A beautiful loop: An active inference theory of consciousness

A beautiful loop: An active inference theory of consciousness

One. Introduction

Consciousness is perhaps the biggest mystery in science. At a certain point, most fields of inquiry find that the strange capacity of organisms to experience cannot be overlooked. Books, articles, and media discussing the nature of consciousness abound, with unique perspectives emerging from psychologists, neuroscientists, philosophers, phenomenologists, computer scientists, biologists, physicists, and contemplatives. Yet, most would agree that consciousness remains an inconvenient enigma in a world that otherwise seems reducible to things, objects, patterns, and equations.

Equally, it is clear that the tools of science can reveal something about the nature of consciousness. Thousands of experiments attest that consciousness has predictable characteristics, predictable correlates, and fluctuates under predictable conditions. Thanks to this growing evidence base, an array of impressive theories of consciousness have emerged in recent years

Abstract

Can active inference model consciousness? We offer three conditions implying that it can. The first condition is the simulation of a world model, which determines what can be known or acted upon; namely an epistemic field. The second is inferential competition to enter the world model. Only the inferences that coherently reduce long-term uncertainty win, evincing a selection for consciousness that we call Bayesian binding. The third is epistemic depth, which is the recurrent sharing of the Bayesian beliefs throughout the system. Due to this recursive loop in a hierarchical system (such as a brain) the world model contains the knowledge that it exists. This is distinct from self-consciousness, because the world model knows itself non-locally and continuously evidences this knowing. Formally, we propose a hyper-model for precision-control, whose latent states (or parameters) encode and control the overall structure and weighting rules for all layers of inference. These globally integrated preferences for precision enact the epistemic agency and flexibility reminiscent of general intelligence. This Beautiful Loop Theory is also deeply revealing about altered states, meditation, and the full spectrum of conscious experience.

These theories have many strengths and explanatory power but a general consensus in the scientific community does not exist. Even the metaphysical assumptions underlying the science of consciousness still result in fierce debates.

Here, we aim to contribute to these theories by building on a promising general theory of organisms, known as active inference or predictive processing, under the free energy principle. Several others have proposed that active inference may provide solutions to different features of conscious experience. Nevertheless, it remains unclear whether active inference can satisfy the conditions for a theory of consciousness. Yet others have proposed that we should think of active inference as providing "theories for consciousness science, rather than theories of consciousness per se". The question arises, what conditions would active inference need to satisfy in order to cross the theory of consciousness threshold? Why is it that the theory is so successful in explaining perception, cognition, and action, but not consciousness itself?

To address these questions, we propose three conditions that seem necessary for consciousness and show how some active inference systems satisfy them. The first condition is a generative world model, or epistemic field. This provides the 'space' or contents that can be known, hence the term epistemic. The second condition is inferential competition, which determines what becomes conscious and why it is coherent. The third and final condition is epistemic depth, which refers to the fact that the epistemic field is recursively, and widely, shared throughout the system. These three conditions together form what we call a 'beautiful loop' that allows the generation of a coherent and aware world model. As we will see, this idea has some similar characteristics to "broadcasting", "information integration", and "fame in the brain", albeit with important differences.

Below, we will introduce one condition at a time. We will then show how active inference can provide a parsimonious explanation for a range of cognitive processes and states of consciousness when these conditions are satisfied. Our view also implies that the most basal or minimal form of awareness is a highly simplified (nearly contentless) world model knowing itself recursively. Therefore, a first person perspective, self-modeling, and agency, are not prerequisites of awareness, but are instead common forms or features of conscious world models.

To be concise, we will avoid an extensive literature review. However, as noted above, many features of the theory are consistent with elements of other theories of consciousness such as Global Neuronal Workspace Theory, Higher Order Theories, Recurrent Processing Theory and Integrated Information Theory. Links with existing theories will be made throughout. The strength of our approach will be in showing how the interactions of a minimal set of computational assumptions within active inference may provide the ingredients for consciousness, with implications for understanding various states from lucid dreaming to meditation, psychedelics, and artificial intelligence.