The perception of reality absent physical stimuli—hallucinations—exemplifies how our brains construct rather than merely capture the world around us. Recent computational neuroscience approaches have illuminated the mechanisms behind these false percepts, particularly through predictive coding and Bayesian inference frameworks. These models suggest hallucinations emerge when the delicate balance between our brain’s expectations and actual sensory input becomes disrupted. This report examines how various computational mechanisms contribute to false inferences that manifest as hallucinatory experiences, with evidence supporting both perceptual dysfunction and belief-processing aberrations.
Predictive Coding and the Precision Imbalance
The predictive coding theory proposes that perception operates as an active process of hypothesis testing rather than passive sensory reception. In this framework, the brain continually generates predictions about incoming sensory data and updates these predictions based on prediction errors—the mismatch between expectations and actual sensory input. Hallucinations emerge when this delicate balance shifts too heavily toward prior expectations at the expense of sensory evidence.
Multiple empirical studies demonstrate that hallucination-prone individuals exhibit stronger reliance on perceptual priors during ambiguous sensory tasks. These individuals show increased susceptibility to laboratory-induced hallucinations, suggesting that overweighting of prior beliefs relative to sensory evidence represents a transdiagnostic mechanism underlying hallucinatory experiences. One particularly illuminating experimental paradigm called the “Conditioned Hallucinations” task revealed that participants who experience hallucinations in daily life were more likely to report hearing sounds that weren’t actually presented during the experiment34. These findings remained consistent across both clinical and non-clinical populations.
At the heart of this imbalance lies a key computational mechanism called precision weighting—the brain’s assignment of confidence levels to both predictions and prediction errors. Precision can be conceptualized as the inverse of uncertainty; highly precise signals are weighted more heavily in perceptual inference. Hallucinations may result from either overly precise prior beliefs or reduced precision of sensory evidence that contradicts expectations35.
Sensory Precision Deficits
Recent research provides compelling evidence for the role of reduced sensory precision in hallucination formation. Participants with recent hallucinatory experiences as well as those with higher hallucination-proneness demonstrated higher stimulus thresholds, lower sensitivity to stimuli presented at the highest threshold, and lower response confidence—all consistent with reduced precision of sensory evidence410. This finding suggests that both reduced sensory precision and increased prior weighting are independently related to hallucination severity.
As one study explains, “a down-weighting of the precision of sensations (i.e., silence) that contradict the expected percept (i.e., a voice)” can lead to false perceptions5. This precision imbalance causes the brain to favor its predictions over contradictory sensory information, potentially resulting in the perception of stimuli that aren’t objectively present. Importantly, this mechanism helps explain why hallucinations often occur in noisy or ambiguous environments where sensory signals are inherently less precise.
Active Inference and Generation of False Percepts
The active inference model provides a more comprehensive computational framework for understanding hallucinations by incorporating the role of action selection in perception. Under active inference, agents not only form predictions about sensory input but also actively sample their environment to gather evidence for their beliefs about the world5. This perspective treats perception and action as inseparable aspects of the same process—minimizing surprise by making the world conform to expectations.
Computer simulations based on this framework demonstrate that hallucinatory percepts can emerge when an agent expects to hear a voice in the presence of imprecise sensory data5. When applied to auditory verbal hallucinations (AVH), this model suggests that a false inference that a voice is present, despite the absence of corresponding auditory input, indicates the domination of prior beliefs over perceptual inference5.
The active inference account particularly addresses the interactive nature of many hallucinations. For example, people who experience auditory hallucinations often engage in dialog with their voices. This interactive quality emerges naturally from the active inference framework, as the agent’s actions (such as listening or speaking) influence their perceptual inferences about hidden states of the environment5. Crucially, these actions are driven by the same predictive process that generates perceptions.
Through mathematical modeling using Markov decision processes, researchers have formally expressed how hallucinations arise from the interaction between action and perception. In this formulation, the content of and confidence in prior beliefs depends on beliefs about policies (sequences of actions like listening and talking) and on beliefs about the reliability of sensory data5.
Different Types of Hallucinations Based on Belief Structures
Computational models have helped differentiate between distinct types of hallucinatory experiences based on the nature of the underlying belief disturbances. “In-context hallucinations” occur when individuals cannot use sensory information to correct prior beliefs about hearing a voice, but their beliefs about content (such as the sequential order of a sentence) remain accurate1112. In contrast, “out-of-context hallucinations” emerge when hallucinating subjects also have inaccurate beliefs about state transitions, leading to disordered hallucinated content reminiscent of the bizarre hallucinations sometimes observed in conditions like schizophrenia1112.
This distinction helps explain the spectrum of hallucinatory experiences—from those that seem plausible given the context to more bizarre manifestations disconnected from environmental contingencies. For example, a person experiencing in-context hallucinations might hear a voice saying contextually appropriate phrases, while someone with out-of-context hallucinations might perceive jumbled or semantically incoherent speech.
The computational mechanisms underlying these different manifestations involve varying degrees of precision imbalance at different levels of the perceptual hierarchy. Simulations show that subjects with inaccurate beliefs about state transitions but an intact ability to use sensory information do not hallucinate and resemble prodromal patients—individuals who experience attenuated psychotic symptoms before developing full psychosis1112. This suggests that the progression from prodromal states to frank psychosis may involve a gradual shift in the balance between sensory precision and prior beliefs.
Reconciling Contradictory Findings: The Hierarchical Processing Solution
A significant challenge in the computational understanding of hallucinations is the apparent contradiction in research findings: some studies implicate weakened prior beliefs in psychosis, while others find stronger priors in hallucinations389. This apparent disconnect becomes comprehensible when considering the hierarchical nature of perceptual processing.
Rather than having uniformly strong or weak priors throughout the perceptual system, individuals may have different precision imbalances at different levels of the perceptual hierarchy and across different sensory modalities39. For example, in the hierarchical processing of speech, weak priors at a lower level might fail to constrain sensory noise, while strong priors at a higher level might generate false perceptions based on expected patterns38.
When illusions are not perceived by patients with schizophrenia, it could be that they fail to attenuate sensory precision, enabling prediction errors to ascend the hierarchy to induce belief updating8. These un-attenuated prediction errors may induce a particular sort of high-level prior belief that becomes the hallucination8. Indeed, research suggests that psychotic individuals with hallucinations utilize different priors than those without hallucinations, even within the same task8. People with hallucinations have strong perceptual priors that are not present in psychotic patients who do not hallucinate, who instead may have weak priors8.
The dynamic interaction between lower-level sensory processing and higher-level beliefs creates a complex landscape in which hallucinations can emerge through multiple computational pathways. This perspective helps reconcile seemingly contradictory findings and underscores the importance of considering the full hierarchical context of perceptual inference.
Neurobiological Underpinnings of Computational Aberrations
The computational mechanisms described above have plausible neurobiological correlates. The main neurotransmitter alterations thought to underlie predictive coding abnormalities include hypofunction of cortical NMDA receptors, dysfunction of gamma-aminobutyric acidergic neurons, and elevated striatal dopamine D2 receptor activity3. These neurochemical changes affect the precision-weighting mechanisms that balance prior beliefs against sensory evidence.
Researchers theorize that maladaptive priors may be encoded in upper levels of the processing hierarchy with weighting modified by dopaminergic signaling, whereas lower-quality sensory evidence could result from reduced integrity of white matter connections such as the arcuate fasciculus or alterations in cholinergic tone4. These neurotransmitter systems provide biological mechanisms through which computational parameters like precision can be regulated in the brain.
In schizophrenia specifically, these computational alterations may relate to loss of synaptic gain control in superficial pyramidal cells, changes in the excitatory-inhibitory balance in sensory cortex, cortical gray matter loss, and disrupted corticothalamic connectivity3. These neurobiological changes align with the computational account of hallucinations as arising from precision imbalances in predictive processing.
Clinical Implications and Future Directions
Understanding hallucinations through the lens of computational mechanisms carries significant clinical implications. Research shows that the relationship between prior overweighting and hallucination propensity is not merely a static trait but rather a state-sensitive marker that fluctuates with symptom severity3. This dynamic relationship suggests that measuring changes in perceptual prior weighting could potentially serve as a biomarker for tracking hallucination susceptibility or treatment response.
Notably, patients with psychosis who do not experience hallucinations do not show the same pattern of prior overweighting, indicating specificity of this abnormality to hallucinations rather than psychotic illness more broadly3. This specificity further supports the centrality of subpersonal prior mechanisms in hallucination formation and suggests targeted interventions might be developed to address this specific computational deficit.
The predictive coding account of hallucinations offers a unifying framework that spans from normal perception to pathological states, emphasizing the continuum of perceptual experiences rather than categorical distinctions38. This perspective encourages a more empathic approach to clinical hallucinations by recognizing them as extreme manifestations of normal perceptual mechanisms rather than entirely alien phenomena.
Conclusion
The computational mechanisms leading to false inferences in hallucinations represent a compelling demonstration of how automatic brain processes operating below conscious awareness profoundly influence perceptual experience. When subpersonal priors become too strong or precise relative to sensory evidence, they can generate percepts detached from physical reality—hallucinations.
Several key computational mechanisms contribute to these false inferences: overly precise prior beliefs relative to sensory evidence; down-weighting of the precision of contradictory sensory information; aberrant encoding of precision due to neurobiological alterations; hierarchical imbalances between lower and higher-level processing; and distinct mechanisms for in-context versus out-of-context hallucinations.
As research in this area advances, improved understanding of these computational mechanisms may lead to novel interventions targeting the precision balance between prior beliefs and sensory evidence. By recognizing hallucinations as arising from fundamental perceptual inference processes rather than categorically distinct phenomena, we gain not only scientific insight but also a more nuanced and compassionate perspective on the hallucinatory experience—potentially opening new avenues for treating distressing hallucinations across various clinical conditions.