Motor and operative intentionality in computational phenomenology: from shallow to deep active inference models.

Computational phenomenology has been recently presented as a novel and promising approach to modelling human experience with the use of computational techniques. More specifically, as it has been discussed in recent work, it links phenomenological descriptions and analyses of lived experience and generative modelling techniques, most prominently, through Active Inference. It is argued that the experiential disclosure of the world and time-consciousness can be cast in terms of free energy minimization and inferential dynamics. It has not been addressed yet, however, whether and to what extent Active Inference may be able to model fundamental structures that constitute what phenomenologists call 'motor intentionality', i.e. our general embodied and practical openness to our experiential world. This openness is taken to underlie all experience. To address this gap in the literature, we propose to distinguish two different kinds of practical aspects within Active Inference. First, we discuss the approximation to a true posterior as a form of active process that can be linked with some of the core characteristics of what phenomenologists call 'passive syntheses'. Secondly, we discuss the process of action selection under likelihood mappings and preferences and link it with the constitution of a field of affordances of an embodied subject, as well as with the idea that the body is 'the first prior' within an Active Inference model. This embodied constitution is the core of what phenomenologists call 'motor intentionality'. Insofar as these two practical aspects of Active Inference are processes undertaken within the model, we emphasize the temporality that underlies them in terms of subjective temporality and objective time.