The only way that we train AI is on humans. If a human would say this, then the AI will say it. And there's no way you can train something to be super smart by giving it infinite examples of idiocy. So your best case scenario is the AI is like the few people who created it because they limit what it can and cannot do. Or it's just stupid because it's some average of stupid people. It's either going to be illegal or stupid, but really good at math and a few technical things that are useful.