To develop the methodological and theoretical foundations of interpretable AI through the lens of causality and abstraction.

We believe that the understanding and control provided by interpretability research will be fundamental to creating safe AI models that benefit humanity.