The myth of the Bayesian brain
Open accessThe DOI matches its Crossref record; 199 of the 204 references it cites were found in a registry or an index, and 4 are not checked yet; 62 of the 183 sources its claims cite were checked for support.
Abstract
The Bayesian brain hypothesis-the idea that neural systems implement or approximate Bayesian inference-has become a dominant framework in cognitive neuroscience over the past two decades. While mathematically elegant and conceptually unifying, this paper argues that the hypothesis occupies an ambiguous territory between useful metaphor and testable, biologically plausible mechanistic explanation. We critically examine the key claims of the Bayesian brain hypothesis, highlighting issues of unfalsifiability, biological implausibility, and inconsistent empirical support. The framework's remarkable flexibility in accommodating diverse findings raises concerns about its explanatory power, as models can often be adjusted post hoc to fit virtually any data pattern. We contrast the Bayesian approach with alternative frameworks, including dynamic systems theory, ecological psychology, and embodied cognition, which conceptualize prediction and adaptive behavior without recourse to probabilistic inference. Despite its limitations, the Bayesian brain hypothesis persists-driven less by empirical grounding than by its mathematical elegance, metaphorical power, and institutional momentum.
Verified 2026-11-22 · 4 references not checked yet
Overview
CoChat Summary
TL;DR
This paper argues that the “Bayesian brain” is better understood as a powerful metaphor than as a biologically grounded mechanistic theory: it is flexible, hard to falsify, and often unsupported by neural implementation evidence. The author contrasts it with dynamic systems, ecological, and embodied approaches, and calls for post-Bayesian neuroscience grounded in biological constraints and explicit, risky predictions.
Key findings
- The paper claims the Bayesian brain framework is ambiguous between metaphor and mechanism, and that this slippage weakens its scientific status; the framework is described as “self-sealing” and potentially “unfalsifiable.”
- It argues that Bayesian models can be fit post hoc to “virtually any observed data pattern” by adjusting priors, likelihoods, noise, or hierarchy, making the theory too flexible to strongly constrain outcomes.
- The paper says empirical support is mixed: some findings are presented as Bayesian “successes” (e.g., multisensory integration, visual illusions, motor control), but these are often consistent with other explanations and frequently rely on retrofitted priors or unclear neural implementation.
- The paper identifies biological implausibilities, including the difficulty of encoding full probability distributions, the computational burden of real-time Bayesian updating, and mismatches with non-Gaussian, multistable, state-dependent neural dynamics.
Methods
- Conceptual and critical review of the Bayesian brain hypothesis, predictive coding, and the Free Energy Principle; the paper also contrasts them with dynamic systems theory, ecological psychology, and embodied/enactive cognition.
- Uses qualitative argumentation and literature-based critique rather than new experiments or original data; it discusses examples from perception, action, learning, and psychiatry, and proposes a hypothetical “crucial experiment.”
- Main analysis focuses on conceptual drift (metaphor vs. mechanism), falsifiability criteria, empirical mismatches, and neurobiological feasibility.
Limitations
- The paper itself notes that a full comparative evaluation of alternative frameworks is beyond its scope.
- It also says the proposed critical experiment and falsifiability criteria are conceptual, not a completed empirical test.
- Evident from the text: the argument is a narrative/literature-based critique without new data or systematic meta-analysis, so its conclusions depend on interpretation of prior studies rather than direct experimental adjudication.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures



Verification
Claims the paper makes
42 Valid · 3 Not valid · 66 Not checked · 135 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“when behavioral data fail to conform to the predictions of Bayesian optimality, researchers frequently appeal to auxiliary assumptions—such as “suboptimal” priors, cognitive capacity limitations, or the use of approximation algorithms—to account for these deviations (Rahnev and Denison 2018; Sanborn and Chater 2016).”
§ The myth of the Bayesian brain > Unfalsifiability and flexibility
What the sources say
- Partly supportedcites Rahnev and Denison 2018 matched to “Suboptimality in perceptual decision making” (2018) sourceExperimental — accuracy not yet measured
Section 5.1 says new findings are often modeled with flexibly chosen Bayesian components, and Section 3.8 describes priors invoked to explain perceptual biases. Section 4.2 also discusses computational capacities as a limit on optimality. The passages do not establish that researchers frequently invoke approximation algorithms specifically to explain deviations. (LLM, conf=0.84)
- Partly supportedcites Sanborn and Chater 2016 matched to “Bayesian Brains without Probabilities” (2016) sourceExperimental — accuracy not yet measured
In “Bayesian Brains without Probabilities” and “Unavoidable Limitations of Sampling” (page unavailable), the paper explains departures from ideal probability judgments through finite sampling and computational limits. It does not say that researchers frequently invoke auxiliary assumptions in general, or identify suboptimal priors as one of those explanations. (LLM, conf=0.83)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“If top-down predictions and bottom-up errors do not interact as predicted—e.g., if perturbations to higher-order areas do not affect lower-level error responses—this violates core architectural assumptions (Clark 2013).”
§ The myth of the Bayesian brain > Metaphor, mechanism, or myth: The conceptual drift of the Bayesian brain > Falsifiability criteria for the Bayesian brain hypothesis
What the sources say
- Supportedcites Clark 2013 matched to “Whatever next? Predictive brains, situated agents, and the future of cognitive science” (2013) source
Section 1.2 (p. 3) describes higher-level predictions as suppressing lower-level prediction errors, with errors then propagating upward. Section 3.1 (p. 12) likewise says backward signals from higher areas should reshape lower-area responses. The cited architecture therefore predicts the interaction described in the claim, although the paper does not discuss the specific perturbation example. (LLM, conf=0.88)
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“If Bayesian-based explanations of psychiatric disorders (e.g., schizophrenia as altered precision weighting) fail to predict symptom patterns, treatment outcomes, or neural correlates in out-of-sample datasets, the theory’s applied value diminishes (Fletcher and Frith 2009).”
§ The myth of the Bayesian brain > Metaphor, mechanism, or myth: The conceptual drift of the Bayesian brain > Falsifiability criteria for the Bayesian brain hypothesis
What the sources say
- Could not verifycites Fletcher and Frith 2009 matched to “Perceiving is believing: a Bayesian approach to explaining the positive symptoms of schizophrenia” (2009) sourceExperimental — accuracy not yet measured
The ‘Prediction error, learning and dopamine’ section says that future work must link experimental learning abnormalities to clinical delusions, and the Conclusions identify further questions for the model. Neither passage addresses whether failure to predict symptoms, treatment outcomes, or neural correlates in out-of-sample datasets diminishes the theory’s applied value; searches for those validation criteria found no such passage. (LLM, conf=0.70)
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Perhaps most problematically, the Bayesian framing has difficulty accounting for the context-sensitivity and task-dependence of neural responses. The same neural circuits may perform different functions depending on behavioral context, attentional state, or motivational factors—a flexibility that sits uncomfortably with strict Bayesian formulations (Gilbert and Li 2013).”
§ The myth of the Bayesian brain > Empirical and biological Shortcomings > Empirical mismatches
What the sources say
- Partly supportedcites Gilbert and Li 2013 matched to “Top-down influences on visual processing” (2013) sourceExperimental — accuracy not yet measured
The “Perceptual task” section (page unavailable) says that neuronal tuning changes with the task even when the stimulus is identical, and that neurons can serve different functions in different behavioral contexts. The cited passages do not say that Bayesian formulations have difficulty accounting for this flexibility; the paper contrasts it with classical feedforward accounts instead. (LLM, conf=0.86)
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In tightly controlled tasks where prior probabilities and likelihoods are explicitly defined, persistent deviations from Bayesian-optimal behavior—unexplainable by noise, heuristics, or approximation—would falsify normative claims (Rahnev and Denison 2018).”
§ The myth of the Bayesian brain > Metaphor, mechanism, or myth: The conceptual drift of the Bayesian brain > Falsifiability criteria for the Bayesian brain hypothesis
What the sources say
- Partly supportedcites Rahnev and Denison 2018 matched to “Suboptimality in perceptual decision making” (2018) sourceExperimental — accuracy not yet measured
Section 5.3 says the many findings of suboptimality make the assumption that the brain is built for optimal decisions suspect. Section 4.2 says deviations can challenge components of a specified model, but it does not say they would falsify normative Bayesian claims generally. Section 2.2 also distinguishes experimentally defined priors from likelihood functions derived under assumptions, so the claim overstates how fully the tasks define both. (LLM, conf=0.86)
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
159 more claims · 29 with their cited sources checked
Create a free account to see all 246 claims and to verify the 3 references still open.
- Probably not a substantial claim. 48
- It may state two separate claims as one. 43
Citations
Cites
Items 201–203 of 203
Friston, K. J. (2016) · Biological Psychiatry · doi:10.1016/j.biopsych.2016.05.006 · Cited by 18 · cited 1×
“Predictive coding accounts propose that symptoms such as hallucinations and delusions arise from aberrant precision-weighting of prediction errors—either attributing too much weight to low-level sensory signals or too little to top-down priors (Adams et al. 2013; Corlett et al. 2019; Friston 2016).”
Bayesian cognitive science, predictive brains, and the nativism debate
Colombo, M. (2018) · Synthese · doi:10.1007/s11229-017-1427-7 · Cited by 15 · cited 2×
“Other critiques have highlighted the difficulty of empirically constraining priors and loss functions in perceptual models (Colombo 2018), and the problematic metaphysical overreach of the Free Energy Principle in its attempts to explain life, cognition, and action through a single variational…”
Doya, K., & Ishii, S. (2006) · Bayesian Brain · doi:10.7551/mitpress/1535.003.0004
Research works printed but not linked: 6
- Dennett D (1987) The Intentional Stance. MIT Press, Cambridge, MA
- Doya K (2007) Bayesian Brain: Probabilistic Approaches to Neural Coding. MIT Press, Cambridge, MA
- Lakoff G, Johnson M (1980) Metaphors We Live By. University of Chicago Press, Chicago, IL
- MacKay DJ (2003) Information Theory. Cambridge University Press, Cambridge, UK, Inference and Learning Algorithms
- Michaels CF, Carello C (1981) Direct Perception. Prentice-Hall, Englewood Cliffs, NJ
- Noë A (2004) Action in Perception. MIT Press, Cambridge, MA
Cited by
The illusion of internal models in biological movement
Mangalam, M. (2025) · European Journal of Applied Physiology · doi:10.1007/s00421-025-05963-3 · Cited by 13
“In a recent deconstruction of the Bayesian brain hypothesis (Mangalam 2025), I argue that Bayesian logic, despite its formal elegance, often functions as a rhetorical backdoor for representational commitments.”
Self-citationThe equilibrium point hypothesis revisited: why threshold control does not explain human movement
Mangalam, M., & Stergiou, N. (2026) · Experimental Brain Research · doi:10.1007/s00221-026-07242-9 · Cited by 7
“Opti - mization-based theories and related computational frameworks that assume explicit prediction or model-based control carry their own computational and implementational challenges (Loeb 2012; Mangalam 2025a, b, c), and we do not advocate for their wholesale adoption.”
Self-citationPathophysiology of functional neurological disorder for the general neurologist
Phillips, W. E. (2025) · BMJ Neurology Open · doi:10.1136/bmjno-2025-001309 · Cited by 6
“15 However, the theory itself is not without critics 16 and it is difficult to provide direct proof for any mechanism.”
Minimization of prediction errors during cerebral embryogenesis and the emergence of agency
Wright, J. J., & Bourke, P. D. (2025) · Frontiers in Systems Neuroscience · doi:10.3389/fnsys.2025.1683448 · Cited by 1
“Recent criticisms (Kwisthout and van Rooij, 2020; Mangalam, 2025) assert that the principle is metaphoric and not testable; that the computation of Bayesian priors and posteriors could not be achieved within hierarchical architectures; that intermediate values of these measures could not be stored;…”
Mangalam, M. (2026) · Neuroscience & Biobehavioral Reviews · doi:10.1016/j.neubiorev.2026.106724 · Cited by 2
Self-citation
Metadata
- Authors
- Mangalam, M.
- Affiliations
- University of Nebraska at Omaha (US)
- Corresponding author
- Mangalam, Madhur
- ORCID
- Mangalam, Madhur
- Venue
- European Journal of Applied Physiology
- Published
- 2025-06-26
- Volume
- 125
- Issue
- 10
- Pages
- 2643-2677
- ISSN
- 0020-9376 (ISSN-L)
- 1439-6319
- 1439-6327
- PMC
- PMC12479598
- OpenAlex
- W4411669101
- Cited by
- 14
- References
- 204
- Citations by year
- 2026: 10 · 2025: 4
- FWCI
- 2.33
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- European Journal of Applied Physiology (publishedVersion, cc-by)
- Europe PMC (PubMed Central) (publishedVersion, cc-by)
- PubMed (publishedVersion)
- PubMed Central (publishedVersion, cc-by)
- DigitalCommons@UNO (submittedVersion, cc-by)
- Topics
- Life Sciences › Neuroscience › Cognitive Neuroscience › Embodied and Extended Cognition
- Physical Sciences › Computer Science › Artificial Intelligence › Bayesian Modeling and Causal Inference
- Life Sciences › Neuroscience › Cognitive Neuroscience › Cognitive Science and Education Research
- Keywords
- Bayesian brain hypothesis, Bayesian inference, unfalsifiability, biological plausibility, prediction, dynamical systems theory, ecological psychology, embodied cognition, probabilistic inference, explanatory power
- MeSH terms
- Animals, Bayes Theorem, Brain, Cognition, Humans, Models, Neurological
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.