nucleation

LIT-1364: On the minimal theory of consciousness implicit in active inference

3 min read

Properties5
StatusActive — read in full 2026-10-22 (NOTE-1062) from arXiv:2410.06633v2, the submitted version; worth reading as the clearest statement of the theory of consciousness that active inference models already assume: conscious contents are posterior beliefs about world, body or brain, and a posterior is conscious when it has the capacity to inform discrete, counterfactual policy selection, slowly and precisely enough. It also reviews the simulations of rivalry, access, no-report, interoception, valence and meta-awareness, and names two predictions close to its core. Its main limit is scope: the criterion is one of conscious access, read off models built to report, and the authors hold it only for humans and similar animals. A review with three small simulations of its own; neither prediction is tested.
Version2
Tagsconsciousness, neuroscience, probabilistic-modeling
Cited as “paper” byNOTE-1062 — Whyte et al. — On the minimal theory of consciousness implicit in active inference
Cited as “source” byTHEORY-883 — Under active inference, the contents of consciousness are posterior beliefs about the hidden states of world, body or brain, and a posterior is conscious when it has the capacity to inform discrete, counterfactual policy selection, at a timescale slow enough to abstract from sensory flux and with enough precision to tell alternatives apart; on this account no conscious content changes without a change in the posterior, and a stimulus given precision by an attention or saccade policy should reach awareness sooner than a retinally matched one

Graph View

Whyte, C. J., Corcoran, A. W., Robinson, J., Smith, R., Moran, R. J., Parr, T., Friston, K. J., Seth, A. K. & Hohwy, J. (2026). On the minimal theory of consciousness implicit in active inference. Physics of Life Reviews 56: 4–28. doi:10.1016/j.plrev.2025.11.002. arXiv:2410.06633.

Key takeaways

From the reading, NOTE-1062. Page numbers are arXiv v2’s.

  • A theory read off the models. Starting from active inference, not from a property of consciousness, the paper argues that the existing models of consciousness paradigms share a “hard core”. It is set out in Lakatos’s terms, with most of the explanatory work left to the “protective belt” (p. 28).
  • The core. (1) Minimising variational and expected free energy, which covers “both conscious and unconscious” behaviour (p. 25). (2) Conscious contents correspond to the posterior over hidden states, so no content changes without a change in the posterior. (3) Conscious access occurs at “the discrete-continuous interface” (p. 26): a posterior is conscious when it has “the capacity (in a counterfactual sense)” to inform policy selection at the right timescale and precision (p. 27).
  • Its scope. Held “necessary and sufficient” for content to become conscious in humans and similar animals; the authors “remain agnostic” about systems “vastly different to ourselves” (p. 27). The third part is “an inference to the best explanation” (p. 27).
  • The models reviewed. Rivalry and Troxler fading as attention-driven precision, hemineglect as lost novelty, access and the P3b, a no-report model whose prefrontal and P3b-like responses drop while the stimulus stays reportable, homeostatic and interoceptive models, emotion concepts, valence and meta-awareness (pp. 13–22).
  • New here. A rivalry model that fits Levelt’s laws (Appendix 1); a predicted reward violation of them (Appendix 2); and a sleep model in which lower transition precision removes the P3b-like response and spares the MMN, which “does not in-itself explain why participants lose consciousness” (p. 24).
  • Predictions. Reward should trade off against stimulus precision in rivalry, and “To See is to Look”: a stimulus given precision by an attention or saccade policy should reach awareness sooner than a retinally matched one (pp. 29–30).

Why this is here

The owner’s exchange of 2026-10-21, “Experience and interpretation”, asked whether experience is constituted by interpretation or produced by it. The exchange cited this paper for extracting a minimal, testable theory of consciousness from what active inference models have in common. It took the paper as one route from computational organization to stated conditions on experience, which is the step the thesis needs. Bears on QUESTION-254 and QUESTION-048.

What the reading changes. The exchange’s description is accurate as far as it goes. What it leaves out is that the conditions are conditions of conscious access in humans. The paper does not claim to say which organizations are experiential, and it grants that non-conscious creatures may run active inference (p. 26). For QUESTION-254 it is a candidate on the organizational side only, and its criterion is tied to report, so it does not meet CLAIM-411’s defeater. It bears on QUESTION-048 through its no-report model.

Filed from this work on 2026-10-22: THEORY-883, its criterion of conscious access.

Backlinks