On the Minimal Theory of Consciousness Implicit in Active Inference (2410.06633v2)

Abstract: The multifaceted nature of subjective experience poses a challenge to the study of consciousness. Traditional neuroscientific approaches often concentrate on isolated facets, such as perceptual awareness or the global state of consciousness and construct a theory around the relevant empirical paradigms and findings. Theories of consciousness are, therefore, often difficult to compare; indeed, there might be little overlap in the phenomena such theories aim to explain. Here, we take a different approach: starting with active inference, a first principles framework for modelling behaviour as (approximate) Bayesian inference, and building up to a minimal theory of consciousness, which emerges from the shared features of computational models derived under active inference. We review a body of work applying active inference models to the study of consciousness and argue that there is implicit in all these models a small set of theoretical commitments that point to a minimal (and testable) theory of consciousness.

• The paper proposes a minimal theory of consciousness by leveraging active inference and free energy minimization.
• It maps shifts in inferred states to changes in conscious experience, aligning policy selection with conscious access.
• The approach integrates diverse experimental models to promote empirical validation and unify varied consciousness phenomena.

A Minimal Theory of Consciousness via Active Inference

The paper "On the Minimal Theory of Consciousness Implicit in Active Inference" proposes a unique approach to the longstanding challenge of understanding consciousness by leveraging the active inference framework. Traditional methods in neuroscience tend to isolate specific facets of consciousness, such as perceptual awareness or global states, leading to theories that are often difficult to compare. This paper attempts to construct a minimal and testable theory of consciousness by utilizing the generality of active inference—a framework for understanding behavior as approximate Bayesian inference.

Active inference provides a broad network for modeling diverse behaviors by minimizing variational and expected free energy. The central thesis posited by the authors is that, although active inference is not a direct theory of consciousness, it offers a comprehensive framework through which the multifaceted nature of consciousness might be modeled. The paper systematically reviews various models applying active inference to consciousness, suggesting that these models inherently possess common theoretical commitments that point towards a minimal theory of consciousness. This theory is grounded in two objective functions: variational free energy, representing an organism's optimization of its internal model with reference to incoming data, and expected free energy, which accommodates future state planning through the reduction of risk, ambiguity, and novelty.

A critical feature of this theory is the mapping between the inferential components of active inference and conscious experiences, with a focus on the approximate posterior over states, $q(s)$, as the correlate of consciousness. According to the theory, all changes in conscious contents must correspond to changes in the inferred states of the world, body, or brain. Notably, the conscious/unconscious distinction is aligned with the mechanisms of policy selection inherent in active inference, associating conscious access with posterior beliefs processed at a discrete level within a generative model.

The paper details how the rigorous application of free energy principles might illuminate diverse consciousness phenomena, ranging from sensory processing and perception to states of self-awareness and meta-awareness. Examples cover binocular rivalry, visual neglect, and homeostasis-driven interoceptive phenomena, each modeled under the auspices of active inference.

Furthermore, the authors explore the theoretical implications of their approach using Lakatos’ framework of scientific research programs. They posit that active inference's hard core—its commitment to free energy minimization—functions as the unifying theory of consciousness, while the protective belt consists of specific model assumptions and simulations. A productive theory, they argue, should remain theoretically expansive and subject to empirical testing without degenerating into mere ad hoc modifications.

The paper concludes with implications for future research. It advocates for the open-ended development of active inference as a theory of consciousness, emphasizing the necessity of empirical validation and model specificity. Future developments could see the integration of more experimental observations, facilitating distinctions between conscious and unconscious processing across varying domains. Crucially, the authors envision active inference as a tool not only to model but to potentially unify the scientific understanding of consciousness in its many forms.

In speculative terms, the paper suggests that active inference could eventually provide a comprehensive theoretical edifice that supports both empirical data correlation and predictive capability. This represents a significant step forward in addressing what has been termed the 'hard problem' of consciousness, offering a mathematically grounded method for exploring conscious processes with a breadth that encompasses the full complexity of conscious experience.