Born to act: Deferred action and desire as active inference.

The active inference framework (AIF) considers the brain as a generative model guiding behavior under the imperative of minimizing the model's variational free energy. Computationally, this is accomplished by hierarchical Bayesian inference. The theory views organisms as doxastic agents, which has drawn the criticism of being insufficient to explain conative agents motivated by desire. Specifically, it has been noted that the concept of desire is not isomorphic with belief and, therefore, fits poorly with AIF. In this paper, we build on previous work that suggests a path to integrating conation in AIF and present three arguments. First, the dichotomy between belief and desire is unnecessary. To that end, we define desire as a hierarchical inference that starts from a domain-general inference on the agent's affective dynamics (affective charge) and descends to contextualized inference on the precision of action policies. We suggest that this hierarchy is implemented by a coordinated activity of the intrinsic brain networks: default mode, action mode, executive, and salient. Second, we argue for a central role that deferred action plays in the process of desire by allowing for affect-dependent awareness of the agent's states over different timescales. Third, we suggest that the proposed model of desire and deferred action has ramifications for understanding psychopathology, which we frame as the malfunction of deferred action and desire and use obsessive-compulsive disorder and depression as examples. This view entails that disorders of affect and motivation are subjectively desired despite their associated suffering.