Tuesday, January 26, 2016
Minsky RIP
Marvin Minsky was among the pioneers of artificial intelligence, but his BIG idea never got developed properly. He was working with ANALOG devices that directly simulated neurons. All the money is in digital supercomputers, so the analog line has been largely abandoned except for some work by Carver Mead. A lot of the work on Perceptrons and similar devices in the '50s and '60s was in Russia. I wouldn't be surprised if it continued and advanced quietly; Russian military and intel types have wisely hesitated to adopt overcomplicated and hackable digital methods.
The digital version of AI ... software that can adjust its own code and constants to work better ... can do lots of useful tasks, but isn't really intelligence by any proper definition. No matter how big and fast the CPU gets, it still can't approximate the near-infinite layers of SIMULTANEOUS AND SMOOTH feedback in an animal or plant. It's self-defeating. Can't have both. A digital system needs more hardware and code to approach smoothness. The more it tries to approximate analog behavior, the more it sprawls, requiring longer transit time between its pieces.
Labels: Carver, defensible spaces
¶ 6:41 AM
