Computational Psychiatry of Delusions
Bayesian Inference, Predictive Coding & Hierarchical Psychopathology
The Brain as an Inferential Engine
Computational psychiatry conceptualizes delusions not as arbitrary cognitive failures, but as statistically optimal adaptations by an inferential engine attempting to resolve fundamentally altered parameters of uncertainty, precision, and prediction error. Under the Bayesian Brain Hypothesis and Predictive Coding Framework, the brain continuously generates top-down predictions to explain away bottom-up sensory prediction errors (PE).
Posteriors integrate top-down prior weights (ω1) and sensory likelihood weights (ω2).
Hyperdopaminergia elevates sensory precision (π), assigning immense importance to random noise.
Delusions emerge as high-level priors formulated specifically to "explain away" unpredicted signals.
Psychotic patients exhibit the "Jumping to Conclusions" bias, deciding after ≤ 2 draws of evidence.
The 3-Stage Pipeline of Delusion Formation
Delusion development follows a recognizable computational and phenomenological trajectory from initial prodromal ambiguity to incorrigible fixation.
Stage 1: Delusional Mood
Phenomenology: Wahnstimmung
A primary disruption in dopamine-mediated precision causes the brain to assign high precision (ω2) to mundane sensory data. Ambient background noise and minor social cues become pregnant with uninterpretable, hyper-salient meaning, placing the individual in an agonizing state of high Free Energy and unconstrained ambiguity.
Stage 2: Delusional Snap
Phenomenology: Aha-Erlebnis
To halt the relentless upward flow of prediction errors and minimize systemic Free Energy, the cognitive hierarchy formulates a novel, highly structured explanatory hypothesis (e.g., "I am under government surveillance"). This newly minted high-level prior immediately contextualizes the noisy signals, providing immense relief.
Stage 3: Fixation
Phenomenology: Incorrigibility
Because the new prior must explain an abnormally broad array of noisy inputs, it is assigned extraordinarily high weight (ω1 >> ω2). The belief becomes impervious to contradictory empirical evidence. Disconfirmatory evidence is systematically discounted, dismissed as irrelevancies, or actively integrated into the narrative.
Parameter Shift across Illness Stages
This chart compares relative weighting parameters across normal inference, the prodromal phase (Wahnstimmung), and delusional fixation. In the prodrome, likelihood weighting (ω2) spikes while prior stability drops. Upon delusional consolidation, prior weighting (ω1) reaches pathologically high levels while likelihood sensitivity drops.
Interactive Bayesian Belief & Free Energy Simulator
Adjust the computational parameters below to observe how the brain updates its posterior belief state and minimizes variational free energy over successive trials.
Higher values represent rigid, fixed top-down expectations.
Higher values represent high sensory precision / aberrant salience.
Discrepancy between environmental input and baseline model.
Resolving the "Delusion Paradox"
How can psychotic patients show weakened priors in visual processing while simultaneously maintaining rigid, overweighted priors in cognitive belief formation?
The delusion paradox is resolved through Hierarchical Segregation across intrinsic neural timescales (INT). In healthy brains, priors operate seamlessly across both rapid sensory processing and slow cognitive abstraction. In psychosis, decreased synaptic gain at middle cortical layers impairs low-level perceptual priors (making patients resistant to optical illusions like the Hollow Mask), while compensatory, hyper-rigid priors emerge at higher prefrontal layers to prevent catastrophic computational instability.
Operates on short Intrinsic Neural Timescales. Failure to attenuate bottom-up sensory noise results in aberrant salience and paradoxical resistance to visual illusions (e.g., perceiving the Hollow Mask accurately as concave).
Operates on long Intrinsic Neural Timescales in PFC. High-level priors become structurally rigid to absorb persistent prediction errors, resulting in fixed delusions and heightened susceptibility to conditioned hallucinations.
Probabilistic Reasoning: The "Jumping to Conclusions" Bias
Deconstructing reasoning aberrations in psychosis using the classic Beads Task paradigm (85:15 vs 60:40 ratios).
Draws-To-Decision (DTD) Distribution
In probabilistic reasoning tasks, subjects request beads from hidden jars before deciding which jar is selected. Between 40% and 70% of delusional patients exhibit the JTC bias, making a definitive decision after drawing 2 or fewer beads (DTD ≤ 2).
Mechanisms of JTC
Initial confirmatory evidence triggers a massive, disproportionate increase in likelihood weight (ω2).
Rapid discounting of preceding evidence leads to extreme belief instability prior to decision locking.
Impaired Subjective Value of Information (VOI) tracking in right DLPFC reduces dynamic sampling motivation.
Circuit & Agency Mechanics
Microcircuit loop reverberation and sensory attenuation failures driving positive symptoms and alien control.
The Circular Inference Model
Jardri & Denève (2013)
Normative cortical function relies on GABAergic inhibition to prevent recurrent excitatory signals from being overcounted. Breakdown of the Excitatory/Inhibitory (E/I) balance causes signals to reverberate infinitely across cortical hierarchies.
Active Inference & Delusions of Agency
Sensory Attenuation Deficits
Executing self-generated movement requires transiently suppressing (attenuating) incoming sensory feedback. In psychosis, sensory attenuation fails, leaving self-generated actions feeling externally caused.
Healthy subjects overestimate force applied to themselves due to sensory attenuation. Patients with schizophrenia match force accurately, failing to attenuate self-generated tactile feedback.
Neurobiological Substrates of Computational Parameters
Mapping formal mathematical parameters onto neurochemical pathways and cellular circuits.
A major strength of computational psychiatry is its capacity to bridge synaptic pathophysiology with subjective phenomenology. Modulations in neurotransmitter systems directly map onto distinct mathematical variables within the hierarchical predictive coding framework.
| Neuromodulator | Computational Parameter | Clinical Phenotype |
|---|---|---|
| Striatal Dopamine | Prediction Error Precision (π) | Aberrant Salience, Wahnstimmung |
| Insular Glutamate / NMDA | Prior Weighting (ω1) | Delusional Conviction, Conditioned Hallucinations |
| GABA Fast-Spiking Interneurons | E/I Microcircuit Balance | Circular Inference, Disorganization |
| Acetylcholine / Norepinephrine | Expected vs Unexpected Uncertainty | Environmental Context Tracking Failure |
Clinical Translation & Precision Psychiatry
By reframing delusions as parameter dysregulations, computational psychiatry paves the way for Computational Phenotyping. Rather than diagnosing broad, heterogeneous categories, clinicians can isolate specific numerical parameters (e.g., ω1 overweighting vs NMDA hypofunction) to tailor pharmacological interventions. Furthermore, therapies such as Metacognitive Training (MCT) and Cognitive Behavioral Therapy for psychosis (CBTp) can be mathematically optimized to retune precision parameters, helping patients tolerate uncertainty and prevent premature belief locking.