This presentation will discuss the physical and computational substrates of consciousness as suggested by Integrated World Modeling Theory (IWMT). I will first review how neural synchrony may entail marginalization over Bayesian networks, implementing the implicit calculations of joint beliefs and establishment of marginal message-passing regimes within a predictive processing context. I will describe how association cortices may correspond to shared latent (work)spaces within an autoencoding framework, potentially structured according to principles of geometric deep learning. Along these lines, I will further describe how Scott Aaronson’s “expander graph” critique of Integrated Information Theory may actually provide surprisingly strong support for the use of Phi in identifying necessary (but not sufficient) conditions for realizing potentially conscious systems. I will then discuss how IWMT’s original proposal of the realization of consciousness via alpha-synchronized subnetworks may have contained an error, with subjective experience potentially entailed by beta-complexes generating inferences at faster speeds with more restricted scopes. Finally, time permitting, I will present a model of how pleasure and pain may correspond to conflicting predictions within these subnetworks, potentially suggesting mechanisms by which such processes could be modified with relatively low-tech interventions.