The Bayesian brain

The Bayesian brain is a dominant theoretical framework in contemporary cognitive neuroscience: the brain combines its expectations (priors) with sensory signals to infer the most probable cause of its inputs. It is a powerful and productive model, but one that remains debated.

Definition

Within this framework, perception does not passively receive reality: the brain predicts it, tests it and updates it constantly. Two streams meet. The top-down stream, carried by priors, brings together past experiences, internal knowledge and models, goals and expectations, context and beliefs, as well as a precision assigned to hypotheses. The bottom-up stream carries sensory signals and correction signals: vision, hearing, touch, smell, taste, interoception (cardiac, visceral, homeostatic signals), perceptual salience, input precision, and ascending prediction error. The higher hierarchical levels generate descending predictions about the probable causes of the signal; the lower levels send back corrective signals weighted by their precision.

Bayesian inference

Perception results from an implicit and automatic inference. The brain minimises prediction error by adjusting its internal models, and sometimes by acting on the world (active inference). Perception is the implicit inference of the most probable cause of the inputs. The subject does not consciously choose the perceived world: it already imposes itself as inferred.

Models and applications to altered states

The plate applies this framework to consciousness-altering substances, each disrupting one cog of inference. LSD and psilocybin produce a relaxation of descending priors (REBUS). N,N-DMT corresponds to an explosion of prediction error and a rapid reconstruction of the model. Ketamine produces a rupture of integration, dissociation and an alteration of filtering. 5-MeO-DMT causes a deep interoceptive disturbance. The Bayesian framework thus provides a common language for thinking about these disruptions of inference, without essentialising them.

Debates and limits

The plate does not hide the objections. The framework is theoretical and dominant, but does not constitute a consensus as an ontology of the brain. Its testability and falsifiability are debated. Its strength lies in its great unifying power; its fragility lies in an interpretive flexibility that is sometimes excessive (a framework that explains everything risks predicting nothing refutable). Alternatives exist: ecological perception (Gibson, Chemero) and enaction (Varela, Thompson). In short, the brain does not passively receive reality: it predicts it, tests it and updates it constantly.

Sources

• Helmholtz (1867), perception as unconscious inference.
• Friston (2010), the free energy principle.
• Clark (2013), the predictive brain.
• Hohwy (2013), the predictive mind.
• Seth (2013), interoceptive inference and perception (and Being You, 2021).
• Carhart-Harris and Friston (2019), the REBUS model and psychedelics.
• Gibson and Chemero, ecological perception; Varela and Thompson, enaction.