02 / Theoretical neuroscience · Peer-reviewed research

Published 2025Two companion reviewsOpen-access record

Conscious
active inference.

I co-authored two theoretical reviews with Michael C. Wiest in 2025. They examine quantum dynamics as a possible implementation of planning and perceptual cycles in active inference.

Read Paper I ↗Read Paper II ↗

Role: Co-author
Publication: Computational and Structural Biotechnology Journal
Papers: Two companion reviews · 2025
Framework: Active inference × quantum dynamics × Orch OR

Argument arc objection → perturbation → mechanistic hypothesis → discriminating experiment

Formal levelWhat must be computed?

Active inference specifies belief updating and policy selection under variational free-energy minimization.

Mechanistic candidateWhat could implement it?

The papers evaluate quantum dynamics and Orch OR as candidate physical implementations, not prerequisites of active inference itself.

Standard of proofProposed experimental tests

A quantum account must outperform viable classical alternatives on discriminating, reproducible intervention predictions.

00 / From process to mechanism

Computational objectives, neural processes, and physical mechanisms

The project begins by separating three levels of explanation. Active inference supplies a normative objective. Process theories map belief updating onto neuronal populations and message passing. Mechanistic models must still explain how real biophysics realizes those computations.

Scientific basis: Ma et al. 2006; Friston et al. 2006, 2017; Friston 2018; Parr, Pezzulo & Friston 2022. [3] [4] [5] [6] [7]

Conceptual diagram

Normative

Active inference

What beliefs and policies minimize variational free energy?

Computational objective

Process

Neural message passing

How could populations encode states, prediction errors, policies, and precision?

Algorithmic mapping

Mechanistic

Biophysics

What physical dynamics actually perform the required computation?

Physical implementation

Active inference does not require quantum mechanics at the normative level. The research question begins at the mechanistic level: what physical dynamics could implement temporally deep inference?

01 / The decoherence challenge

Decoherence estimates and the neural timescale

Tegmark calculated extremely short decoherence times for candidate neural quantum states, roughly 10⁻¹³ to 10⁻²⁰ seconds, compared with neural dynamics on roughly 10⁻³ to 10⁻¹ second scales. The calculation became a central physical objection to quantum models of cognition. [1]

Model comparison · logarithmic time axis

Tegmark / 200010⁻²⁰–10⁻¹³ s

Modeled decoherence for the candidate neural superpositions he analyzed.

Hagan et al. / 2002≈10⁻⁵–10⁻⁴ s

Recalculation for a different proposed microtubule superposition. Longer values require further model assumptions.

Neural / cognitive≈10⁻³–10⁻¹ s

Reference range used to frame the biological timing problem.

The numerical gap depends on the physical model being evaluated. Tegmark and Hagan et al. modeled different candidate superpositions and reached very different estimates. These are calculations, not direct measurements of coherence in living neurons. [1] [2]

Tegmark / 2000

One physical model

Environmental interactions rapidly decohere the superpositions he modeled, leaving a large separation from neural processing times.

Primary source ↗

Hagan, Hameroff & Tuszyński / 2002

A different model

They argued that Tegmark had modeled a different superposition from the Orch OR proposal and obtained longer estimates under altered assumptions.

Primary source ↗

Model dispute: calculation alone could not determine which physical description, if either, applies in living neurons.

02 / Anesthesia as a perturbation

Microtubule stabilization and anesthetic sensitivity

Volatile anesthetics act through multiple targets. Microtubules are one candidate. The cited experiments tested whether microtubule stabilization changes loss-of-righting-reflex latency.

Values reported in cited source

2024 · rats+69 s

Mean delay in loss of righting reflex after epothilone B under 4% isoflurane.

Animals: Male rats
Endpoint: Loss of righting reflex
Effect size: Cohen's d ≈ 1.9

Khan et al., eNeuro ↗

Values reported in cited source

2026 · mice+29 s

Within-subject increase in loss-of-righting-reflex latency reported after 8 mg/kg epothilone B.

Animals: Male and female mice
Endpoint: Loss of righting reflex
Effect size: Cohen's d ≈ 0.8

Huang et al., Neuropharmacology ↗

Endpoint note: loss of righting reflex is a standard rodent behavioral proxy for anesthetic-induced unconsciousness. It is not a direct measurement of phenomenal consciousness.

Observation

Microtubule stabilization changes anesthetic sensitivity.

The rat and mouse experiments report delayed loss of righting reflex after epothilone B. [11] [16]

Supports

A functional microtubule contribution to anesthetic action.

The reported effects are consistent with a microtubule contribution. Classical molecular explanations remain possible.

Does not establish

A quantum mechanism of consciousness.

The experiments do not by themselves demonstrate quantum coherence, entanglement, objective reduction, or a microtubular substrate of phenomenal consciousness.

03 / Our contribution

Our reviews: planning and perceptual cycles

The two companion reviews connect temporally deep active inference to a candidate physical implementation. Paper I focuses on integration over possible future trajectories. Paper II focuses on discrete perceptual cycles and evaluates Orch OR as a proposed mechanism.

Conceptual diagram

Observations

o₁, o₂, … oₜ

Hidden states

s₁, s₂, … sₜ

Candidate policies

π₁ · π₂ · π₃ · … · πₙ

Expected futures

Alternative trajectories weighted under the generative model.

Paper I argues that the integration over alternative future trajectories has a natural formal correspondence with quantum path-integral dynamics. This is a proposed mechanistic implementation, not an empirical demonstration that neurons perform quantum path integration. [13]

Paper I / theoretical claim

Planning across possible futures

We argued that temporally deep active inference requires integration over alternative future trajectories and that quantum path-integral dynamics offer a natural candidate implementation of that computation.

Read Paper I ↗

Paper II / theoretical claim

Discrete perceptual cycles

We evaluated Orch OR as a candidate physical mechanism for temporally structured conscious updating. The paper proposes a mechanistic correspondence; it does not experimentally establish objective reduction in neurons.

Read Paper II ↗

Mechanistic candidate / Paper IIConceptual diagram · theoretical proposal

Where Orch OR enters the argument

Orch OR as a proposed mechanism

Paper II evaluates Orch OR as one candidate physical mechanism for discrete perceptual cycles. It is not a requirement of active inference and is not presented as an established substrate of consciousness.

Evidence boundaryTubulin quantum states, orchestrated coherence, and objective reduction remain theoretical propositions. The schematic visualizes the hypothesis rather than an experimental observation.

Orch OR is shown here as a candidate physical mechanism proposed in Paper II, not as established microtubule physiology. [14]

04 / A disputed signal

The disputed MRI interpretation

A 2022 MRI study interpreted multiple-quantum-coherence measurements as possible evidence of a non-classical mediator associated with brain function. A published 2023 comment argued that the measurements did not justify an entanglement claim. The original authors published a reply.

Interpretation

Possible non-classical mediator

Kerskens and López Pérez reported an MRI signal they interpreted as consistent with a non-classical mediator and entanglement-related behavior.

Original paper ↗

Critique

Classical alternatives remain

Warren argued that multiple-quantum-coherence MRI can arise through classical spin dynamics and that the reported result did not establish entanglement.

Published comment ↗

StatusContested

Independent replication and stronger exclusion of classical alternatives are required. The original authors' published reply is part of that record. [8] [9] [10]

05 / Evidence

Established findings, contested interpretations, and hypotheses

Evidence map

Established

Frameworks and biology

Active inference provides a formal framework for perception, action, learning, and planning.
Neuronal population activity can encode probability distributions and support approximate Bayesian computations.
Quantum coherence can occur in some biological or biomolecular systems under specific physical conditions; that fact alone does not imply quantum consciousness.
Microtubules are intracellular cytoskeletal structures with established structural, transport, and signaling roles.
Volatile anesthetics act through multiple molecular and circuit-level targets.

Refs 3–7, 17–19

Empirically relevant

Perturbation results

Epothilone B delayed isoflurane-induced loss of righting reflex in a 2024 rat experiment.
A 2026 mouse study reproduced the direction of the microtubule-stabilization effect.
These results support a functional microtubule contribution to anesthetic action without identifying that contribution as quantum mechanical.

Refs 11, 16

Contested

Physical interpretation

Whether the 2022 MRI result warrants an entanglement or non-classical-mediator interpretation remains disputed.
Whether biologically relevant microtubule states preserve quantum coherence for the required duration remains unresolved.
Classical molecular explanations remain viable for the anesthetic perturbation results.

Refs 1–2, 8–10

Theoretical

Mechanistic proposal

Orchestrated objective reduction as a generator of conscious moments.
Collective microtubule quantum states as a substrate of consciousness.
Quantum path-integral dynamics as the physical implementation of temporally deep active inference.

Refs 12–15

06 / Falsifiability

Proposed experiments to distinguish quantum and classical models

A useful experiment would separate a specifically quantum mechanism from conventional microtubule biology and classical network dynamics.

Directly measure a candidate microtubule quantum degree of freedom in living neural tissue at a behaviorally relevant timescale.

Perturb that degree of freedom while preserving conventional microtubule structure and transport as far as experimentally possible.

Demonstrate a reproducible change in a prespecified conscious-state measure or tightly defined neural correlate.

Show that a quantum model predicts the intervention effect better than competing classical molecular and dynamical models.

Replicate the result independently under preregistered conditions with explicit negative controls.

Decision rule

A quantum account should earn explanatory status only if it predicts intervention outcomes that viable classical models do not, and those predictions replicate.

Visual appendix / Figure atlas

Conceptual figure appendix

Switch between diagrams of the three explanatory levels, possible policy trajectories, the Orch OR proposal, and experimental requirements.

2. Levels of explanationFrom principle to implementation.

◎

NORMATIVE

What should be computed?

Variational free-energy minimization

Optimality principles

⌘

PROCESS

How is it computed?

Neural message passing / population codes

Algorithms and dynamics

◉

MECHANISTIC

What is the physical substrate?

Candidate physical implementation

Biophysics and materials

Quantum active inference concerns a proposed mechanistic implementation, not a requirement of active inference itself.

Conceptual diagram

Normative, process, and mechanistic levels are kept distinct; the quantum proposal enters only at the mechanistic level.

07 / Published work

Published reviews

Paper IPeer-reviewed review article

Conscious active inference I: A quantum model naturally implements the path integral needed for real-time planning and control

Michael C. Wiest and Arjan Singh Puniani

Computational and Structural Biotechnology Journal, 30, 108–121 · 2025

A theoretical review proposing quantum dynamics as a possible implementation of path integration in temporally deep active inference. It does not demonstrate that implementation in neurons.

Role: Co-author
DOI: 10.1016/j.csbj.2025.09.017

Open article ↗PubMed ↗Full text ↗

Paper IIPeer-reviewed review article

Conscious active inference II: Quantum orchestrated objective reduction among intraneuronal microtubules naturally accounts for discrete perceptual cycles

Michael C. Wiest and Arjan Singh Puniani

Computational and Structural Biotechnology Journal, 30, 94–107 · 2025

A companion theoretical review arguing that Orch OR microtubule dynamics could provide discrete perceptual cycles for conscious active inference. It does not constitute experimental proof of a quantum mechanism for consciousness.

Role: Co-author
DOI: 10.1016/j.csbj.2025.09.016

Open article ↗PubMed ↗Full text ↗

Correction record: 2025 corrigendum correcting equations on page 102 ↗

References / Primary record

Sources behind the argument.

The references include the original studies, published critiques and replies, both reviews, and the Paper II corrigendum.