“Here’s an example: in hospitals, a delivery robot could pass out towels and medication. If it were to get stuck leaving a room, the machine could call a command center where a human technician would figure out the problem and free the robot. Here’s the key: every time a person made a fix, the robot would keep that new information and use it to perform differently the next time it leaves the room. With humans in the mix, this robot could learn from its mistakes and continually push toward 100 percent autonomy.