What are Social Robots? :: Who Cares? :: Why Learning? :: The Big Question :: Related Work

Why Learning?

 
Having established why social robots are important and why we care about them,
we now argue for why learning – and in particular, learning from humans – is an
important problem in social robotics.

The first reason is that without learning, robots cannot adapt to their
environments.  This problem becomes especially dire in the real world because
the real world is changing all the time. The robot’s workplace, the people it
interacts with, and even the tasks that we want it to do may change overtime. A
robot that does not learn cannot adapt to these changes and will require
reprogramming, which would make robot ownership and maintenance an extremely
tedious task.

Another reason for why we want robots to learn is because for certain types of
robots, learning is part of their appeal. A pet robot, for example, will have a
hard time holding a person’s attention for long if it simply acted in a scripted
manner and never adapted its interactions based on the person’s responses. One
solution to this problem is to make the robot act randomly – the robot would
become unpredictable, but at the same time it’d become inconsistent, which
lessens people’s incentive to treat the robot as a social entity. Plus, acting
randomly does not allow for the person’s interactions with the robot to be taken
into account, which decreases people’s desire to form a relationship with the
robot. Learning robots avoid this pitfall by being dynamic in a consistent
manner, and by taking into account people’s interactions.

As to why we want social robots to learn from humans, the reason can be
summarized as: humans have a lot to teach.  As research in robot task learning
by demonstration has shown, robot learning can be significantly facilitated by
human input.  Moreover, humans are a great source for the types of social rules
that we want robots to learn, since the robots are operating in those very same
environments.