A model validation team is challenged by a model owner who argues that an explainable AI technique (LIME) used during validation is itself imperfect and therefore the validation findings are invalid. How should the validation team respond?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — b is correct because explainability techniques like LIME have known approximation limitations, but this does not invalidate their use as one component of a multi-method validation approach — good validation practice triangulates findings across multiple techniques rather than relying on any single method. A would undermine the independence of validation.
Full explanation below image
Full Explanation
B is correct because explainability techniques like LIME have known approximation limitations, but this does not invalidate their use as one component of a multi-method validation approach — good validation practice triangulates findings across multiple techniques rather than relying on any single method. A would undermine the independence of validation. C simply shifts the same argument to a different technique. D is disproportionate for a methodological debate.