Integrated world modeling theory expanded: Implications for the future of consciousness.
Front Comput Neurosci November 24, 2022 Adam Safron 32 citations
Integrated world modeling theory (IWMT) combines the free energy principle with integrated information theory and global neuronal workspace theory to offer a unifying account of consciousness. The article reviews philosophical principles and neural systems behind IWMT, then describes predictive processing models linked to machine learning architectures such as autoencoders, turbo-codes, and graph neural networks. It suggests new ways to estimate integrated information using probabilistic graphical models, flow networks, and game theory, and addresses debates about the physical substrates of conscious and unconscious phenomena. The work also explores the 'Bayesian blur problem'—how discrete experience arises from probabilistic modeling—and discusses attentional selection and the evolutionary origins of consciousness through unlimited associative learning.