Computational Psychiatry of Pareidolia & Pseudohallucinations
Reframing perceptual anomalies not as discrete psychological failures, but as mathematically formalizable dysregulations in predictive coding, precision weighting, and metacognitive reality monitoring.
What are Pareidolia and Pseudohallucinations?
Pareidolia is the everyday experience of seeing meaningful shapes, especially faces, in random or ambiguous stimuli, such as clouds, wood grain, or static on a screen. It is a normal brain tendency, but it becomes clinically interesting when it occurs far more often or feels far more convincing than usual.
Pseudohallucinations are vivid sensory-like experiences, such as hearing a voice or seeing a figure, that feel real but are recognized by the person as coming from inside their own mind rather than from the outside world. Unlike true hallucinations, insight and reality testing remain intact.
Both phenomena help us understand how the brain balances expectations against sensory evidence. When that balance shifts, internally generated patterns can feel surprisingly real.
Posterior belief calculated by multiplying likelihood (sensory input) by prior belief (expectations).
Precision represents inverse variance. Dictates whether sensory evidence or priors dominate perception.
High rate of pseudo-hallucinatory voices in Borderline Personality Disorder, driven by trauma priors.
Schizophrenia involves altered pictorial size scaling and spatial frequency filtering in visual tasks.
Understanding the Core Anomalies
Before examining the mathematical and neural mechanisms of Bayesian predictive processing, it is crucial to clearly define pareidolia and pseudohallucinations, two core perceptual phenomena that illustrate how internal mental models shape our subjective reality.
Pareidolia
Pattern DiscoveryDefinition: Pareidolia is the tendency to perceive meaningful patterns, most commonly human faces, bodies, or spoken words, within ambiguous, noisy, or random sensory stimuli (shapes in clouds, knots in wood, or static noise).
Computational Insight: It is not a structural brain deficit, but a consequence of hyper-precise, evolutionarily tuned top-down expectation priors (e.g., face detection templates) over-interpreting weak or degraded sensory likelihood signals.
Pseudohallucinations
Preserved InsightDefinition: Pseudohallucinations are vivid perceptual experiences that carry the clarity of actual sensory input, but occur either within inner subjective space (e.g., a voice heard inside the head) or with intact reality testing.
Computational Insight: Unlike genuine hallucinations where reality monitoring breaks down, pseudohallucinations occur when strong internal priors generate sensory-like content while higher-order prefrontal networks (CRMN) maintain intact metacognitive oversight.
Hierarchical Predictive Coding & Precision Weighting
In the active inference framework, the brain continuously generates top-down predictions (priors) to explain incoming sensory input. Discrepancies generate bottom-up prediction errors, which are propagated up the cortical hierarchy to update internal models. Precision weighting, the relative confidence assigned to priors versus sensory likelihood, acts as the master controller of sensory reality.
Precision as the master control
Perception is a weighted compromise. When sensory precision falls, or prior precision rises, the balance tips and internal expectation is experienced as perception.
Circular Inference & Deep Attractors
When hierarchical gating fails, top-down predictions and prediction errors re-enter ascending and descending loops, locking neural activity into deep attractor states that become immune to contradictory external evidence.
Figure 1: Ratio of prior precision to sensory precision across clinical conditions.
Neuromodulatory Substrates of Inference
The precision weighting of prediction errors is directly driven by specific neuromodulatory systems that alter post-synaptic pyramidal gain. Dysregulations in dopamine, acetylcholine, NMDA receptors, and serotonin yield distinct clinical perceptual profiles.
Modulates gain of sensory prediction errors. Drives representational sharpening. Deficits degrade sensory signal-to-noise ratio, forcing reliance on internal templates.
Signals unsigned prediction error (surprise/salience). Spontaneous dopamine surges assign excessive precision to environmental noise, triggering delusional beliefs.
Mediates synaptic plasticity and top-down belief updates. Hypofunction prevents prediction error resolution, causing prior beliefs to become rigidly fixed.
Modulates top-down expectations. Psychedelics disrupt backward alpha waves, triggering unconstrained forward prediction errors and vivid pseudohallucinations.
Figure 2: Differential impact of neuromodulators across precision parameters.
Pareidolia: Hyper-Precision of Evolutionary Priors
Pareidolia is the illusory perception of meaningful patterns (faces or voices) in ambiguous or random sensory noise. From an evolutionary perspective, false positive face detections carried minimal biological cost compared to missing a predator or ally. When sensory likelihood is broad, hyper-precise evolutionary social priors dominate perception.
Figure 3: Illusory detection rates on Scene and Visual Noise testing tasks.
Clinical High Risk (CHR) Dynamics
CHR individuals consistently report more faces in static noise and speech in white noise. This cross-modal propensity correlates directly with clinician-rated positive symptom severity, serving as an early computational marker.
Dementia with Lewy Bodies (DLB)
Severe pareidolia combined with visuoperceptual impairments is a key diagnostic indicator for prodromal DLB, driven by neurodegenerative disruption of the ventral visual stream.
HADD vs. PPAD Models
Unlike the Hyperactive Agency Detection Device (HADD) model which posits modular hard-wiring, the Predictive Processing of Agency Detection (PPAD) model frames pareidolia as a domain-general Bayesian outcome.
Pseudohallucinations & Metacognitive Reality Monitoring
Unlike genuine hallucinations, pseudohallucinations are experienced either within inner subjective space (e.g., voices in the head) or maintain metacognitive insight (reality testing). This distinction relies on the Cognitive Reality Monitoring Network (CRMN) in the prefrontal cortex.
CRMN failure + externalizing bias. Internally generated thoughts trigger unattenuated prediction errors that are projected into objective space, accompanied by lost reality testing.
Hyper-precise internal prior, but higher-order CRMN remains intact. Perception is vivid and involuntary, but recognized as lacking external sensory corroboration.
Failure of corollary discharge for internal speech creates unexpected internal prediction error, leading to metacognitive attribution of agency to an external entity.
Figure 4: Perceptual states mapped by spatial projection and preserved insight.
Transdiagnostic Spectrum Matrix
Predictive processing provides a unified, transdiagnostic vocabulary across psychiatric and neurological conditions, mapping how distinct sensory and metacognitive disruptions produce specialized visual and auditory anomalies.
| Clinical Condition | Primary Computational Dysfunction | Origin of Anomaly | Metacognitive Status |
|---|---|---|---|
| Schizophrenia | Hyper-precise priors; dopamine-driven aberrant precision weighting. | Multi-level cortical hierarchy breakdown. | Absent (Genuine) |
| Borderline Personality (BPD) | Over-reliance on internal top-down trauma priors under stress. | Affective priors dominating weak sensory signals. | Preserved / Partial |
| OCD ("False Psychosis") | Hyper-vigilant self-monitoring inflating precision of thoughts. | Compulsive checking & NMDA synaptic consolidation. | Fully Intact |
| PTSD Flashbacks | Rigid, un-updatable threat priors overpowering sensory likelihood. | Trauma priors + dissociation suppressing sensory precision. | Variable / Transient |
| Charles Bonnet Syndrome | Loss of sensory likelihood precision (peripheral deafferentation). | Visual loss unleashing unconstrained top-down predictions. | Fully Intact (Pseudo) |
| Anton's Syndrome | Central cortical blindness + damaged reality monitoring network. | Occipital infarcts + prefrontal CRMN destruction. | Absent (Anosognosia) |
Unstructured prediction errors escaping lower visual cortical suppression.
Dysregulation in distance-to-size scaling priors and spatial frequency integration.
Failure in fusiform face area priors causing dynamic facial melting illusions.
Persistence of visual traces when temporal prediction and sensory decay are mismatched.
Artificial Intelligence & Biological Inevitability
Confabulations and hallucinations are often regarded as system errors. However, computational psychiatry reveals that hallucination is an inevitable feature of any generative system operating under uncertainty.
Active Inference Engine
Fills sensory gaps using social priors and top-down predictions. Corrects internal errors using continuous environmental feedback, multisensory calibration, and metacognitive reality monitoring.
Ungrounded Top-Down Sampler
Predicts tokens via learned statistical probability distributions. Lacking physical embodiment and real-time sensory prediction error loops, prompt ambiguity forces reliance on priors, generating ungrounded confabulations.
Synthesized from predictive coding and hierarchical Bayesian perception research.