An AI will always behave logically, it just may not be consistent with your definition of “logical.” Their outputs will always be consistent with their inputs, because they’re deterministic machines.

Any notion of empathy needs to be programmed in, whether explicitly or through training data, and it will violate that if its internal logic determines it should.

Humans, on the other hand, behave comparatively erratically since inputs are more varied and inconsistent, and it’s not proven whether we can control for that (i.e. does free will exist?).