talk with our admissions team to get started.
Contact Us
Pand Health
Mechanistic Analysis of Psychosis & Cognition

The Algorithmic Mind: Computational Psychiatry of Cognitive Biases

Clinical psychiatry has traditionally relied on descriptive symptom clusters to categorize delusional ideation. Computational psychiatry bridges neurobiology and phenomenology by modeling the brain as a hierarchical Bayesian statistical engine. Delusions and belief inflexibility are deconstructed into quantifiable algorithmic parameters: altered sensory precision, learning rate asymmetries, and hierarchical prediction error failures.

Key frameworks: Hierarchical Gaussian Filter (HGF) • Drift-Diffusion Modeling (DDM) • Dual-Rate Reinforcement Learning

JTC Bias Phenotype
1–2 Beads

Average draws to decision (DTD) in active delusions versus 5–7 in healthy controls (Garety et al., 1991).

BADE Mechanics
π → ∞

Hyper-precise top-down priors in lateral OFC suppress bottom-up negative prediction errors (Petrovic & Sterzer, 2023).

Confirmation Bias Ratio
αC ≫ αD

Asymmetric reinforcement learning where confirmatory learning rate far exceeds the disconfirmatory rate (Lefebvre et al., 2017).

Mood Disorder Inversion
α⁻ > α⁺

Inverted learning asymmetry in depression driving anhedonia and negativity bias (Pike & Robinson, 2022).

1. Jumping to Conclusions (JTC) Bias

From hasty probabilistic heuristics to perceptual noise and aberrant salience in sequential sampling.

The jumping to conclusions bias is traditionally evaluated with the beads task, where participants sample beads from two hidden jars to determine the origin jar. Healthy controls gather five to seven beads before deciding; patients with delusional schizophrenia often commit after one or two. Computational modeling by Moutoussis et al. (2011) showed that JTC is not caused by overestimating the subjective cost of gathering information, but by heightened decision noise relative to the task rules.

Draws to Decision (DTD) Distribution

Mean data requests before final commitment across clinical groups.

Data synthesized from Garety et al. (1991) and Fine et al. (2007).

Drift-Diffusion Accumulation Trajectories

Simulated evidence accumulation reaching decision boundaries under aberrant salience.

Simulated DDM trajectories based on Stuke et al. (2018) and Hillier et al. (2025).

Aberrant salience and drift rates (v)

Applying drift-diffusion modeling to perceptual tasks such as random dot motion reveals that hallucination- and delusion-prone individuals show significantly higher drift rates (v) and reduced boundary separation (a). Aberrant dopaminergic firing imbues noisy sensory signals with hyper-precision, forcing evidence accumulation to rapidly bridge the collapsed decision boundary.

Resolving the delusion severity paradox

Baker et al. (2019) demonstrated a critical dissociation: while schizophrenia broadly is associated with hasty, un-incentivized decision-making tied to cognitive impairment, delusion severity specifically tracks a distinct inferential mechanism, the erratic, over-weighted updating of beliefs in response to noisy evidence.

2. Bias Against Disconfirmatory Evidence (BADE)

Resolving the delusion paradox through multi-level hierarchical predictive processing.

While JTC explains how false beliefs form rapidly, BADE explains why delusions persist for decades. In multi-stage narrative or visual interpretation tasks, as disambiguating evidence refutes an initial "lure" scenario, healthy controls drop their plausibility ratings. Patients with schizophrenia maintain high plausibility for refuted lures and show liberal acceptance of absurd scenarios (Woodward et al., 2008).

Multi-Stage Scenario Plausibility Ratings

Lure scenario devaluation across progressive disambiguation stages.

Adapted from the Woodward et al. (2008) interpretation inflexibility paradigm.

The Delusion Paradox Resolved

Petrovic & Sterzer (2023) hierarchical coding architecture

High cortical hierarchy (lOFC / prefrontal)Hyper-precise prior (π_high → ∞)

Compensatory creation of rigid, hyper-precise high-level priors (delusions) to impose order on a chaotic bottom-up sensory stream.

↑ Top-down suppression ↓ Bottom-up disconfirmation

Low cortical hierarchy (sensory cortices)Blunted negative PE (δ⁻ = 0)

NMDA interneuron dysfunction degrades negative prediction error signals. Incoming disconfirmatory data is assigned near-zero precision and explained away.

A delusion is not a simple learning failure, but a top-down defense mechanism shielding the brain from bottom-up sensory chaos.

3. Confirmation Bias & Asymmetric Reinforcement Learning

Quantifying dual learning rates (αC vs. αD) and fronto-striatal decoupling.

In standard Q-learning models, human learning is best captured by splitting the unified learning rate into asymmetric parameters for confirmatory outcomes (αC) and disconfirmatory outcomes (αD). Healthy individuals hold a mild confirmatory bias that optimizes reward harvesting against neural noise. In schizophrenia, fronto-striatal decoupling turns this into an exaggerated, rigid confirmation bias.

Asymmetric Learning Rates Across Profiles

Confirmatory (αC) versus disconfirmatory (αD) parameter weights.

Parameters modeled on Lefebvre et al. (2017) and Pike & Robinson (2022).

Trial-by-Trial Value Updates in IPST

Simulated belief retention under false initial instructions across 20 trials.

Model simulating the instructed probabilistic selection task (Doll et al., 2014; Frydecka et al., 2022).

Fronto-striatal mechanisms in the IPST taskParticipants are given false initial instructions about stimulus values. Healthy controls rely on accumulating striatal dopamine prediction errors to dynamically unlearn those instructions, while patients with schizophrenia remain computationally locked in due to near-zero disconfirmatory learning rates driven by fronto-striatal decoupling (Frydecka et al., 2022).

4. Neurocircuitry & Transdiagnostic Applications

Mapping molecular pathways to computational parameters and mood disorder inversions.

NMDA

Glutamatergic Hypofunction

NMDA receptor deficits on cortical interneurons selectively attenuate negative prediction errors (δ⁻), preventing neural encoding of disconfirmatory evidence and underpinning BADE.

DA

Striatal Dopamine Hyperfunction

Aberrant phasic dopamine release amplifies positive prediction errors (δ⁺), imbuing benign environmental events with hyper-salience and elevating drift rates (v).

5-HT

Serotonergic Modulation in MDD

In major depressive disorder, learning asymmetry is inverted (α⁻ > α⁺). Prolonged SSRI administration recalibrates serotonin levels, gradually resetting baseline expectations (Browning et al., 2022).

Transdiagnostic inversion: negativity bias in depression

Asymmetric learning rate parameters extend across diagnostic boundaries. In major depressive disorder, the typical optimistic confirmation bias is inverted into a severe, pathological negativity bias (Pike & Robinson, 2022). Depressed patients update expected environmental values rapidly following losses and punishments while assigning near-zero learning rates to positive surprises, mathematically formalizing clinical anhedonia and learned helplessness.

5. Metacognitive Resolution & Precision Translation

Decoupling subjective certainty from objective performance, and targeted therapeutic avenues.

Metacognitive Sensitivity Ratio (meta-d′ / d′)

Ability to distinguish correct from incorrect judgments.

Signal detection theory modeling based on Koller & Cannon (2021) and Hillier et al. (2025).

Metacognitive blindness and overconfidence in errors

Type 2 signal detection analyses show that individuals on the psychosis continuum have a profound deficit in metacognitive resolution (Hillier et al., 2025). Rather than global overconfidence, patients lack the internal sensitivity to differentiate their own correct and erroneous judgments.

Because confidence signals are decoupled from objective accuracy, subjective certainty cannot serve as an internal feedback signal to modulate learning rates, trapping the person in self-reinforcing delusion loops.

The metacognitive training (MCT) backdoor approach

Directly challenging delusional content triggers defensive BADE. MCT (Moritz & Woodward, 2005) uses neutral tasks, such as resolving blurred pictures or unfolding benign cartoons, to make patients explicitly aware of algorithmic flaws like JTC and BADE, engaging prefrontal executive control to override noisy automatic inference.

Precision psychiatry: computational stratification

Profile A: aberrant salience phenotype

Driven by high sensory noise and low decision boundaries (a). Responds optimally to dopaminergic blockade to lower drift rates (v) and restore boundary separation.

Profile B: hyper-precise prior phenotype

Driven by blunted negative prediction errors (δ⁻). Requires glutamatergic/NMDA modulation to restore disconfirmatory belief updating and lower high-level prior precision.

References

  • Adams, R. A., Stephan, K. E., Brown, H. R., Frith, C. D., & Friston, K. J. (2018). The computational anatomy of psychosis. Frontiers in Psychiatry, 4, 47.
  • Baker, S. C., Konova, A. B., Daw, N. D., & Horga, G. (2019). A distinct inferential mechanism for delusions in schizophrenia. Brain, 142(6), 1797-1812.
  • Doll, B. B., Waltz, J. A., Cockburn, J., Brown, J. K., Frank, M. J., & Gold, J. M. (2014). Reduced susceptibility to confirmation bias in schizophrenia. Cognitive, Affective, & Behavioral Neuroscience, 14(3), 715-728.
  • Fine, C., Gardner, M., Craigie, J., & Gold, I. (2007). Hopping, skipping or jumping to conclusions? Cognitive Neuropsychiatry, 12(1), 46-77.
  • Friston, K. J., & Brown, H. R. (2013). Predictive coding, active inference and the brain. International Review of Psychiatry, 25(6), 647-657.
  • Frydecka, D., Piotrowski, P., & Bielawski, T. (2022). Confirmation bias in the course of instructed reinforcement learning in schizophrenia-spectrum disorders. Brain Sciences, 12(1), 90.
  • Garety, P. A., Hemsley, D. R., & Wessely, S. (1991). Reasoning in deluded schizophrenic and paranoid patients. Journal of Nervous and Mental Disease, 179(4), 194-201.
  • Hillier, C., Weber, N., & Balzan, R. P. (2025). Overconfidence or resolution in psychosis: a Bayesian reanalysis. Cognitive Neuropsychiatry, 30(3), 186-198.
  • Koller, W. N., & Cannon, T. D. (2021). Paranoia is associated with impaired novelty detection and overconfidence in recognition memory judgments. Journal of Abnormal Psychology, 130(3), 273-285.
  • Lefebvre, G., Lebreton, M., Meyniel, F., Bourgeois-Gironde, S., & Palminteri, S. (2017). Behavioural and neural characterization of optimistic reinforcement learning. Nature Human Behaviour, 1(4), 0067.
  • Moritz, S., & Woodward, T. S. (2005). Jumping to conclusions in delusional and non-delusional schizophrenic patients. British Journal of Clinical Psychology, 44(2), 193-207.
  • Moutoussis, M., Bentall, R. P., El-Deredy, W., & Dayan, P. (2011). Bayesian modelling of jumping-to-conclusions bias in delusional patients. Cognitive Neuropsychiatry, 16(5), 422-447.
  • Petrovic, P., & Sterzer, P. (2023). Resolving the delusion paradox. Schizophrenia Bulletin, 49(6), 1425-1436.
  • Pike, A. C., & Robinson, O. J. (2022). Reinforcement learning in depression: A review of computational research. Psychological Medicine, 52(12), 2231-2244.
  • Stuke, H., Weilnhammer, V., Heinz, A., & Sterzer, P. (2018). Hallucination- and delusion-like experiences are associated with increased precision of sensory evidence in perceptual inference. Schizophrenia Bulletin, 44(4), 896-905.
  • Woodward, T. S., Moritz, S., Cuttler, C., & Whitman, J. C. (2008). A bias against disconfirmatory evidence is associated with delusions in schizophrenia. Psychiatry Research, 161(2), 213-221.

Keep reading

All research & library pages →

This page was medically reviewed by Eric Wexler M.D., Ph.D. on August 14, 2026.