Intrinsic fluctuations in global connectivity reflect transitions between states of high and low prediction error

Abstract

While numerous researchers claim that the minimization of prediction error (PE) is a general force underlying most brain functions, others argue instead that PE minimization drives low-level, sensory-related neuronal computations but not high-order, abstract cognitive operations. We investigated this issue using behavioral, fMRI, and EEG data. Studies 1A/1B examined semantic- and reward-processing PE using task-fMRI, yielding converging evidence of PEs global effects on large-scale connectivity: high-PE states broadly upregulated ventral-dorsal connectivity, and low-PE states upregulated posterior-anterior connectivity. Investigating whether these global patterns characterize cognition generally, Studies 2A/2B used resting-state fMRI and showed that individuals continuously fluctuate between ventral-dorsal (high-PE) and posterior-anterior (low-PE) dynamic connectivity states. Additionally, individual differences in PE task responses track differences in resting-state fluctuations, further endorsing that these fluctuations represent PE minimization at rest. Finally, Study 3 combined fMRI and EEG data, and the study found that the fMRI fluctuation amplitude correlates most strongly with EEG power at 3-6 Hz, consistent with the PE network fluctuations occurring at Delta/Theta oscillation speeds. This whole-brain layout and timeline together are consistent with high/low-PE fluctuations playing a role in integrative and general sub-second cognitive operations.