Jesse Hoogland–AI Risk, Interpretability