Shape of subjectivity: an active inference approach to consciousness and altered self-experience

Abstract

How should we understand the place of the mind in the natural world? Can the relationship between the contents of consciousness and the underlying mechanisms be identified? This thesis approaches the question of consciousness and the self through the framework of active inference. According to predictive processing approaches to brain function, brains are essentially prediction machines. On this view, perception and action are underpinned by inferential mechanisms that implement a hierarchical generative model, constantly attempting to match incoming sensory inputs with topdown predictions or expectations. Predictive processing is thought to offer a first glimpse of a unified theory of the mind—uniting perception, action, and cognition under a single theoretical framework. In particular, active inference, under the free energy principle, has emerged as the most explanatorily powerful approach in predictive processing. In this thesis, I develop a conceptual framework within active inference for understanding consciousness and phenomenal selfhood (broadly, the ‘sense of being a self’) in terms of an “allostatic control model”. I made the case that phenomenal selfhood arises from a hierarchically deep inference about endogenous control of ‘selfevidencing’ (survival-relevant) sensory outcomes.

I apply this account to develop a new understanding of the relationship between self-consciousness and consciousness. Based on the allostatic control model, I posit a novel theoretical model of how psychedelic drugs can lead to ‘selfless’ experiences. I then apply the allostatic control model to characterise the contrastingly dysphoric and euphoric selfless experiences that can arise in depersonalisation disorder and meditation practice. Based on these accounts, I consider the possibility of a theory of consciousness within this active inference, analysing whether selfless experiences pose a threat to an active inference theory of consciousness understood in terms of selfmodelling mechanisms. I argue that selfless experiences do not pose a threat to an active inference theory of consciousness, rather selfless states can be informative as to how consciousness should be understood in active inference. Consciousness emerges as fundamentally affective on this view, where (in normal experience) hierarchically deep self-modelling mechanisms function to ‘tune’ organisms to opportunities for adaptive action across multiple interlocking timescales.