The term subpersonal priors refers to prior beliefs or expectations encoded at the level of the brain’s unconscious, automatic processes, rather than at the level of conscious, personal awareness. These priors operate within the brain’s generative model, which underpins active inference by predicting and interpreting sensory inputs to guide behavior.
In the context of active inference, subpersonal denotes cognitive processes that occur below the level of conscious thought and are attributed to neural or computational mechanisms rather than to the person as a whole. For example, while a person may consciously decide to act in a certain way (a personal-level decision), their brain’s subpersonal systems are constantly generating predictions and updating beliefs based on sensory data without conscious input36.
By absorbing incentive values into these subpersonal priors, active inference avoids explicitly representing rewards or goals as separate entities. Instead, preferred outcomes are implicitly encoded within these priors as having higher probabilities, shaping action selection automatically based on what is most likely to minimize uncertainty or achieve desired states1