Does the brain learn like AI models? Not quite.

Researchers compared how deep learning models and human brains process visual information using fMRI and brain recordings. Backpropagated gradients (the learning signals in neural networks) did predict brain activity in higher visual areas, but the timing and spatial order were wrong—gradients flow in different patterns than what brain hierarchies would predict. This suggests AI and brains may converge on similar visual representations despite using fundamentally different learning mechanisms.