Microsoft: Generative Causal Testing — AI hypotheses about the brain validated by scanner
Generative Causal Testing (GCT) is a two-phase AI framework that converts opaque brain activity prediction models into testable hypotheses — and then verifies them with real fMRI experiments on human subjects.
This article was generated using artificial intelligence from primary sources.
What is Generative Causal Testing?
Generative Causal Testing (GCT) is a two-phase research framework that addresses a key problem in neuroscience: machine learning models for predicting brain activity yield accurate results, but do not explain why. GCT converts these opaque models into concrete, testable hypotheses.
How does AI ask the brain questions?
In the first phase, an LLM generates synthetic stories — short textual stimuli — targeted at activating specific brain regions. fMRI (functional magnetic resonance imaging, a technique that measures blood flow in the brain as a proxy for activity) then tests these stimuli on real subjects in the scanner. The results confirm or refute the hypothesis, creating a closed loop between the AI model and empirical experiment.
Three regions, new micro-areas
The system was validated on 3 subjects. GCT successfully distinguished 3 adjacent regions responsible for processing spatial places — the retrosplenial cortex (RSC), the parahippocampal place area (PPA), and the occipital place area (OPA) — which were difficult to distinguish using classical methods. In addition, the method revealed new prefrontal micro-regions not previously described in the literature.
Collaboration and implications
The project is the result of collaboration between Microsoft and researchers from UC Berkeley, UCSF, and Columbia. The approach opens a path toward systematic, AI-guided brain mapping — where the machine does not replace the neuroscientist, but proposes which experiments to conduct next.
Frequently Asked Questions
- What is Generative Causal Testing and how does it work?
- GCT is a two-phase method in which an LLM generates synthetic fMRI stories targeting specific brain regions, and the resulting hypotheses are then confirmed through real scanner experiments on subjects.
- What concrete results did GCT achieve on subjects?
- The system distinguished 3 adjacent place-processing regions (RSC, PPA, OPA) and discovered new prefrontal micro-regions, validated on 3 subjects.