I don't think that a root cause analysis like this is good for anything else than improving the algorithm.

It's the same as with face detection. The algorithms are better far superior to humans, but they fail at occasions where a human would not even think twice