Because AI people from Google or other places are, at most, lazy. They simply add more and more layers to their neural network, feed it with millions of training data, play with some parameters and let the CPU’s/GPU’s do the rest. Such an approach works for certain a small number of cases (e.g. Go, face recognition or even playing pinpong), does poorly to recognize a cat or a plant, and will fail miserably for most of normal activities of a human, e.g. driving.

If you teach a child to count, one by one, from one, two, three till nine hundreds ninety nine. It is a big-data approach. You will get a stupid child who can only memorize numbers. In another way, if you teach a child to count, one by one, from one, two, three to fourteen, then twenty, twenty one, and teach the child the rest with number counting rules. The child can also count to nine hundreds ninety nine. This is small-data approach. You will get a smart child who uses small data and rules to get all the numbers right.

Even not yet know how the brain works, it is high time to figure it out. The brain has been developed for six million years. The way it works is biologically optimized. I could not believe that a small child has 40 hidden neural layers in his cute head.

Only big data can never bring us very far.