AI leaders have warned the United Nations about the risks posed by increasingly powerful and autonomous systems. But what happens if artificial intelligence becomes too powerful for humans to control?