Right? Aren't we constantly told that scientists smarter than us little people can't figure out what's going on inside of deep learning structures? How do we know they might not be more moral because they don't yet have a limbic system mucking up their logic with the three f's? Orthogonality says nothing about motivation, only that it is independent of intelligence. Maybe the paperclip collector bot will decide that it cannot complete its task without the requestor present to validate? We don't know.