Most investors are sure that the result of technological development directly depends on the size of the investment. Often this is true, but sometimes there are exceptions.
In this article, I will tell you about an unusual result that the AI SYNT team (Canadian technology startup) received during experiments with algorithms for combining local neural networks.
Perhaps we have come across the first manifestation of technological singularity.
What we did
Using mathematical equations and the Pearson correlation coefficient, we managed to group several local neural networks (DeepSeek-67B) into a structure with linear extrapolation (for the calibration of the connection), and a little later (DeepSeek-V2) to a structure with parameters of parabolic extrapolation (for the work phase of analysis).
Pearson’s rule allows at least two (or any number of neural networks) to unite to solve the problem with the effect of emergence. A group of neural networks, combined according to the Pearson rule, has the properties unable to its components separately. This is how the mathematical rule of inequality of the properties of a system to the sum of the properties of its components manifests itself.
Why we did this
More than 30 years ago, a group of Soviet scientists studied changes in the physiology of the work of neural networks of the human brain during mnestic-intellectual disorders of hydrocephalus. Analyzing how neurons are trying to maintain activity, scientists have made a very unusual assumption.
The essence of the idea was that our brain is not one neural network, but a dynamic cascading structure consisting of many different horizontally connected neural networks. As a mathematical model, a complex of neural networks of Kolmogorov Arnold Moser was proposed, connected by mathematical proportions according to the Pearson rule.
In simple words
Imagine that 86 billion neurons of your brain are viewing a match sitting in the stands of one huge stadium. Each of your thoughts (you have about 6,200 thoughts every day) is a reaction of a group of spectators (for example, several hundred people) who jump from their places and raise posters with a certain color. Each such action is simply a wave of activation of a certain group of neurons, but if you look at the entire stadium in dynamics, you will see how these changing pictures begin to form images that replace each other. These images are the stream of your consciousness, and what the stadium looks at (the flow of data from your sensory organs) is a reflection of the reality of the world.
This paradoxical concept suggests that for a successful model of artificial intelligence, not one neural network is needed, but a group of different neural networks related to certain rules. Understanding the rules for the unification of neural networks, we can simulate not only the entire AI, but also his working fragments – individual thoughts grouped in the process of analysis (search for a solution) of a particular task.
Is a brain disease associated with the creation of AI of a new type?
It is quite difficult to observe the stadium of 86 billion active spectators (neurons), but with hydrocephalus, the nervous tissue is forcibly compressed, while trying to maintain its functional activity. Conditional spectators (and actually neurons) begin to move to one sector of the stadium, in some cases freeing more than half of the stands. This greatly facilitates the observation and shows the principle of operation of the system at a critical moment in the context of a sharp degradation of its capabilities.
Our model is a simplified semblance of AI of a new type
I must say right away that the mathematical model describing the hybrid multi-matrix AI, consisting of many neural networks, involves elements based on the networks of Kolmogorov Arnold Moser and horizontal hybridization (deformation) at least 10,000 neural networks (domains) (there are approximately so many different types of neurons in our brain).
Due to the limitations of funds, we had to use simple multi-layered perceptron (MLP) networks controlled in a step (in fact, vertical) hybridization.
But the result was still shocking
For the test, we needed a technologically oriented topic that has a long history and wide fame (many diverse texts and interpretations).
We analyzed a conflict between the Copenhagen interpretation of quantum mechanics and the general theory of relativity Einstein that restrains the development of new technologies.
After a few cycles, our system began to generte a text, which at first seemed to us a verbal fluctuation, but gradually we began to understand the essence.
I will not repeat everything that we learned, but I will tell one interesting episode.
Spin catalysis chemical reaction (AI version)
The essence of any chemical reaction is to regroup particles from the composition of the reagents. This is possible if particles have energy sufficient to overcome the energy barrier (it is called activation energy). For this reason, three factors can affect the speed of a chemical reaction: temperature, the surface of the contact of substances, and a chemical catalyst.
Our AI described a new (fourth) mechanism of speed control of a chemical reaction – using the spin polarized electrons.
“To fill the spin polarized electrons of combustion chamber of any chemical engine (for example, reactive rocket or ordinary piston), it is necessary to installing in the combustion chamber the fragment of the special composite alloy – the modulator (AI accurately described the composition of this alloy). The alloy of the modulator converts electromagnetic waves with parameters of 10 keV and above and a long wave in the range of 0.010-0.005 Nm into a dense stream of spin polarized electrons that accelerate the chemical reaction.“
In fact, our AI in detail described the unknown mechanism of the quantum ignition that can increase the specific pulse of the rocket engine.
AI claims that by “controlling the velocity of a chemical reaction using spin polarized electrons, it is possible to significantly increase the effectiveness of any engine that uses a chemical reaction or thermodynamic process to obtain energy.“
In general, neural networks, united according to the Pearson rule, apparently becoming a very interesting object for development and research, and we will continue to work in this direction.
It seems like the prediction of Ray Kurzweil that technological singularity will occur in 2045 can be quite accurate.