What if robots learn to say no?

Could robots one day refuse to follow orders? That is precisely what scientists Gordon Briggs and Matthias Scheutz want them to do. Their goal is to find a balance between obedience and disobedience: if a robot has a good reason for not carrying out orders, it should be able to refuse them.

In the videos below, a robot called Dempster communicated with researchers from Tufts Human-Robot Interaction Lab in a number of compromising scenarios. Dempster had been programmed to calculate the risk of propositions in a very similar way to humans. A number of basic questions, such as “Do I know how to do that?” and “Do I have to do that?” lead on to less basic ones, such as “Does this violate any normal principal?” (such as harming oneself or others).

Put more simply, Dempster is asked to walk off a ledge. After a brief moment to think about it, the robot’s answer is “no”, along with a concrete explanation as to why not. Dempster is also subject to persuasion: when offered a way to survive the fall, the robot changes its mind.

A machine’s ability to think for itself in a compromising situation has been the subject of much discussion lately, particularly when it comes to the ethical choices that would go into self-driving cars, for example.

In this video featuring Dempster the robot, a request to walk towards the edge of the table elicits a surprising response.

In this clip, a robot called Shafer refuses to walk into an obstacle.

In this final video, Dempster is asked to disable his obstacle detection. “You are not authorized to do that,” the robot says.

Have you read?
Who will win – robot, man or woman?
3 ways robots can change manufacturing
Could a robot do your job?

Author: Donald Armbrecht is a social media producer and freelance writer.

Image: A “Nao” humanoid robot by Aldebaran Robotics sits in a corner while its batteries are being charged during a presentation at a branch of the Bank of Tokyo-Mitsubishi UFJ (MUFG) in Tokyo. REUTERS/Thomas Peter