Robots that learn like infants - a path to true artificial intelligence?
After decades of research in Artificial Intelligence,
the most intelligent machines are still limited to very narrow
tasks they have to be specifically designed for.
Humans, in contrast, can perform evolutionarily entirely novel tasks easily.
I will argue that it is not the proficiency in complex tasks that
makes us intelligent, but the ability to learn them without being built for it.
The idea of Developmental Robotics is to let robot learning start
where also human learning starts: at an unskilled baby stage, during
which the robots explores, socially interacts with people, and learns.
I will discuss how this approach could lead us to a better understanding
of our own intelligence as well as general artificial intelligence.
As an example I will show how insights from infant psychology helped
to design highly efficient machine learning algorithms for the control
of a biomorphic robot arm for physical human-robot interaction in industrial
scenarios - and how the success of this work could inform developmental
Note: The robot in the picture is the one Matthias uses in his research example. He cannot bring this one but will bring a small humanoid for illustration and interactive purposes.