Improving AI With Baby-Brain Power
The ultimate goal of artificial intelligence research is to make a computer with the ability to learn, hypothesize, and grow just like a human does. Typically the researchers involved are adults, so they give the AI the same cognition they have, but now some are looking elsewhere for inspiration, and for good reason. The brain of a human child is the best learning machine we know of. A well-known example of this is the ease with which children can learn a language compared to an adult.
Researchers at the University of California, Berkeley have been studying how children learn in a variety of ways, with the hope of giving computers the same abilities. One of the tests involved two jars of lollipops. The lollipops were either pink or black, and the jars contained a mix biased to one color or the other. A lollipop was removed from each jar and hidden, so the infant could not see which color it was. This was to make the infant have to rely on the information from the jars about which lollipop was the one they wanted. In most case the child crawled to the lollipop from the mostly-pink jar, as they wanted the pink lollipop.
Another test involved a toy that activated differently, depending on the color blocks set on it. A red block would make it light up while a green block would make it spin. A blue block could do both. Despite being told that information about the blue block, the children still often came to the appropriate conclusion that it was the more common red and green blocks being placed on the toy together that would make the toy both light up and spin.
This kind of exploratory and probabilistic reasoning the children use to learn is what the researchers hope will allow for more advanced and more AIs. After all, if we want artificial intelligence to be similar to human intelligence, it makes sense to get AI to the same place we were as we grew up.