The issue is that it's still utterly beholden to the data-set in ways humans are not, because the underlying method cannot improvise, because the results are so statistically chaotic that it cannot be trusted to adapt in the field to anything without the model breaking.
Again, it will require a...