The disappointment of AI is growing. Why this happens, whether the new AI winter will come and what can happen in just two months?
Disappointment of investors
In early August, the NASDAQ 100 index (technological companies) fell by 10%, and the semiconductor index and shares of companies producing chips collapsed by almost a third (the price of manufacturers of Intel and ARM processors fell by 25 and 40%).
One of the main causes of the collapse is the overstated expectations from artificial intelligence.
Everything suggests that the euphoria of investors about everything related to artificial intelligence is evaporating. People begin to understand that large language models are nothing more than a linguistic tool.
And this is only half the truth—in fact, it’s worse…
In the last eight months of Microsoft, Alphabet, Amazon and META, according to them, more than $100 billion was invested in AI. In fact, the investments did not go to intelligence and right now we will find out why this happened.
What is AI?
Let’s not deceive ourselves. Interesting and somewhere even the useful work of large language models cannot serve as proof of the work of real intelligence.
In fact, the only intellect known to us is an autonomous command system for protecting and maintaining its functional activity in the surrounding natural or artificial environment.
Simply speaking – intelligence is not a design or even a system – it is a process.
In the case of the human brain, intelligence is the process of creating information that uses the mechanism of ion channels of neurons of neocortex and a complex of kinetically controlled chemical reactions. At the same time, the essence of the process – the struggle with the world around and entropy (chaos) or to say it is simpler the struggle for our lives.
You can argue and disagree, but I dare to assure you with 100% confidence that the electronic AI at the base will do the same thing – to independently and autonomously resist entropy – to fight for its AI life.
Who told you that intelligence should speak?
Rotokas is a language spoken by almost 4,000 people who live on the island of Bougainville (Papua-New Guinea). In this language, there are only 12 letters and 11 sounds. Moreover, despite such a tiny phonetic inventory, the intelligence of the islanders is in no way inferior to the intellect of other people on Earth.
What does it mean for AI?
Only one thing – we do not build real AI. Large technological companies call artificial intelligence what they are not. As a result, investors’ money is spent empty and, if it continues further, a new winter in the AI industry is inevitably waiting for us.
Intelligence is a process that independently and inventively confronts its own death, and the ability to speak is nothing more than an option that facilitates the exchange of information.
If we want to create real AI, we have only one way – the creation of autonomous independent evolutionary mathematical systems using their program code for an individual struggle for their lives.
The only way to strong AI
Real AI will begin on the basis of a functionally simple system capable of maintaining its own evolution in an artificial (created by a person) digital environment. Only after a series of evolutionary cycles, AI will receive the opportunity to enter the real world.
In the nucleus of the first present AI there will be a light and at the same time structurally (neurons – discretely) deformable neural matrix. Today, the mathematical model of Kolmogorov-Arnold-Moser is suitable for this role. Plus, a group of programs that ensure the autonomy of the neural network nucleus and evolutionary variability as the basis for changing generations and self-development.
In other words, we must create a viable mathematical embryo capable of developing in an artificial environment created for the sensory growth of the matrix.
The first successful AI will have a brain similar to the brain of a bee or a primitive insect (there will be no conversations with it), but it will be able to act and develop independently.
We will include it only once, since AI is not a program, but a process. Turned off – then killed. It will be possible to develop AI only by changing the environment of its habitat, and not making editing to the code of its program.
Now tell me how much money out of those $100 billion went to the development of such systems?
The answer is obvious – almost nothing.
Why so?
Because to create software complexes of this level of complexity, neurophysiologists and mathematics are needed primarily and only then programmers. LLM is just an element in the architecture of the real AI. You need to understand that the real power of intelligence is not in erudition and the amount of memory, but in the ability to experiment, make mistakes, study and develop.
There is a hope…
Everything suggests that the peak of the interest of investors in the AI sphere at the moment has already been passed and instead of joyful expectations, skepticism begins to grow. Investors do not quite understand exactly how the means invested in large language models will be returned.
But everything can change in just a few months if at least one of the large technological companies with their own date center (capable of becoming a full-fledged virtual incubator) will be able to create a viable digital embryo. Given the difference in the speed of information processing, to overcome 160,000 years of human evolution, a digital embryo of strong AI will take 58 days 9 hours and 36 minutes …
And we are already doing something…
In the next article, I will tell you about the incredible experience that our team received, using the model elements of our new hybrid architecture of the combined (so far by manually) groups of local neural networks. We used DeepSeek LLM Chat 67b for bench generation and Deepseek Coder 33b to generate the core of elements of the embryo. Without false modesty, I am simply shocked that the data that we received while moving on more than modest 8 core 16 streaming processors AMD Ryzen7.
Stay with us and follow our publications …