This interesting TED talk comes to us from a fellow named Hod Lipson. This is a man studying how robots can evolve and learn like biological systems. In my opinion, this is one of the most interesting fields right now. The general idea is to see how software in certain circumstances can either learn through trial and error, or evolve through random mutations much like organic organisms. This approach is different than the previous plan of trying to hard code artificial intelligence; we're letting it evolve "naturally" instead of trying to figure everything out ourselves.
As Hod Lipson shows us, this new strategy is coming a long way. Already he's been able to create robots that can learn how to walk and self-replicate. Who knows what the future holds?