How Do We Learn Things

Machine learning is a hot topic today. Based on our hiring experience. Almost all the internship candidate wants to do something related to machine learning. I believe huge bubbles in the machine learning area. It was obvious in the past couple years that machine learning has not progressed much, at least as much as how we expected it would achieve. If machine was not learning as we planned, what’s the gap between human intellegence and machine?

Well, to answer that question, we must first have a reflection on how we learn things. A very simple example is how baby learns to recognize stuff around, such as an apple. Usually the parent takes an apple and swings it in front of the baby and say, ‘APPLE’. Suppose we are teaching machine the same thing, how will a machine parse these information? First it take videos, i.e. a sequence of graphs that contains the apple, parent’s hand, and surrounding environment. As the app was shaked by the hand, computer can make a diff of two pictures and get to know that the movinging stuff, which looks round and red, is actually the object that we are talking about. Then a microphone on the machine will receive the sounds of ‘APPLE’ and make some correlation between the sound wave of ‘APPLE’ with the object that’s round and red.

The aforementioned process seems totally doable with today’s technology. However, the biggest difference is that a baby tends to learn this automatically. But machine needs human code to conduct the learning. In this sense, it’s the programmer of the machine who asked the machine to learn rather than the machine was learning. This is similar to that you are writing with a pen, but you can not say that the pen was writing.

How can we make machine learning things automatically? It would be helpful to see how baby was learning things automatically. It’s something similar to the penalty function. If baby learn’s something, or the baby sees that parent’s are smiling when they can repeat ‘APPLE’, their inner body will generate some chemistry to make them happy. This happy feeling becomes the momemtum that drives babies learning. In general, such chasing for happiness, as well as other emotions such as scary of lose, uncomfortable with hungry are the inner trigger of human learning.

In short machine learns because programmed to do so. But human beings learn because we have emotions.