Did sama not simply consider that, by the time AI becomes so powerful as to be dangerous, we might also be able to program an AI with the opposite function, ie 'save humanity', and watch it fight the 'bad AI'?
This is an important counter to this strange fear-mongering trend. Whatever advances a supposed intelligent machine can use, people will first put it to their own best use - including defense, not from autonomous machines, but from other people using these advances. We'll have a lot of experience and good tools to protect ourselves.
