AI-generated videos to maximally drive a target brain region
Points and comments are a snapshot, not live.
NEvo evolves AI videos to maximally drive target brain regions, mapping social-dynamic selectivity.
NEvo uses a 'digital twin' encoding model to predict neural responses and then evolves video prompts via genetic algorithms to maximize activation in targeted visual brain regions. The method operates in two stages: first finding the strongest still image, then animating it into a two-second clip. Synthesized videos outperform handcrafted localizer clips and natural videos. Searchlight analysis from V1 to aSTS reveals a gradient from simple patterns to social interaction. Abstract probes confirm that optimizing for pSTS generates face-like characters, while MT optimization yields pure motion.
What commenters are saying
Commenters are uneasy, comparing the work to supernormal stimuli and the short story 'BLIT', warning of potential misuse for addiction or control. Many see parallels with social media algorithms already exploiting attention. A few defend the research as a tool for neurorehabilitation or insight into brain function. Some dismiss the videos as underwhelming and the technique as overhyped. There is a split between those who see inevitable harm and those who argue the science itself is neutral, with applications depending on incentives.