Researchers at the Human-Robot Interaction Lab run by Tufts University are teaching robots how to disobey direct orders. In this video, one of the researchers walks a robot through a series of commands including "walk forward." The adorable little robot notes that will put it in danger because there is "no support," so it hedges, effectively telling the roboticist no. After the researcher assures the little robot he'll catch it, it cheerfully walks to and over the edge. Fortunately for human-robot relations, the researcher is true to his word, though it does elicit a quick "Ouch!" from the robot.
The team believes that teaching robots to decline orders improves their reasoning, which will help guide AI through contradictory or confusing directions from their human counterparts.