Mechanistic interpretability

A research field focused on understanding how AI models, especially deep learning networks, arrive at their decisions by analyzing their internal components.