New study links consciousness theory to self-organization in living neural networks

A new paper from researchers at the University of Tokyo (Mayama, Akita, Shimizu, Takano, and Takahashi) presents an empirical connection between two influential but previously separate frameworks in neuroscience: Integrated Information Theory (IIT), which ties consciousness to a system's causal integration, and the Free-Energy Principle (FEP), which describes self-organization through variational free-energy minimization. The team grew dissociated neuronal cultures on multi-electrode arrays and trained them to infer the location of a hidden signal source by observing repeated patterns of stimulation. Over time, the networks improved their inference accuracy while their variational free energy decreased, consistent with FEP predictions.

Crucially, the researchers computed an IIT-inspired proxy for integrated information and tracked the size of the ‘main complex’ (the set of elements with maximum integrated information). Both measures followed a non-monotonic, hill-shaped trajectory — rising early during learning then declining as the network stabilized. The integrated-information proxy correlated most strongly with Bayesian surprise (the divergence between prior and posterior beliefs) rather than with accuracy or free energy alone. Using an Ising model analysis, the team showed that Bayesian surprise and integrated information can be jointly amplified near shared positive critical modes, suggesting a mechanistic basis for the observed dynamics. This work provides a rare empirical point of contact between IIT and the FEP, and hints at how belief updating and integration might co-evolve in living neural systems.