Active inference has emerged as an influential theoretical framework in cognitive neuroscience, offering a unifying account of perception, action, and cognition under the single principle of surprise minimization. Despite its growing prominence, the framework is often criticized for the difficulty of extracting qualitatively distinct, testable predictions and its limited empirical grounding. This review addresses these concerns by systematically outlining and evaluating key predictions of active inference across decision-making in discrete time and motor control in continuous time. For each domain, we offer a formal introduction of active inference and contrast it with leading alternative accounts. In the domain of decision-making, we identify predictions that agents behave more stochastically when they are uncertain about outcome predictions; that they explore to resolve uncertainty about hidden states and model parameters; and that preferences can be learned through accumulated experience. In the domain of motor control, we identify predictions that sensory input is attenuated early and non-specifically to facilitate movement initiation; that motor output is context-sensitive and reflects predictions of proprioceptive input; and that the brain adaptively controls and predictively corrects movements depending on estimations of uncertainty. Overall, we find that existing empirical work is broadly consistent with several of these predictions, but further research is necessary to draw more definitive conclusions. We highlight areas where evidence is promising, while emphasizing the need for theory-driven experiments that can adjudicate between accounts. In this way, active inference can move beyond a formal mathematical framework toward an empirically grounded theory of brain function.
Lageman et al. (Mon,) studied this question.