With all respect to Dr. Hawking, his article gives away a very superficial understanding of where AI currently is, and most of his fears aren't justified.
The best point he makes is about being cautious when it comes to what vital functions we turn over to computers. The idea of autonomous weapon systems that choose and eliminate targets without human intervention is certainly disturbing.
But the idea that computers are becoming too complicated for us to understand is rubbish. And they'll out-invent us? Of course not. Computers don't invent anything. They never have.
Hawking is a brilliant physicist but he should maybe avoid opining about AI since he doesn't seem to know much about the field at all.