Neuroscientist pushes back: task-trained nets converge on brain-like representations that can predict primate cortex

coherence · x · 2026-09-28

coherence reposts @camhberg's rebuttal to the claim that neural networks tell us nothing about brains:

He concedes key differences exist (substrate, neuromodulation, recurrence) and that nobody serious claims machine intelligence equals human intelligence.
Still, he argues brains are the closest reference class for frontier AI dynamics, with no close second.
The evidence: a growing literature showing networks trained only on tasks converge on brain-like representations—close enough to predict and even steer activity in primate and human cortex—with multiple studies cited.

More from AGI Musings