re: Elon Musk's thoughts on artificial intelligence threat

It's a program, so it will never be "self-aware". To be self-aware you really need to have organic functions with nervous and endocrine system that evolved over millions of years. AI is a few decades old.

However, as any program, it can most definitely kill by algorithmic evolution. It's not that it will "evolve" to think, but it may adapt to input and produce different scenarios that originally programmed.
Most basic program failures is those plane system.s The software that Boeing auto-pilots used, crashed two planes. So putting AI in charge of vital life-depending things is going to be very lazy and short-sighted.