AI Explainability

“Perhaps we need to solve the first problem first, which is just understanding when models are not performing, or like when there are errors or failure cases before we can understand why.”