[antipaul]

It’s not only outputs/labels that provide “insights”.

Knowing how the outputs relate to the inputs is where most new insights could come from.

For example, what feature (input) is driving the “failure” of the machine (output/prediction)?

This is where ML explainability comes in.