Machine & Infant Learning

My 7 month old daughter started talking just recently.

Okay, so we can't understand what she's saying, but she's got the cutest way of saying it! It sounds a bit like "gnab gnab blah", repeated over and over again. She's clearly processing our speech and working it into her neural pathways, trying to reproduce it.

The whole process of infant learning has been wonderfully fascinating to watch. They come into life with very little other than an amazing capacity to learn and integrate what they observe from their environment, and work their way upward from there. There's a field of AI that maintains that's the way we should teach computers to think, too, by building in the ability to learn, then putting them into an environment and teaching them how to solve a series of problems.

The main problem with that is our limited understanding of how infant learning actually takes place. The traditional view of an infant is that they're relatively stupid, and learn slowly. Having observed my daughter as she's gone from infant to baby stage, I can say with confidence that infants are incredibly smart and able to learn quickly.

There's been some research going on in this area. There's an article on ScienCentral News about research done in how quickly infants can learn, while others have done research showing that infants have complex emotional responses from the start (sorry, I've lost my link to that one).

What all this means for machine learning is that we can only scratch the surface of what's possible, since we don't understand the complexities of infant learning yet. Theoretically, if we understood infant learning completely, we could then model that in a computer to create a computer that could learn how to interact with its environment.

Anyone want to raise a baby computer?