Laukkonen, R., Friston, K. & Chandaria, S. (2025). A beautiful loop: An active inference theory of consciousness. Neuroscience & Biobehavioral Reviews 176: 106296. doi:10.1016/j.neubiorev.2025.106296. Preprint: PsyArXiv, doi:10.31234/osf.io/daf5n.
Key takeaways
From the reading, NOTE-1063. Page numbers are preprint v3’s.
- Three conditions. A unified generative world model, an “epistemic field”; precision-weighted competition to enter it, in which “coherence and boundedness are a central criterion” (p. 6), called Bayesian binding; and epistemic depth, the world model “reflected back through the underlying hierarchy” (p. 9). Each is offered as necessary.
- Epistemic depth as a hyper-model. A higher-order set of parameters Φ forecasts and broadcasts the precision every level should have, and is updated by errors on that forecast (pp. 11–15, Table 1). The paper separates this system-wide loop from local “parametric depth” and from metacognition: “this is achieved through recursion not metacognition” (p. 20).
- Awareness without content, and without a self. “Awareness depends on epistemic depth more than configurations of contents” (p. 20). Minimal phenomenal experience is maximal depth with a contentless model (p. 22); cessation in meditation is a model that has become incoherent (p. 25). A first-person perspective, self-modelling and agency are “not prerequisites of awareness” (pp. 2–3).
- Implementation. Neuromodulatory “sprays” and endogenous electric fields are proposed as the broadcast channels (p. 17).
- AI. Current systems lack epistemic depth above all (p. 27). The main text concludes that a system meeting the conditions should be judged conscious (p. 29); note 11 says the conditions do not guarantee “phenomenal experience in the subjective sense” (p. 29).
- No test. There are no data and no simulation. The paper calls itself “first and foremost a conceptual one” (p. 18).
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 its extra organizational condition: higher-order modelling of precision, beyond inference as such. It took that as one candidate for what makes an interpretive organization experiential, and noted the paper’s use of altered states and meditation as evidence. Bears on QUESTION-254.
What the reading changes. The exchange’s description is right: the paper adds higher-order modelling of precision to inference as such. Two qualifications. The altered states and meditation are redescribed in the theory’s terms, not offered as tests of it, so they are illustrations, not evidence. And the paper denies that a first-person perspective is needed, which bears on how TERM-072’s senses of perspective enter the constitutive thesis.
Filed from this work on 2026-10-22: THEORY-882, its three conditions.