Subjective experience is multifaceted, making it hard for traditional neuroscientific theories of consciousness to be compared because each focuses on different aspects like perceptual awareness or global states. This work instead starts from active inference, a first-principles framework that models behavior as approximate Bayesian inference, and builds a minimal theory of consciousness from shared features of computational models derived under active inference. By reviewing studies that apply active inference models to consciousness, the authors identify a small set of theoretical commitments implicit in these models, pointing toward a minimal and testable theory of consciousness.
A family of functional connectivity measures based on tracking the 'center of mass' between two brain signals outperforms other measures at decoding conscious visual perception from magnetoencephalography data. These measures generalize across brain regions central to both Integrated Information Theory and Global Neuronal Workspace Theory. Neural mass models simulating each theory's hypothesized dynamics showed that both the GNWT-based model (featuring delayed ignition dynamics) and the IIT-based model (relying on synchronous sensory dynamics) captured the observed connectivity patterns. However, the presence of ignition dynamics independent of task-demand conditions contradicts IIT predictions, lending tentative support to GNWT. The work introduces a framework for systematically identifying and testing neural correlates of conscious vision.