Why the revolution in neural networks is much more important than it seems

We'll have to start a little further back. For many centuries, the basis of any European (and, accordingly, Russian too) worldview was philosophy. It was an indispensable element in the horizons of any public intellectual. In fact, the post-war generation was the first to break away from this tradition – and it paid off. On many issues, modern public intellectuals, lacking knowledge of the history of thought, appear completely ignorant.

Many concepts that seem banal to us have specific authors. For example, the idea that “nature is written in the language of mathematics,” which is natural for modern techies, was expressed by Galileo. Before that, for centuries, Aristotle reigned supreme, who put forward the thesis that the worlds of mathematics and physics are orthogonal, and physics can only be comprehended by the senses. The simplest formal thesis was centuries ahead of human history.

At the moment, all science is written in the language of mathematics, and mathematics is a strict and harmonious building, built on a long chain of logical conclusions from several self-evident axioms. We know that the entire IT infrastructure is built from a few formally simple rules of discrete mathematics and formal logic.

Now a little closer to the topic. Even though AI is created using IT and mathematics, there is really no understanding of how it actually works. Of course, there are transformer models, various gradient descent algorithms, tensors and matrices, but we have no idea what exactly do they mean. This is definitely not Galileo's model. The human brain and human culture are structured very differently. All the explanations I have been able to find about why calculating the statistical probability of the next token in the chain leads to an imitation of human thinking, which is very effective in many areas, come down to the concept of “emergence”. But this very emergence hardly appears either in physics or in philosophy. In fact, the explanation “because it happened that way” would be no less convincing.

In fact, quite unexpectedly, we have a new model of cognition, a new model of thinking, previously unknown to people, and therefore the consequences of its emergence are unknown to us. Just as Aristotle’s thoughts stopped physics for a thousand years, and Galileo’s thoughts, on the contrary, revived it, AI will bring large-scale and completely unpredictable consequences, since its thinking is a black box for us. We can adjust certain parameters, but why and how this works is unclear. Just like we have a DNA code, but we only partially understand why heredity is coded the way it is, and we are still very far from fully designing a living organism.