Beautiful mind vs sterile formula, the pathway to better AI.

A way of thinking about a particular concept is called a paradigm. It is a model, or a framework, that shapes our perspective in order to interpret the world.

People, until we acquire a paradigm for a particular way of thinking, tend to ignore the facts. They do not see it, or sometimes, even worse, they see it through the lens of their past.

Reading hundreds of books and studying massive amounts of varied materials gives us a mental library of various, sometimes contradictory, paradigms that we can choose based on the probability associated with circumstances for a given situation.

This is the difference between a “hammer” seeing only the nails and a beautiful mind.

In the real world, we get many millions of inputs every second, our brain is very fast at selecting the most important signals which are then applied to paradigms that we have developed via learning. But, that is not the whole story.

In sciences, the method of finding the pieces of data that best characterize a certain phenomenon is called Principal Component Analysis (PCA).

However, filtering a multitude of signals into a sterile formula carries a risk.

In our minds, there is a constant humming of the background ideas that we might not be conscious of at all, but which skew our decisions. These could be the result of our past learning, and experiences.

The risk of oversimplification is significant in Machine Learning, too.

When we find concepts that are mathematically reducible to a simple principle or an equation, we prefer to use them. They are easier to implement and validate. This is especially true in most autonomous driving companies. The formulas seem safer, but the question is, are they really safe?

The recent developments in Natural Language Processing (NLP) and specifically in ChatGPT taught me a lot about what I intuitively believed to be true all along. Namely, the importance of fuzzy background information in addition to principal data components.

It is really worth having a chat with AI on the ideas that are controversial in nature, for example on some new scientific discoveries that are not fully accepted by the larger body of scientists. The AI agent (i.e. ChatGPT 4) navigates the intellectual minefield gracefully and is able to hold multiple conflicting ideas at the same time, a sign of intelligence, some say.

By the way, I do not care about the discussion on the subject of whether today’s AI can be considered intelligent. Such conversations miss the point and are boring to me.

In summary, I believe that the formulas and principal components are very important, at the same time I like the approach of providing the AI with broader data to make their probabilistic decision.

I also believe there is a place for strict formulas in some computations, whereas we will have to teach autonomous cars how to drive and how to recognize probable risks along the way.