I would expect any safety-critical application to be have a higher bar for explainability. As an extreme example, if image recognition is used for anti-aircraf

I would expect any safety-critical application to be have a higher bar for explainability. As an extreme example, if image recognition is used for anti-aircraft defense, I think it would be nice to understand how it differentiates between an enemy plane vs. a domestic air carrier. Knowing it's accuracy is probably not enough.