Natasha Dantzig of TED (Technology, Entertainment, and Design) writes,
"We've released a new
talk by Hod Lipson (Flash video format). He demonstrates a few of
his cool little robots, which have the ability to learn, understand
themselves and even self-replicate. At the root of this uncanny demo is
a deep inquiry into the nature of how humans and living beings learn and
evolve, and how we might harness these processes to make things that
learn and evolve." For more see the Cornell Computation Synthesis Lab
website, the Talking
Robots interview with Hod Lipson, and our previous story on resilient introspective
I think the best way forward will be a combination of design and evolution. Even within simulations on powerful computers evolution is a slow process, and so far Alife research has failed to deliver on the promise of being able to evolve significantly intelligent organisms.
Smarter evolutionary search methods, such as the MOSES algorithm used within Novamente, which try to characterise the fitness landscape could be useful for producing very adaptable machines. As is mentioned in the video the best search methods involve forming hypotheses in a non-random way (like a scientist or a detective) which can be used to quickly eliminate large areas of the space of possibilities.