Why would we program AI that wants to harm us? Because we might not know how to do otherwise.

https://www.cold-takes.com/why-ai-alignment-could-be-hard-with-modern-deep-learning/