A Simple Experiment Shows How Hard It Is To Get Robots To Behave Ethically

If we want robots to be the complex autonomous agents promised to us in sci-fi movies, they need to be able to interpret the (often complicated) subtexts of situations they might encounter in the course of carrying out their duties.

Footnote reports on an interesting experiment that explores a robot's "ethical" potential:

A large robot comes out of an office mailroom carrying a package marked "Urgent" to deliver to the boss upstairs. After navigating down the hall at maximum speed, it discovers someone is already waiting for the elevator. If they cannot both fit in the elevator, is it acceptable for the robot to ask this person to take the next elevator so it can fulfill its urgent delivery duty? What if, instead, there's someone in a wheelchair already riding the elevator? Should the robot ask this person to vacate the elevator to make room for it to complete its delivery task?

These are pretty straightforward problems for a human to solve. Purely by virtue of being alive and recalling past experiences, one can generally draw easy conclusions on how to handle these situations. But this is much less true for a robot that only knows how to do what you teach it to do.

Since robots are programmed by individuals (and each individual might very well have a different response to these situations) you can see how complex it becomes to teach a robot to behave "ethically." A person waiting for the elevator ahead of a robot making an urgent delivery may not object to letting the robot take the elevator first. Alternatively, if a person in a wheelchair is already on the elevator, they're probably going to feel less pressure to give up their ride to a robot, regardless of its urgent delivery.

AJung Moon and others at the Open Roboethics Initiative decided to solve this problem by involving "various stakeholders - industry members, government policymakers, users, and academic experts - in the design process. By incorporating their perspectives on robot behavior into a machine-learning algorithm, our research explored an empirical method for designing behavior in situations where the ethical action for a robot isn't always immediately clear."

Put another way, experimentalists took a survey of what various people thought would be the most appropriate way to handle the robot-and-human elevator scenarios described above - when the robot should yield to the human, when the human should give up a spot on the elevator to the robot. By and large, the robot was instructed that it was most appropriate to engage in a dialogue with people it encountered in the elevator scenario. This data was incorporated into a behavioral model for the robot and the following video was filmed of the experiment:

Here's how this experiment boils down: ethical behavior is more than "a set of inflexible rules." It's behavior that depends on context and a number of environmental factors. It's most likely that advanced "ethical" robots will not come about by developing some sort of hardwired ethical system, but instead will come from "the development of robots that can better communicate and negotiate with the people they encounter to reach mutually agreeable outcomes."