Skip to main content

Memory is more dominant than generalization

 Artificial intelligence seeks machines that can generalize very well. When you train some model like a system for image recognition, the goal is to find some general representation of the input, so that the model is smaller than the data required to train it.

From our understanding of human intelligence, it might seem that this is what is going on in the human mind because most of the time we do not recall facts from our brains like computers do, rather we are constantly generating the facts from "models" we have developed as we encounter information in the real world. 

If generalization is what we are really doing all the time then we could have developed a theory that unifies all the forces of nature by now, i.e. a unified field theorem. But the fact that we are still unable as a species to achieve such grand unification as well as smaller unifications like finding the core model that controls prices in the stock market, I must say that the primary function of our intelligence is not generalization but more of memory.

Previously I had said that the human mind is not able to fully reproduce the exact details of the information it has stored and what it usually produces looks like something that was generated rather than recalled verbatim, so if the human mind seems to be generating stuff how is it that we are not super generalizers?

The thing here is that we have confused a patchy memory for generalization ability. It is true that the brain is able to find underlying similarities between different things, which is what we call generalization, but what I am saying here is that this ability is not as developed as we would like to think, because if it was as developed as we think it is then we could have solved all unification problems.

Despite the fact that we cannot reproduce exactly that which we have mentally consumed, we are in no way doing generalization all the time. Most of the time it is all about memory and even low-level creative abilities can be viewed as remixing our stored memories and experiences rather than finding general underlying structures. 

The primary generalization ability of the brain could actually be the fact that it stores information in a unified format, information from varied sense organs is all transformed into electrical impulses which are represented as neural structures. It is because of this unified storage that sometimes we are able to generate effects that seem like we might be doing some kind of generalization because the unified storage of sense impulses enables the brain to mix information and generate information that gives the illusion of high generalization.

I am not implying that there is no true generalization going on in the brain. If there was no generalization we would not have E = MC^2. What I am saying here is that memory is the more dominant ability we possess and despite the fact that it is choppy memory it is in no way true generalization. 

The ultimate goal of intelligence would seem to be the ability to generalize very well and accurately, the ability to see the underlying deep structure between disparate looking things. This is what we are trying to achieve in mathematics. 

Comments

Popular posts from this blog

Next Steps Towards Strong Artificial Intelligence

What is Intelligence? Pathways to Synthetic Intelligence If you follow current AI Research then it will be apparent to you that AI research, the deep learning type has stalled! This does not mean that new areas of application for existing techniques are not appearing but that the fundamentals have been solved and things have become pretty standardized.

How to become an AI researcher

Artificial Intelligence is all the rage these days. Everyone is getting on the bandwagon but there seems to be a shortage of AI researchers everywhere these days. Although many people are talking about doing AI not many people are actually doing AI research.