Preparing to act follows Bayesian inference rules

Bayesian brain theories suggest that the human brain sets up predictive models to anticipate sensory evidence. Here, we propose that these models inform the entire functional hierarchy, scaling all the way down to the motor system. To test this hypothesis, 62 participants completed a probabilistic discrimination task while receiving bilateral single-pulse transcranial magnetic simulation (TMS) over the two M1s. Motor-evoked potentials (MEPs) were recorded to assess motor excitability for prior-congruent vs. incongruent actions. Results demonstrated that prior expectations shaped cortico-spinal excitability (CSE) prior to action execution, inhibiting prior-incongruent response preparation. Notably, drift-diffusion modeling revealed that trial-by-trial MEPs variations significantly influenced decision-making bias. Finally, interindividual differences in CSE modulations uncovered two distinct predictive strategies - the believers' style, which heavily relies on priors, and the empiricists' style, who downplay their role - driven by schizotypal and autistic traits, respectively. These findings suggest integration of predictive models into action representation, with prior-driven CSE modulations reflecting interindividual differences in predictive styles.