A prominent theory holds that cognition works by minimizing prediction errors through Bayesian inference, with attention understood as optimizing the precision of those error signals. While this account explains many attention-related phenomena, it fails to accommodate certain forms of voluntary attention. The authors argue that advocates of Bayesian prediction error minimization have overreached by claiming it is all the brain ever does, and that the theory's tools, though powerful, are insufficient for a complete explanation of attention.
The word 'attention' is often used loosely, but this paper treats it as a term defined by its role in explaining empirical psychological phenomena, particularly those involving reaction-time modulations. Evidence shows that such modulations are linked to processing that stands in various relations to consciousness. The psychological phenomena that explain these modulations should not be identified with the causes of consciousness, nor do they operate exclusively within conscious awareness. If such explanations are to shed light on how and when consciousness occurs, they must do so within a more complex explanatory theory.