DARPA is teaching some manners — to robots

By Gary Elinoff, contributing writer

Researchers at DARPA are seeking to create smoother relations between robotic systems and humans, and because nothing enhances communication like good manners, they’re developing a kind of “finishing school” for robots. The hope is that our artificial friends may be empowered with what amounts to intuition with which to guide their interactions with us flesh-and-blood types.

According to Reza Ghanadan, the program manager at DARPA, “If we’re going to get along as closely with future robots, driverless cars, and virtual digital assistants in our phones and homes as we envision doing so today, then those assistants are going to have to obey the same norms we do.”

Establishing norms

Toward this end, members of Ghandan’s group have decided to teach robots some manners. The hope is that then, they will be able to judge what’s proper to do and what isn’t. They’ve done this by studying what humans recognize as proper social norms.

So what are these norms? They start with the simple things that we all do, or at least that those of us with good manners do. Examples include silencing a ringing phone and speaking quietly in the library, knowing when to say thank you, and so on — all of the small things that most of us learn, little by little, starting in childhood.

DARPA is establishing a charm school for robots. Source: DARPA.

The overall aim of the DARPA-funded project involves creating a representational framework of the norms in human behavior and to study how those norms are registered in the human mind and what triggers them. The next step will be to model the results in a way that that can be impressed onto machines. Finally, a machine-learning algorithm that allows the machine, when presented with a new situation, to use what it has already learned to react properly in that situation. The machine will then have created an entirely new norm, one that it wasn’t explicitly taught.

Creating new, conditional norms

According to Ghanadan, the trick will be to impose norm processing onto computational architectures. There are an enormous number of these norms that most of us follow without being aware of doing so. Some of them are universal, and some are relevant only in certain situations. Some situations bring on a whole host of them, even if they don’t relate to each other in other contexts. And sometimes they conflict with each other. They are continually being updated, modified, or dropped.

Of course, robots can already do quite a lot, like guessing what email message is spam and deducing which X-ray is to be tagged for a physician’s immediate attention. But to get beyond that point, robotic systems will need to be able to co-mingle, activate, and respond to established norms, as well as learn new ones, and know when to apply them. Just as moral human beings do.