It's doing something else than the most logical deterministic assumption, so clearly it must be trying to be "smart" ?

How often does a badly implemented algorithm that should in theory just work, get labeled AI because in practice it returns opaque results occasionally?