I always thought that area of research had the coolest name, too: “mechanistic interpretability”