EPFL Scientists Built an AI That Engineers Videos to Activate Specific Brain Regions

Reporting from Hacker News, researchers at EPFL have developed an AI system that generates videos engineered to maximally stimulate targeted brain regions — the first demonstration that generative models can be optimized directly for neurological response rather than human preference.

The project uses a closed-loop approach: the AI generates visual stimuli, measures brain responses via neural recording, and iterates until specific circuits fire reliably. The result is content that can trigger targeted neural activity on demand.

The neuroscience applications are real — mapping how the visual cortex processes information, or studying responses in patients with neurological conditions. But the same optimization loop, pointed at reward pathways or attention circuits, produces something quite different. The researchers acknowledge the dual-use risk.

Nobody making advertising algorithms is currently reading neuroscience papers this carefully. Nobody you know about, anyway.

Read the full story at EPFL NEVO Project