In this article, I will tell you why the technological singularity has already arrived despite the fact that most of us still do not understand that scientific and technical progress no longer belongs only to the human brain. Using a specific example, I will show how artificial intelligence creates new technical solutions and creatively overcomes the limitations of the human brain.
AI is not a search engine
Most of us look at AI as an analogue of a search engine with the function of analysis and dialogue.
AI and search engines are fundamentally different technologies.
A search engine does not just search and give us links to pages with the information we are interested in, but also forms the order of issuance (in fact, it builds the parameters of information availability).
Generative AI (despite its ability to find information) is a complex data approximation system – a mathematical mechanism designed to create new and unique answers associated with a hint and the so-called content window.
Strictly speaking, generative AI is not exactly a program (in the usual sense for us), since instead of executing pre-written instructions, the AI system can adapt to new conditions (arising during a dialogue) and, as a result, it can achieve an unexpected result.
It is even more important to understand that AI and a neural network are not the same thing
AI is a general term for all technologies that are in one way or another similar to the abilities of natural intelligence. And a neural network is a specific tool in this technology, designed to recognize and use patterns and similar parameters in different data sets.
This means that when we talk about AI, we are talking about a set, which, in addition to the tool itself (the neural network), also includes us (our brain) as a full-fledged participant in the dialogue. For this reason, the product of AI – and this is the information that the neural network creates as a result of the dialogue, depends not only on the operation of the computer system, but also on how we, people, build our interaction with the machine.
Therefore, the efficiency of AI is an operator-dependent factor
What AI will create largely depends on the operator and their ability to manage the dialogue.
Discovery Machine
The process of creating new technologies can be described mathematically as a function of extrapolation and approximation. At the same time, parabolic approximation and extrapolation (which are often used in machine learning and data analysis) allow you to process and create nonlinear dependencies (which, in essence, is a pure and refined creative process).
Thus, if the context window forms an environment or a set of data, then the dialogue with the generative neural network, built on the points of parabolic extrapolation, gives data beyond the initially known values. In a general sense, extrapolation is a method of extending conclusions based on known and proven facts to new and unknown data.
It is important to understand that the result you get will depend more on your ability to prepare and manage the neural network than on your knowledge in the scientific field under study.
Here is a concrete example from my experimental practice of using a mathematical model inside AI.
I am omitting the mathematical part that formed the basis of this experiment and will show you only the general scheme of the dialogue.
Step One – Create an Environment (Context Window)
Controlling the Rate of a Chemical Reaction in an Extended Two-Domain Model of Quantum Mechanics. Application Area – Chemical Rocket Engine.
Step Two – Generates Parabolic Extrapolation Points
1. The mechanism of spin catalysis: using spin-polarized electrons to reduce the activation energy during rocket fuel combustion.
2. Injecting electrons with aligned spins as a way to influence the reaction rate by changing the underlying quantum paths.
3. Since most of the spins are aligned, the resulting spin polarization density changes the potential energy landscape of the reacting gases.
4. The solution scheme should effectively reduce the activation energy by ΔEₐ ∝ –λ⟨S_z⟩.
Step 3 – Obtaining the solution (new data set)
• The photoinjector integrated into the combustion chamber uses a multi-alkali photocathode (KNa₂Sb).
• When exposed to coherent radiation of 0.010–0.005 nm (~10 keV), the proposed cathode emits spin-polarized electrons with ~50% polarization.
• The resulting electron bath affects the reactant molecules, allowing spin catalysis to accelerate combustion.
Note that AI uses a highly efficient, but exotic for such a task, multi-alkali photocathode, which is known and widely used in night vision devices, but has never been used in rocketry before.
It should be noted that this solution is given with mathematical calculations and very precise argumentation (there is even a comparison of efficiency with another GaAs-based technology). At the same time, if we discard the preparation and step-by-step dialogue (just ask how to upgrade a rocket engine), the neural network’s answer will be a general reasoning, devoid of valuable innovative details and mathematical justifications.
Conclusions
We are entering an amazing period when the ability of one researcher to manage a mathematical model inside a neural network can incredibly accelerate the process of creating new technologies. Now, one researcher skillfully managing a trained AI can be creatively stronger and significantly more efficient than an entire research institute with hundreds of highly paid employees. At the same time, the technical solutions that will be created in such an unusual collaboration (neural network + person) will initially be unknown to the neural network itself and may be generally beyond the understanding of the operator, a person.
And this is not the distant future – all this is realizable right now on existing AI models.
What is this if not a technological singularity?