Understanding the brain with AI-driven explanations and experiments

A new generative causal testing (GCT) model has been developed through collaboration between Microsoft Research and several universities. This model translates brain response predictions into short verbal explanations, enhancing our understanding of how different brain areas respond to language. Experiments demonstrated that GCT can confirm known selectivity in place processing and identify micro-regions in the prefrontal cortex that respond to specific concepts. This addresses the issue of black-box models in large language networks, which are often unreadable.