Understanding the brain with AI-driven explanations and experiments
TL;DR
Microsoft Research introduces Generative Causal Testing: an LLM turns black-box brain models into short hypotheses such as „food preparation” or „location names”. The loop is then tested in scanners: an LLM writes stories designed to activate a target brain region, and three participants hear them during fMRI scans. GCT confirmed known language selectivity, separated nearby place-processing regions such as RSC, PPA and OPA, and surfaced tiny prefrontal micro-regions.
Nauti's Take
This is where AI in science gets genuinely useful: not as a magical answer machine, but as a tool that proposes sharper experiments. Still, the framing needs caution.
Three participants are a small base, and Microsoft presents the story with predictable polish. The stronger point is the method: black-box prediction gets translated back into hypotheses that can be tested against the real world.
Briefingshow
Many AI models can predict brain activity surprisingly well, but they do not explain what a region is actually responding to. GCT tries to close that gap by turning model patterns into testable verbal hypotheses. The important part is the closed loop: generate a hypothesis, create a stimulus, test it in the scanner.
Sources