A beautiful loop: An active inference theory of consciousness

gist

Ruben Laukkonen, Karl Friston and Shamil Chandaria argue that active inference yields consciousness when a world model recursively knows itself. Their Beautiful Loop Theory sets three conditions: a unified reality model, competition among inferences to enter it, and epistemic depth, where beliefs are shared back through the whole hierarchy. A global hyper-model that predicts precision at every layer formalizes that depth. The paper applies the account to attention, sleep, meditation and artificial intelligence. This is the accepted preprint of a 2025 Neuroscience and Biobehavioral Reviews article.

ideas

• A simulated world model forms the epistemic field. Hierarchical precision-weighted prediction error minimization builds a model of reality that fixes what the organism can know or act upon.
• Coherence decides what becomes conscious. Inferences compete, and those that fit with parallel and adjacent inferences while reducing long-term uncertainty get bound into experience, which the authors call Bayesian binding.
• Epistemic depth is a recursive loop. Beliefs are shared throughout the hierarchy so the world model contains the knowledge that it exists, which the authors separate from self-consciousness.
• A hyper-model controls precision globally. It forecasts the precision profile every layer should use, and errors on those forecasts fuse into one organism-wide signal.
• Altered states test the theory. Minimal phenomenal experience in meditation is modeled as sharp precision on an empty world model, which reads as vivid yet contentless awareness.

quotes