OnIntelligence

How does a computer learn?

Since I just recently bought the book OnIntelligence by Jeff Hawkins and have not really started reading, it seems to be about time to share some of my understanding of the learning patterns of the brain.

So far I have spent some thoughts on how a computer could learn, by learning I am also talking about recognizing patterns and taking these to predict future events or in a simple case words.

This simple case is where I always start off with since it does not seem to be to far away from the first things a human learns. Apart from walking, which I cannot use since I am missing some essential hardware. So understanding how the brain learns by understanding how it learns language would be a good approach.

The first time I thought about the learning process of the brain in computational terms was during my time in High School (well since I live in Germany: Gymnasium) as we had an in class discussion about the evolution of language and we also talked about how computers would be able to learn and if they would be at all.

My final solution on this looked at that time just the same that it does now. It even comes up with a way to predict the next words and it makes it clear why it is difficult for people who have learned a lot have a hard time learning new stuff.

The solution looks somewhat like storing every newly learned word in the memory (may be represented by a MySQL database) and connecting it to a number that represents the number of times the word was ever recognized, this number will be initialized to 0. Every time the word is read/heard after that, the number in the memory will be increased by 1 so if specific words come up more often they will be the first to be accessed. For example the word “hello” might be used most times, so upon creating a new sentence with a new person it might be considered to be an essential part of sentences.

The next step would be to link the words to other words within a sentence. This is probably a bit more difficult but can be done within databases. For each pair of words that was used by someone the AI (well programmed brain) would have to save a new connection between these words. So basically saving those two indexes of a database in another table.

Basically we have just tried to create a brain. So far it can only save words but not really speak any usefull sentence. There is one thing that people talking about an AI never really leave a comment on. Every human being, may it be a child or an adult, is being tought things. In the early years of our lives we even have a system to teach children to understand grammar and vocab. It might be in mathematics or in some language.

So to have a functional programmed brain, one will have to teach it. For human beings we spend about 10 years on education. So to get the same results we would need about 10 years of education to teach the programm about the same amount of things.

This is the main thing to think about. Once someone creates a programmed brain (it may or may not be based upon my model) we will have to teach it the basic things. Maybe it would be usefull to send them to school, although this might a bit weird for the other students.

This is what I think of how a computer learns. I hope some of you have come across some new ideas but I will now go ahead and read OnIntelligence to find out what thoughts Jeff Hawkings had on this topic.