In the last year since I became a parent, I’ve had the wonderful opportunity to observe how my daughter acquires new skills and learns about the world around her.

“Boundary seeking” is how I think of her main strategy for learning efficiently.

When she first encounters a new object, she pushes it to its limits. One day last month she found a tote bag and played with it for the better part of an hour. She dragged it around the room, she whipped it around her body as fast as she could, she put her toys into it, she stepped on it, and of course, she put it on her head, which she thought was hilarious. (Don’t worry, the tote bag is made from a breathable mesh and she was totally supervised the whole time.)

By sampling over a large distribution space, she was able to find the most interesting things that she could do with the tote bag. She got lots of information about the way that the tote bag behaves under a wide variety of conditions. As she discovered new, interesting behaviors (like putting the bag over her head) she spend more time exploring them.

This is a very efficient way to learn. Essentially, you sample as big a space as you can, then focus on the areas of that space that are most interesting to you. My theory is that these “interesting” areas of the space represent boundaries on a manifold, where one behavior changes into another. (Think, for example, about how the bag might be good for carrying things when you carry it loosely, but when you swing it around your head it makes a whooshing noise and it’s more fun.) The most efficient way to explore a high-dimensional space is to move along boundaries, then use simple predictive heuristics to fill in the gaps between boundaries.