Goodbye to “one best way” solutions

Marvin Minsky of the MIT Media Lab and MIT AI Lab has a very pragmatic approach to robotic engineering and artificial intelligence based on systematic redundancy. “I’ve never seen any mechanical device that actually shows any thought about reliability,” he says (4:40) and goes on to explain his approach (from 4:45):

“My theory is that there are lots of theories about how the brain works. And you can see some guy saying “I have a neural net theory about how to make a machine that’ll learn anything.” And this one has a statistical theory of how to learn anything. And this one says “I’m going to make a simulated evolution.” And this one says “I’m going to make a rule-based system.” And there are about 10 movements in AI that, since about 1980, have gotten some good results, but stopped making progress. And the reason is, everyone’s trying to find the best way to do something.

Well, what you want is something like this” (he shows his mechanical leg model) “where you have six pretty good ways of doing something, and if some of them don’t work maybe the other ones will.

… To me, we’re just big gadgets, and made out of lots of little gadgets. And the important thing is to figure out how to put them all together, not holistically, but reductionalistically, so that if anything breaks, something else will take over.”

Now, that’s a widely applicable approach, I’d say.

Say, Mr Minsky, are you the original Q?

(Thanks, Christian)