An Active Inference view of cognitive control

Cognitive control allows the control of though processes and provides top-down guidance to overt behavior towards distal goals. We cast cognitive control within Friston's "active inference" framework, suggesting that it can be mechanistically realized with a nesting of optimizations (i.e., free energy minimization loops) over time, with one overt loop and one (or more) covert loop(s). The covert loop produces "controlled predictions" that support the imagination of future scenarios and the planning of distal goals. Once formed mentally, distal goals can be achieved by providing priors (over sets of desired states and transitions) to the overt loop. This approach is compatible with the view of cognitive control (and prefrontal cortex) as providing flexible top-down guidance to overt behavior besides stereotyped stimulus-response mappings. At the same time, in this approach the control of though processes consists in covert action-perception loops that obey to the same optimization principles and use the same resources as the overt action-perception loop, rather than modularized processing. We propose that the nesting of optimization processes is realized in prefrontal hierarchies and is functional to the achievement of (distal) goals at different time scales.