HONOLULU — The notion of ownership comes so easily to humans that even preschoolers have got it down. Robots, on the other hand, often struggle to grasp such abstract concepts.
Now researchers have programmed a well-mannered robot that can learn who owns what, as well as what it’s allowed to do with people’s belongings.
Teaching robots ownership-related etiquette “is really, really important,” says Matthias Scheutz, a computer scientist and human-robot interaction researcher at Tufts University in Medford, Mass., not involved in the work. You can’t just send out robots “who are blissfully unaware of who owns what.… If I instruct a robot ‘build a fence,’ and it goes to the neighbor’s and starts [stealing] boards, that’s not what we have in mind,” he says.
The new socially conscious robot, to be described February 1 at the AAAI Conference on Artificial Intelligence, can learn who owns what from explicit statements as well as its own observations. The robot learns its code of conduct from direct orders and generalizing from specific examples. For instance, if the robot is told not to touch several objects that it knows belong to specific people, the robot deduces that, as a general rule, it shouldn’t touch owned objects.
Artificial intelligence researcher Xuan Tan and colleagues at Yale University tested the robot’s manners in experiments with blocks on a table. In one session, Tan played with only the red blocks, leading the robot to infer that these blocks belonged to the same person. When Tan instructed the bot to throw away everything on the table and the machine reached for a red block, Tan stopped the robot, saying, “That’s mine.”
Now aware that it should not chuck Tan’s belongings, and assuming that the rest of the red blocks belonged to Tan as well, the robot cleared the table of everything but red blocks. Later, when study coauthor Jake Brawer directed the robot to throw out a red block, the bot replied, “Sorry, I am forbidden to throw it away if it is owned by Xuan.”
Robots may have more difficulty discerning who owns what in scenarios cluttered with far more objects of much wider variety than blocks on a table. So “the extent to which [this] really would scale up to more realistic contexts is an open question,” says Scheutz. But he called it a good “first stab” at imbuing robots with an appreciation for ownership.