Auto companies are pouring billions of dollars into artificial intelligence research, largely with the goal of creating safe self-driving cars. How much autonomy to give AI remains, however, an open question. At one end of the spectrum, it’s possible to give AI complete control over all perception and driving decisions — AI directly bridges sensors and steering. This “end-to-end AI” approach has been successfully demonstrated by Nvidia Corp., and evangelized by computer hacker George Hotz. At the other end of the spectrum, AI can be used not at all, or in limited ways to augment elaborate software. It is this “if-then” approach that drives Google Inc.’s cars.