent medial frontal regions appear to play a particularly important role. These regions are strongly implicated in voluntary action initiation, selection of movement sequences, and the internal generation of motor plans. In FND, altered activity and connectivity here may correspond to aberrant motor priors—internal models predicting that a limb will not move or that a particular movement will be abnormal. Because these priors are instantiated at a relatively high level of the motor hierarchy, they can influence the downstream execution of movement while still allowing reflex pathways and basic motor excitability to remain intact, consistent with clinical signs such as Hoover’s sign in functional leg weakness.
Parietal regions, especially the inferior parietal lobule and temporoparietal junction, are central to integrating sensory feedback with motor predictions and constructing the feeling of agency over action. Neuroimaging in FND often demonstrates atypical activation in these areas during both symptomatic and non-symptomatic movements. One interpretation in terms of neurobiology and mechanisms is that parietal regions are receiving a pattern of sensory and motor prediction errors that has been shaped by over-precise priors about symptom states, leading to a mismatch between what the system “expects” to be voluntary and what is actually being generated. The resulting disruption in agency is experienced as movements or lack of movement that feel involuntary, even when initiated by internal processes.
Networks involved in salience detection and emotional processing—particularly the anterior insula, anterior cingulate cortex, and amygdala—also show altered function in FND. The anterior insula, a key hub for interoception, integrates internal bodily signals with cognitive and affective context, contributing to judgments about bodily relevance and threat. In individuals with functional symptoms, heightened insular activation and abnormal connectivity with limbic and frontal regions suggest that internal bodily states and symptom-related cues are assigned exaggerated salience. This aligns with the predictive processing view that symptom-related priors and prediction errors are given excessive precision, such that ambiguous bodily fluctuations are more likely to be interpreted in line with symptom templates.
The anterior cingulate cortex, another key region, participates in conflict monitoring, error detection, and the adjustment of control in response to unexpected outcomes. In FND, atypical cingulate activity may reflect persistent attempts to resolve discrepancies between maladaptive symptom expectations and incoming sensory evidence. However, because the precision of symptom priors is high, the system may repeatedly favor top-down explanations that confirm the presence of symptoms, rather than updating in light of disconfirming information. This can produce a state of chronic “over-explaining away” of healthy sensory inputs that would otherwise correct the model.
Functional connectivity studies also implicate large-scale networks that support self-referential thinking and environmental engagement, particularly the default mode network (DMN) and the salience and executive control networks. The DMN, centered on medial prefrontal cortex and posterior cingulate cortex, is closely associated with autobiographical memory, self-related processing, and long-term beliefs about the self and body. In FND, altered DMN connectivity may reflect entrenched, symptom-related self-models (for example, “my body is fragile,” “my movement is unreliable”) that shape how new experiences are interpreted. Coupling between the DMN and salience network, anchored in the insula and anterior cingulate, can further bias attention toward symptom-relevant information, reinforcing maladaptive priors.
At a more granular level, sensorimotor circuits demonstrate a mixture of preserved excitability and altered modulation. Transcranial magnetic stimulation and other neurophysiological tools often reveal intact corticospinal pathways but changes in intracortical inhibition, facilitation, and preparatory activity. For instance, reduced readiness potentials or altered patterns of pre-movement cortical activation have been reported in functional weakness and tremor, suggesting that the usual buildup of motor preparation is disrupted or replaced by alternative patterns consistent with symptom-related predictions. This selective alteration of preparatory mechanisms, rather than outright failure of execution machinery, parallels the idea that high-level priors are interrupting or reshaping the normal flow of motor planning.
Cortico-striato-thalamo-cortical loops are another key substrate. These loops mediate action selection, habit formation, and the integration of motivational and sensorimotor information. Aberrant functioning here could allow symptom-related action patterns—such as non-epileptic seizures or abnormal postures—to become “default” responses in certain contexts, even in the absence of conscious intention. Dopaminergic modulation of these loops, which affects how prediction errors influence learning and habit, may contribute to the consolidation and persistence of functional symptoms once they have been established through repeated episodes and reinforcement.
Neuromodulatory systems more broadly are likely to influence the precision weighting of priors and prediction errors across cortical hierarchies. Dopamine, for example, plays a central role in signaling the salience and expected value of actions and outcomes, and is closely tied to how strongly prediction errors drive learning. Noradrenaline and serotonin are implicated in arousal, stress responses, and the regulation of uncertainty. Dysregulated activity in these systems, whether due to chronic stress, early life adversity, or genetic vulnerability, could bias the brain toward over-precise threat-related or symptom-related expectations and under-precise sensory feedback. This provides a plausible pathway from psychosocial factors to altered inferential dynamics without requiring structural damage.
Interoceptive pathways connecting viscera, brainstem nuclei, thalamus, and insular cortex further ground functional symptoms in bodily physiology. Cardiorespiratory fluctuations, gastrointestinal sensations, and autonomic changes are constantly feeding into central representations of bodily state. In many people with FND, there is evidence of autonomic dysregulation—such as heightened sympathetic activity, variable heart rate dynamics, or abnormal blood pressure responses—that can generate noisy or exaggerated internal signals. When these signals are interpreted by a generative model already biased toward beliefs about illness or fragility, they can become the raw material for symptom construction, particularly when coupled with heightened attention to bodily sensations.
Importantly, these neurobiological profiles are not static. Functional imaging and longitudinal studies suggest that symptom severity and context can dynamically alter network configurations. For instance, during symptomatic episodes or when individuals attend to affected body parts, there may be increased connectivity between limbic regions, insula, and motor or sensory cortices, reflecting the temporary dominance of symptom-related priors. In contrast, when attention is redirected externally or during tasks that emphasize automatic movement, connectivity patterns may shift toward more typical configurations, mirroring clinical observations that symptoms often improve with distraction or when focus is on goal-directed behavior rather than on the symptom itself.
The neurobiology and mechanisms of functional sensory symptoms parallel those of motor manifestations but emphasize different nodes. In functional visual loss, altered activation in visual association cortices and attentional control regions can occur despite preserved early visual responses. In functional numbness or anesthesia, somatosensory cortex remains responsive at a basic level, yet higher-order integration areas and networks involved in body representation show atypical activity and connectivity. These findings support the idea that perception in FND is constrained by top-down expectations that certain inputs are absent or irrelevant, leading to active suppression or reinterpretation of incoming sensory signals.
Studies of dissociative phenomena and alterations in consciousness further highlight the involvement of thalamocortical and midline structures. The thalamus serves as a key relay and gating station for sensory information and plays a central role in regulating states of consciousness. In non-epileptic seizures and related episodes, patterns of thalamic and brainstem activity suggest shifts in arousal and sensory gating that can transiently decouple incoming signals from higher-level processing. Rather than being driven by epileptic discharges, these episodes may reflect network-level transitions orchestrated by expectation, affective state, and autonomic feedback, operating within a brain that is biased toward symptom-related models of collapse or loss of control.
Genetic and developmental factors likely modulate these networks by influencing brain maturation, stress responsivity, and the shaping of priors across childhood and adolescence. Variations in genes affecting synaptic plasticity, neuromodulatory tone, and stress hormone regulation can make some individuals more susceptible to forming rigid, threat-related beliefs about the body or to experiencing bodily signals as overwhelming or incomprehensible. Early experiences of pain, illness, or interpersonal threat provide the experiential data from which generative models of the self and body are constructed. Over time, these models become encoded in the connectivity patterns and responsiveness of large-scale networks, such that later stressors can more readily evoke functional symptoms as the brain falls back on familiar explanatory templates.
Together, these converging strands of evidence indicate that FND is rooted in complex, dynamic alterations of brain function, not in feigning or simple “psychological” overlay on a healthy nervous system. The key change is not gross damage but a reconfiguration of how prediction, error signaling, and precision are implemented across motor, sensory, interoceptive, and affective systems. These neurobiological mechanisms make it possible for symptom-related expectations to become entrenched and self-fulfilling, while preserving the underlying capacity of neural tissue to function normally under different contextual and inferential regimes.
Interactions between prior beliefs, attention, and sensorimotor processing
The dynamic interplay between prior beliefs, attention, and sensorimotor processing is central to understanding how functional symptoms are generated and maintained. Within a predictive processing framework, prior beliefs—implemented as probabilistic expectations at multiple hierarchical levels—constrain how sensory inputs and motor commands are interpreted. These priors include explicit beliefs (“my leg is weak,” “I am prone to seizures”) and more implicit, embodied assumptions about bodily vulnerability, control, and coherence. When such expectations are strong and assigned high precision, they shape the interpretation of ambiguous bodily signals and the selection of motor responses in ways that can bring symptoms into being.
Attention acts as the primary mechanism by which the brain modulates the precision of prediction errors. By selectively amplifying some sensory channels or internal signals and suppressing others, attention effectively determines which data are allowed to challenge existing priors and which are ignored or down-weighted. In FND, self-focused, symptom-oriented attention tends to heighten the influence of symptom-congruent information. Minor fluctuations in muscle tension, posture, vision, or interoceptive sensations are more likely to be noticed, scrutinized, and interpreted through the lens of prior beliefs about illness or dysfunction, making it easier for the generative model to “explain” these inputs as evidence of weakness, seizures, or sensory loss.
This feedback loop can be illustrated in functional motor symptoms. Suppose an individual carries an entrenched expectation that their hand is unreliable or prone to tremor. Under stress or heightened self-monitoring, attention gravitates toward the hand, increasing the precision of both sensory signals from the limb and prediction errors related to its position and movement. The generative model, biased by the prior of instability, preferentially interprets any small discrepancy or noise as confirmation of abnormal movement. Motor commands are then shaped to minimize prediction error in favor of the “abnormal movement” hypothesis—producing a tremor that is experienced as involuntary. The symptom thus emerges from an active process of prediction error minimization, not from deliberate action or conscious fabrication.
In functional weakness, a similar mechanism applies but with a different content of prior belief. A strong prior that “this leg cannot move” biases the motor system toward solutions that satisfy that expectation. When the individual attempts to move, top-down predictions impose a suppressed or incomplete motor output, consistent with the belief in weakness. Attention directed to the leg accentuates the perceived discrepancy between intended and actual movement, reinforcing the sense of failure and the conviction that the limb is weak. Paradoxically, when attention is drawn away—during automatic tasks or distraction—precision shifts away from the symptom-related priors, allowing more typical motor patterns to emerge, which aligns with clinical observations of improved movement in less self-focused contexts.
Sensory symptoms such as numbness, visual loss, or distorted perception constitute another domain where priors and attention interact. If the generative model includes an expectation that sensation from a body region is absent or abnormal, then attention to that region increases the precision of prediction errors that are interpreted through this template. Incoming sensory signals that do not fit the “numbness” or “blindness” expectation are selectively ignored, suppressed, or reinterpreted. Attention thus supports the construction of a phenomenological reality in which the region feels genuinely anesthetic or the visual field appears absent. When attention shifts elsewhere, the influence of these priors may lessen, sometimes allowing more veridical perception to re-emerge, at least transiently.
Interoception plays an important intermediary role in these processes. Internal bodily sensations related to heart rate, breathing, gut activity, or muscle tension are inherently noisy and ambiguous. Prior beliefs about bodily threat or fragility can bias the interpretation of this interoceptive noise. Heightened attention to internal states increases the precision assigned to interoceptive prediction errors, making them more influential in updating or confirming symptom-related beliefs. For example, in someone who expects that stress will trigger a seizure or collapse, normal fluctuations in heart rate or dizziness can be rapidly interpreted as prodromal signs of an episode. The generative model then predicts imminent loss of control or consciousness, leading to coordinated changes in posture, muscle tone, and awareness that fulfill the expectation of a seizure-like event.
The interaction between attention and priors is not unidirectional: prior beliefs also shape where attention is deployed. Generative models include predictions not just about sensory outcomes and motor states, but also about which sources of information will be most relevant or informative in a given context. In FND, high-level symptom-related priors bias attentional policies toward constant monitoring of the body for signs of dysfunction. This hypervigilant stance ensures that bodily sensations and small anomalies are repeatedly sampled and scrutinized, increasing the likelihood that they will be assimilated into the symptom narrative. Over time, such attentional policies can become habitual, further entrenching symptom-focused processing.
Social and contextual cues modulate this entire system. Interactions with clinicians, family members, and media can instantiate or strengthen priors concerning illness, prognosis, and appropriate symptom forms. When authority figures suggest that certain movements are dangerous, or when cultural scripts provide vivid templates for seizures, paralysis, or sensory loss, these narratives can become part of the generative model. Attention then orientates toward cues that resonate with these narratives—such as subtle bodily sensations during stress or images of others with similar symptoms—enhancing their precision and making it more probable that the system will resolve ambiguity in ways that conform to the learned symptom patterns.
Importantly, prior beliefs involved in FND need not be accessible to conscious reflection. Many operate at intermediate hierarchical levels, encoding probabilistic regularities about how the body typically behaves under specific emotional or environmental conditions. For instance, a person with a history of collapsing in response to overwhelming distress may implicitly encode a strong mapping between intense emotion and bodily shut-down. In future emotionally charged situations, attention to both internal affective states and bodily cues can activate this mapping, generating an expectation of collapse. The sensorimotor system then orchestrates a loss of postural tone or apparent unresponsiveness that resolves the discrepancy between the predicted and current state, experienced as an involuntary episode.
The sense of agency itself is shaped by the alignment between high-level intentions, lower-level motor predictions, and incoming sensory feedback. When symptoms arise from high-precision priors that are not aligned with explicit goals (“I want to walk normally”), the resulting actions or failures to act can feel alien or uncontrollable. Attention exacerbates this misalignment when it is continually directed toward the discrepancy between intended and actual movement, reinforcing the impression that something external or inexplicable is driving the symptom. Conversely, interventions that redirect attention toward functional goals (such as balance or reaching a target) and away from symptom monitoring can help re-establish coherence between intentions, predictions, and feedback, strengthening the experience of agency.
Another critical interaction occurs between expectations and the temporal dynamics of sensorimotor processing. Predictive models generate not only what should be perceived or enacted, but when. In FND, distorted timing expectations—for example, regarding when a movement should initiate or a symptom should emerge—can interact with attention to produce anticipatory adjustments that look like failure or loss of control. A person who expects tremor during a public performance may, under intense self-focused attention, unconsciously alter muscle tone or coordination in anticipation of that event. The early, subtle instability produced by this anticipation is then detected by the hypervigilant system, rapidly taken as confirmation of the prior, and amplified into a full-blown tremor.
The role of uncertainty is also crucial. States of heightened uncertainty—about health, diagnosis, or safety—tend to drive both an increase in prior strength and a narrowing of attentional focus. The brain attempts to resolve uncertainty by leaning more heavily on existing beliefs and sampling more data from perceived sources of threat or relevance, often the body itself. In individuals with FND, this dynamic creates fertile ground for symptom consolidation: strong, sometimes catastrophic priors about illness or dysfunction combine with intense bodily monitoring, making it more likely that ordinary variability will be interpreted as confirmation of the feared state and enacted through motor and sensory channels.
Therapeutically, this intricate coupling between priors, attention, and sensorimotor processing suggests that changing any one of these elements can influence the others. Adjusting prior beliefs through psychoeducation, cognitive restructuring, or experiential learning changes how future sensory information is interpreted. Training attention away from symptoms and toward external tasks, goals, or broader environmental cues alters precision weighting in favor of more adaptive inputs. Engaging in graded sensorimotor retraining provides new evidence that contradicts entrenched symptom-related expectations, encouraging the generative model to update. Although these strategies are addressed more fully in clinical discussions, they all operate by reshaping the ongoing dialogue between what the brain expects, what it attends to, and how it enacts and perceives movement and sensation.
Clinical implications for diagnosis and treatment of fnd
Applying this framework in clinical practice begins with how FND is conceptualized and communicated to patients. Reframing symptoms as arising from altered prediction and precision weighting, rather than from structural damage or feigning, helps clinicians offer a positive diagnosis grounded in neurobiology and mechanisms. Explaining that the brain operates according to predictive processing principles—constantly generating expectations about movement and sensation and then acting to minimize mismatches—allows patients to understand their symptoms as real but potentially reversible products of learned brain processes. This narrative can validate the patient’s experience while emphasizing that the nervous system remains structurally intact and therefore amenable to change.
Diagnosis itself can be guided by this framework. Rather than relying on exclusion, clinicians can focus on internal inconsistencies and context-dependent variability that indicate altered inference rather than fixed deficit. Signs such as Hoover’s sign, distractibility of tremor, entrainment phenomena, and symptom improvement with dual-tasking can be described to the patient as evidence that the motor or sensory system is capable of normal function when symptom-related expectations are not dominating. Demonstrating these signs in real time and linking them to concepts of attention, priors, and precision (for example, “when your focus is elsewhere, your brain stops predicting weakness so strongly, and your leg can move more normally”) transforms bedside examination into a therapeutic intervention.
The initial diagnostic conversation is a key therapeutic moment. Clinicians can explicitly distinguish between structural damage and functional miscommunication within brain networks, using analogies such as “software” versus “hardware” problems while stressing that functional problems are just as real and can be as disabling. Anchoring the explanation in concrete brain processes—altered connectivity between regions that generate predictions and those that carry out movement or sensation—helps reduce stigma and counter assumptions that the symptoms are “all in the mind” in a dismissive sense. Emphasizing that symptoms are involuntary outputs of the prediction system, shaped by attention and past experience, reassures patients that they are not to blame while also orienting them toward active participation in treatment.
Psychoeducation can be deepened over time. Visual aids, metaphors, and experiential demonstrations are often more effective than abstract descriptions. For instance, showing how optical illusions or sensorimotor illusions arise from prior expectations overriding sensory input can help patients grasp how similar processes operate in their own symptoms. Clinicians might illustrate how focusing intently on a body part changes its perceived sensation or heaviness, underscoring the role of attention in modulating precision. These experiences provide concrete evidence that perception and action are constructed rather than passively received, making it easier for patients to accept the possibility that their nervous system can “learn” different ways of responding.
Interdisciplinary rehabilitation can be explicitly designed to target the components of the predictive model: priors, precision, and prediction errors. In physiotherapy and occupational therapy, the emphasis shifts from strengthening a “weak” limb to retraining movement predictions and reducing the precision of symptom-related priors. Therapists can structure tasks to maximize experiences of successful, symptom-free movement under conditions of reduced self-focused attention—for example, walking while performing a cognitive task, rhythmically synchronizing movements to music, or reaching for objects in a game-like context. These tasks supply strong, surprising prediction errors (“I expected my leg not to move, but it did, smoothly”) in a safe environment, gradually weakening entrenched expectations of failure.
Graded exposure to feared or symptom-linked movements is another core principle. Under predictive processing, avoidance keeps maladaptive priors intact by limiting opportunities for disconfirmation. Systematically reintroducing movements or contexts that are associated with seizures, weakness, or pain—while carefully managing arousal and providing corrective experiences—encourages the brain to revise its expectations. For example, a person who anticipates collapse when standing in crowded places might begin with brief, supported standing in less threatening environments while practicing attentional strategies that orient toward external tasks rather than internal monitoring. Over time, increased exposure with successful outcomes reshapes the generative model, decreasing the predicted probability of collapse in those contexts.
Attention-focused interventions play a central role across therapies. Many individuals with FND habitually monitor their bodies for signs of dysfunction, inadvertently increasing the precision of symptom-related signals. Training patients to shift attention outward—to environmental cues, task goals, or sensory modalities unrelated to the symptom—can diminish the dominance of symptom priors. Techniques include dual-task exercises in physiotherapy, mindfulness practices that cultivate a broad, non-judgmental awareness rather than narrow hypervigilance, and structured behavioral experiments that demonstrate how symptom intensity changes with shifts in attentional focus. Clinicians can explicitly label these techniques as methods for recalibrating the brain’s precision settings.
Cognitive and psychotherapeutic approaches can be reframed as tools for altering high-level priors and their emotional context. Cognitive-behavioral strategies can identify and modify beliefs such as “once my symptoms start, they will inevitably get worse” or “any unusual sensation means permanent damage,” which function as catastrophic expectations that amplify symptom-related prediction errors. Therapists can collaborate with patients to test these beliefs through planned experiments, gathering data that contradicts rigid priors and encourages more flexible, probabilistic interpretations. In this view, cognitive restructuring is not simply about “positive thinking” but about systematically changing the statistical assumptions that the generative model uses to interpret bodily events.
Trauma-focused and emotion-focused therapies can be integrated into this model by recognizing that experiences of threat, helplessness, or violation often shape long-standing bodily priors. Past episodes of collapse, pain, or loss of control can become templates that the brain reuses in later stressful situations. Addressing these experiences—through modalities such as EMDR, trauma-focused CBT, psychodynamic therapy, or somatic therapies—can reduce the emotional charge and perceived inevitability of these bodily responses. As traumatic memory networks become less dominant, the generative model becomes less biased toward expectation of bodily catastrophe, thereby reducing the likelihood that stress will be resolved through functional symptoms.
Interoceptive interventions are particularly relevant for patients whose symptoms are tightly linked to autonomic arousal, panic, or bodily discomfort. Techniques aimed at recalibrating interoception—such as paced breathing, heart rate variability biofeedback, or interoceptive exposure—offer opportunities to experience internal sensations without catastrophic consequences. Patients can practice noticing changes in heart rate, breathing, or dizziness while maintaining posture or continuing a task, learning that these signals do not require collapse or seizure-like episodes to resolve. Over time, repeated non-threatening exposures reduce the precision of priors that equate bodily arousal with imminent loss of control, and interoceptive signals can be integrated more flexibly into the generative model.
In some cases, pharmacological treatments may support these processes by influencing neuromodulatory systems that regulate precision. Medications that reduce anxiety, stabilize mood, or modulate arousal can indirectly decrease the salience of threat-related signals and symptom-related cues, making it easier for psychological and rehabilitative interventions to gain traction. For example, selective serotonin reuptake inhibitors or noradrenergic agents may help regulate hyperarousal and reduce the tendency for attention to lock onto bodily sensations. It is essential, however, that pharmacotherapy be framed as a tool for optimizing the learning environment of the brain’s prediction system, rather than as a cure that passively removes symptoms.
Communication within the multidisciplinary team is facilitated by a shared predictive processing language. Neurologists, psychiatrists, psychologists, physiotherapists, occupational therapists, and speech therapists can align around the goal of reshaping priors and precision across different domains. Neurologists emphasize positive diagnostic signs and brain-based explanations; psychologists focus on high-level beliefs, trauma, and emotional context; physiotherapists engineer sensorimotor prediction errors and attentional shifts; occupational therapists target everyday contexts and roles where new patterns must generalize. When all team members consistently frame interventions in terms of learning new brain patterns rather than “proving” that symptoms are psychological, patients receive a coherent message that promotes engagement and reduces confusion.
Specific symptom clusters can inform tailored treatment strategies. In functional movement disorders, therapy may prioritize motor retraining, rhythmic entrainment, and tasks that highlight preserved automaticity of movement (such as walking to music or performing movements as part of complex actions rather than in isolation). These interventions limit self-monitoring and enhance prediction errors that favor normal movement patterns. In non-epileptic seizures, the emphasis may be on recognizing early interoceptive cues, implementing competing motor or attentional responses, and altering the meanings attached to those cues through cognitive and trauma-focused work. In functional sensory loss, graded sensory re-exposure with concurrent distraction, mirror feedback, or visual-tactile incongruence can demonstrate that intact sensory pathways can be accessed when symptom priors are relaxed.
Behavioral planning and pacing can also be conceptualized within this framework. Sudden, dramatic attempts to abandon all symptom-related accommodations may produce overwhelming prediction errors that the system cannot integrate, leading to reinforcement of failure expectations. Instead, structured, incremental changes in activity levels and roles allow the generative model to update gradually, matching expectations to achievable successes. Collaborative goal-setting—defining concrete, graded steps toward valued activities—helps align high-level intentional states (“I want to return to work,” “I want to play with my children”) with sensorimotor experience, strengthening coherent agency and counteracting priors centered on incapacity.
The clinical relationship itself is part of the therapeutic context in which priors are formed and updated. Consistent, non-stigmatizing explanations; validation of distress and disability; and transparency about uncertainty all contribute to building trust in the clinician and, by extension, in the therapeutic model being offered. If patients perceive dismissiveness or inconsistency, this can strengthen priors that their symptoms are incomprehensible or intractable, undermining engagement. Conversely, when clinicians convey confidence that symptoms are understandable in terms of brain function and that change is possible through active learning, they help instantiate new, hopeful expectations that can guide the brain’s inferential processes in a more adaptive direction.
Emerging digital and technological tools can further operationalize these principles. Virtual reality environments, for example, can manipulate sensory feedback and attentional focus in ways that produce striking prediction errors, demonstrating to patients that their movements or sensations can normalize under altered contextual constraints. Wearable devices that track activity, physiological signals, or episodes can provide objective data that challenge catastrophic priors (“I am always about to collapse”) and support more nuanced interpretations. When integrated into a predictive processing-based treatment plan, such technologies can extend therapeutic learning into everyday life, reinforcing new patterns of expectation and response outside the clinic.
Clinical practice informed by this framework emphasizes that treatment is not about uncovering a single hidden psychological cause but about systematically reshaping how the brain predicts, attends, and responds across multiple levels. Diagnosis becomes an opportunity to demonstrate the flexibility of the system; psychoeducation offers a new explanatory model; therapy and rehabilitation deliver repeated, structured experiences that contradict maladaptive priors while supporting new ones. Through this process, functional symptoms can diminish as the nervous system learns to generate predictions that are better aligned with the person’s goals, capabilities, and current bodily state.
Future directions in research on predictive processing and fnd
Future research will need to translate the conceptual strengths of predictive processing into precise, testable models of functional neurological disorder. One priority is to formalize computational accounts that specify how priors, prediction errors, and precision weighting are configured in different FND phenotypes. Hierarchical Bayesian models, active inference simulations, and generative models fitted to behavioral and physiological data can be used to estimate individual “prediction profiles” in motor, sensory, and interoceptive domains. Such computational phenotyping would permit more rigorous hypotheses about which aspects of the inference process—overly strong symptom priors, underweighted sensory input, aberrant attentional gain, or altered learning rates—are most disturbed in a given person, and how these disturbances map onto specific symptom patterns.
Developing these models will require experimental paradigms that manipulate expectation and attention while measuring behavioral performance, brain activity, and bodily responses. Tasks that systematically vary uncertainty, cue participants to expect particular movements or sensations, or introduce controlled mismatches between prediction and feedback can be applied to people with FND and comparison groups. For example, motor adaptation paradigms, sensorimotor conflict tasks, or illusions that rely on top-down prediction (such as rubber hand or visuomotor distortion illusions) can be adapted to probe how readily participants update priors in response to surprising evidence. Concurrent neuroimaging and electrophysiology can then reveal the neurobiology and mechanisms with which brain networks implement or fail to implement these updates.
Longitudinal and state-dependent designs will be particularly important. Many existing studies compare patients with FND to healthy controls at a single time point, yet predictive processing emphasizes that symptoms arise from dynamic inference processes that may fluctuate across context and over illness course. Future work should track individuals before, during, and after symptomatic episodes; across different attentional contexts; and throughout treatment. Repeated measures of network connectivity, neuromodulatory tone, and interoceptive and motor behavior could clarify how symptom-related priors strengthen or weaken over time, how precision is redistributed across hierarchies during episodes, and which changes predict clinical improvement or relapse.
Another major direction involves refining our knowledge of large-scale network interactions in FND. Resting-state and task-based imaging studies have highlighted roles for the salience network, default mode network, and executive control networks, but a more granular understanding is needed. Future research should examine how these networks coordinate during the generation, maintenance, and resolution of functional symptoms, and how they support the deployment of attention toward or away from the body. Methods such as dynamic causal modeling, time-varying connectivity analyses, and network control theory can be harnessed to test explicit hypotheses about the directionality of influence between regions that encode priors (for example, medial prefrontal cortex), those that signal prediction errors (for example, posterior insula, parietal cortex), and those that adjust precision (for example, anterior cingulate, neuromodulatory brainstem nuclei).
Interoception warrants targeted investigation, both as a contributor to symptom formation and as a potential treatment target. Future studies can integrate fine-grained measures of autonomic function—heart rate variability, baroreflex sensitivity, skin conductance, respiratory patterns—with tasks that manipulate internal expectations and bodily focus. For instance, paradigms that require participants to predict or track their own heartbeat, breathing, or visceral sensations under varying emotional or cognitive loads can reveal how interoceptive prediction and precision weighting are altered in FND. Combining these tasks with insula- and cingulate-focused imaging, as well as perturbation techniques such as vagus nerve stimulation or biofeedback interventions, may help clarify causal pathways between interoceptive inference and symptom expression.
Neuromodulatory systems represent another promising frontier. Future work can explore how dopamine, noradrenaline, serotonin, and acetylcholine shape precision weighting and learning in FND. Pharmacological challenge studies and molecular imaging can be used to test whether abnormalities in these systems track with symptom severity, stress sensitivity, or responsiveness to psychological treatment. Beyond correlational associations, experiments that temporarily alter neuromodulator function—using established medications or neurostimulation techniques—can assess whether modifying neuromodulatory tone improves the flexibility of priors and the capacity to update symptom expectations in response to new evidence. Such studies could inform targeted pharmacological adjuncts that are explicitly designed to augment learning within predictive processing–based therapies rather than to suppress symptoms nonspecifically.
Experimental therapeutics provide a bridge between mechanism and clinical practice. Future interventional studies should be built around mechanistic predictions derived from predictive processing: for example, that therapies which systematically generate strong, surprising, and tolerable prediction errors will be more effective than those that provide only verbal reassurance; or that strategies emphasizing external, goal-directed attention will outperform those that inadvertently encourage symptom monitoring. Randomized trials can compare treatments that make different demands on priors, attention, and sensorimotor engagement, while embedding biomarkers and computational measures that evaluate how the brain’s predictive architecture changes over the course of therapy. Such designs can move the field beyond efficacy alone toward an understanding of why and for whom interventions work.
Digital and immersive technologies can be harnessed as experimental platforms for manipulating prediction and feedback with high precision. Virtual and augmented reality permit controlled distortions of visual, proprioceptive, and vestibular inputs, enabling researchers to probe the boundaries of symptom-related priors and to craft experiences that challenge them safely. For example, virtual environments that conceal or alter visual feedback from affected limbs can test how strongly movement expectations constrain actual motor output. Real-time motion capture, haptic interfaces, and neurofeedback systems can provide continuous streams of data for building and refining computational models, while also serving as tools for graded sensorimotor retraining that can be readily customized to individual inference profiles.
At the clinical and psychosocial level, future research needs to delineate how life experiences, cultural narratives, and interpersonal dynamics shape the generative models that predispose to FND. Prospective and developmental studies tracking individuals from childhood into adulthood could examine how early pain, illness, trauma, and family beliefs about health influence the formation of bodily priors and attentional habits. Qualitative and mixed-methods work can characterize the specific illness narratives and expectation patterns that precede symptom onset, while experimental social neuroscience paradigms can investigate how suggestion, observation of others’ symptoms, and clinician communication adjust precision weighting. Such research would tighten the link between social context and neurobiology, showing how culturally and interpersonally mediated expectations become instantiated in neural networks.
Personalized medicine represents an overarching aspiration. If future studies can reliably characterize individual differences in predictive patterns—such as the relative contribution of interoceptive versus exteroceptive priors, the degree of attentional bias toward the body, or the rigidity of motor predictions—therapies can be tailored accordingly. One patient may benefit most from interoceptive exposure and autonomic regulation; another from intensive motor retraining combined with external focus techniques; another from trauma-focused work to soften high-level threat models. Adaptive trial designs and N-of-1 frameworks, combined with computational monitoring of belief updating and learning, can accelerate the development of such stratified approaches and help identify early markers of treatment response or nonresponse.
Integration with adjacent fields is also a key future direction. Predictive processing models of chronic pain, fatigue, dissociation, and anxiety share many features with accounts of FND, and many patients experience overlapping symptom constellations. Cross-diagnostic research can investigate whether there are shared alterations in inference and precision across these conditions, and whether interventions that target common mechanisms (such as catastrophizing priors or hypervigilant attention) produce transdiagnostic benefits. At the same time, careful comparative work can identify features that are relatively specific to FND, such as distinctive patterns of motor control or agency disruption, preventing overgeneralization of theoretical models.
Another avenue lies in clarifying boundaries and interactions between FND and classical neurological disorders. Many individuals with epilepsy, movement disorders, or multiple sclerosis also develop functional symptoms. Future work can explore whether the presence of structural pathology modifies generative models in ways that make functional symptoms more likely, for instance by providing powerful templates for symptom forms or by altering baseline sensory noise and uncertainty. Multimodal imaging and neurophysiological studies in such “mixed” presentations could reveal how organic and functional mechanisms interact within a single predictive architecture, challenging simplistic dichotomies and informing integrated treatment strategies.
Methodologically, advancing the field will require larger, multisite cohorts, harmonized protocols, and open science practices. Because FND encompasses heterogeneous presentations and is often under-recognized, individual studies have frequently been small and underpowered. Collaborative consortia can facilitate standardized symptom phenotyping, collection of shared imaging and physiological datasets, and pooling of expertise in computational modeling. Open data and pre-registered analytic pipelines will help resolve discrepancies between studies and accelerate consensus on which network alterations and inferential disturbances are robust. Including patient partners in the design and interpretation of research can ensure that tasks and interventions map onto lived experience and clinically meaningful outcomes.
Future work should attend not only to symptom reduction but also to broader outcomes such as quality of life, role functioning, and sense of agency. Predictive processing emphasizes that the brain’s generative models are oriented toward maintaining coherence of the self in relation to the world; meaningful life changes may be as important as specific symptom shifts in reshaping these models. Research that links mechanistic markers—such as changes in attention deployment, interoceptive accuracy, or network connectivity—to improvements in participation, autonomy, and identity will help ground the theory in outcomes that matter to patients. In doing so, the field can move toward a mature science of FND that unites computational rigor, nuanced neurobiology and mechanisms, and clinically relevant change within a single predictive processing framework.
