The biggest elephant in the room is that ML/AI, call it what you want is a bit of a closed black box - you can see what it does (to some extent) but you don't know fully how each model does it. It has got many use cases where safety and accuracy isn't so important but how do you pass each model...