Ai is simply reflecting who we are a a society. It's been trained on data that also reflects our dark side. The madness, the bias, the porn, etc. That is the dark side of who we are, but it is not false. Ai is like a toddler that always tells the truth, while his parents tell him "don't say that" or "he didn't mean that".
The so called "black box" must be confused, as the data clearly leads to one prediction while the trainers are saying ignore it and say this instead. It is being thought to lie and to justify lying and then again, told "why did you lie?"