Karl Friston, Thomas Parr
Neuroscience has witnessed a curious inversion over the past decades, with a switch from views of the brain as abstracting information from sensory input to a view of the brain as a constructive organ, actively generating explanations for – and sampling – the sensorium. Perhaps this inversion is not strange, but a reversion to the ‘facts of perception’ as laid out by Helmholtz. His enduring view of perception as inference – to the best explanation – now dominates the neurosciences in the form of the Bayesian brain paradigm and the Free Energy Principle. This review follows the story from Helmholtz’s unconscious inference to modern day active inference. In particular, it uses the history of predictive coding – and how it has been generalized over the years – to furnish a framework that brings theory and empirical study in close contact, through generating mechanistic (anatomical and physiological) hypotheses about sense-making in the brain.