A robot is only as smart as its programming. Learning on the go has been the sole purview of living things.
That was, until a team of scientists at UC Berkeley programmed a robot to learn simple tasks through trial and error, just like humans do.
The robot itself, a Willow Garage PR-2, is not new. But researchers applied a relatively new form of artificial intelligence, known as Deep Learning, to give it a kind of primitive learning ability.
With it, the robot or BRETT, (Berkeley Robot for the Elimination of Tedious Tasks) can use visual and sensory information about itself, its environment and the objects before it. It uses them like LEGO, building little neural networks of information, basically figuring out how to do something (how to put two real-life blocks together, say, or put a ring on a peg).