Results from neural networks support the idea that brains use predictions to create perceptions—and that they work that way to conserve power.

Neuroscientists are finding that deep-learning networks, often criticized as “black boxes,” can be good models for the organization of living brains.