What risks does AutoML pose in terms of interpretability?

Main risks of AutoML from an interpretability perspective:

• Model complexity: AutoML often selects complex ensembles or deep models that are difficult to interpret.
• Lack of transparency: Automated hyperparameter and architecture selection can lead to models whose logic is unclear to the user.
• Explainability issues: Standard interpretation methods (e.g., SHAP, LIME) may perform worse or be less informative for models created by AutoML.
• Loss of control: The user does not always understand why AutoML chose a particular model, which complicates trust and debugging.

To mitigate these risks, it is recommended to use AutoML together with interpretation tools and to limit model complexity if transparency is important.