I never understood the fear mongering behind ML systems becoming self-aware. My first research project in Uni was on ML and I had the revelation that although powerful, they were utterly stupid, so I switched to PL. 10 years later, verifying DNNs, my god are they still stupid.

I've always wondered if they really need to be smart or self aware to wreak havoc, however. Theoretically, all you need is a system that can learn operate outside of its parameters, right?