Generative causal testing to bridge data-driven models and scientific theories in language neuroscience
Researchers have developed a new framework called Generative Causal Testing (GCT) to better understand the relationship between large language models (LLMs) and scientific theories in language neuroscience. GCT aims to explain what specific features of language stimuli drive responses in different brain areas, moving beyond the opaque nature of current LLM representations. The framework has shown success in explaining selectivity in individual voxels and brain regions, including newly identified microROIs in the prefrontal cortex, and can differentiate between areas with similar functional selectivity. This approach seeks to bridge the gap between data-driven AI models and formal scientific theories. AI
IMPACT This research offers a novel method for interpreting LLM behavior in the context of neuroscience, potentially leading to more transparent AI models.