Interpretable abstractions of artificial neural networks predict behavior and neural activity during human information gathering

Nature Neuroscience, Published online: 26 June 2026; doi:10.1038/s41593-026-02342-9

D’Ambrogio et al. combine deep learning and symbolic regression to report an interpretable equation of how humans value information. The equation predicts choices and neural activity in anterior insula, cingulate cortex and midbrain nuclei.