Deep Currents

The recent developments in neural networks are making Ray Kurzweil’s “Singularity” look closer and closer every day. Along the lines of the self-aware starfish that made a splash at the TED conference last year, researchers in Leipzig recently released an animation showing three virtual robots — a dog, a pig, and a humanoid — who explore the movement of their bodies and the space around them like a newborn baby. They have no specific instructions except to try to predict their own movement: if I put a foot here, and the other one here, then my body will move forward. The video shows the robots exploring nearly every possible motion and eventually figuring out some of the basics of how to walk. They weren’t told to walk, but they figured it out on their own. This is a radical departure from traditional robots that are given precise instructions for how to carry out a task. These robots won’t fetch you a drink or drive your car, yet, but they have the beginnings of what differentiates every human child from the machines we all use to do our bidding: the inclination to try new things and the memory of how to do what worked. The head researcher calls the software a “plug-and-play brain.” Just imagine what these robots could do with treads that climb walls, or if they were attached to your own brain as a kind of independent-minded appendage.

Written: August 7th, 2008