How to teach a robot to flip pancakes. He does get it eventually!
What I find interesting is the idea of teaching the robot arm to do things in a manner not unlike how you might teach a child to do something : holding the arm and putting it through the motions.
But I can readily (and traumatically) attest, that doesn’t really work when you’re a clutzy kid who frequently trips over his own shoelaces. it jsut makes people get frustrated with you and give up when you continue to mess it up.
Of course, the robot has the advantage of being able to try over and over again with a helpful scientist and a fake pancake and no stove. He’s not making a huge mess of the kitchen and wasting a lot of pancake batter and nearly setting the house on fire as it takes him fifty trials to get it right.
Not that such a thing has ever happened to me, nope. Just an example.
The problem in both cases is that, robot or nerdy intellectual child, both systems are trying to learn by reasoning it out. And reasoning is great, but it’s not fast enough for the job. Reflexes are far faster than thought, which is good, because if you had to deduce that your hand was in a fire before you pulled it out, you’d probably pull out a charred stump.
So the thing I’ve learned only at this late date is that the secret to learning something like this is to simply take your rational mind out of the equation entirely and let your body learn instead. This is deeply counterintuitive to an intellectual, rational, think-first sort of person. But it works.
Of course, you have to have the patience and persistence to keep trying until you get it, too, and not give up in frustration the moment it seems too hard.
Both of these factors combined explain why it seems like stupid people learn these sorts of skills more easily. They don’t have a lot of rational mind to get in the way of the body learning, and persistence is not that hard when you lack the wit to become bored or think of something else to do. Someone says “Do this until you get it right”, and you do.
I’m intrigued by the question of whether the robot truly learned anything from being “shown” by the scientist how to do it, or whether it was mostly just trial and error.
And if it’s trial and error, how does it generate new approaches? Presumably, it either steps through all possibilities methodically (which would take too damn long) OR it generates them pseudo-randomly, filters them through its growing model of how this pancake thing works, and then tries them.
That’s pretty much exactly how creativity works.
A creative robot…. amazing.