As a result, humanity will receive countless brilliant refinements to already existing theories. But no scientific revolution comparable to the discovery of quantum mechanics or the theory of relativity should be expected…
In conclusion, a few words about what has already begun to radically change the life of humanity — artificial intelligence. It is invading spheres that were previously considered the exclusive prerogative of humans because they are connected with creativity and scientific research.
In late March 2026, one of the world’s most prestigious scientific journals, Nature, published an article describing, at first glance, a rather modest result — a certain method for training neural networks showed no improvements compared to others.
Why, then, did such a trivial result attract the attention of the editors of such a respected journal?
An editorial in the same issue of Nature explained that the scientific work had been carried out entirely — from idea to layout — by an artificial intelligence system called The AI Scientist. The creation of this system by the Japanese company Sakana AI represents an attempt to fully automate the scientific process: from literature review and creating research idea to conducting experiments and writing the article itself.
One must appreciate the scale of this ambition. It is one thing to generate mediocre-quality images, videos, music, or even literary texts for public consumption using AI, and quite another to completely replace humans in scientific activity.
It is no secret that many scientists already use AI for the rapid execution of repetitive or labor-intensive parts of research, such as writing programs for data processing, hypothesis testing, “smart” analysis of possibilities, or literature reviews on a given topic. This part of a scientist’s work, which can take up to 90% of their time, is best automated whenever possible in order to free up time for actual intellectual labor — which, it is commonly believed, cannot in principle be automated.
But the system The AI Scientist, which impressed even the seasoned editors of Nature, goes much further. It automates what belongs to the sphere of genuine creativity: proposing scientific hypotheses and analyzing the results obtained.
The journal notes that The AI Scientist was not only able to formulate a hypothesis and conduct an experiment, but also to write an article about the negative result obtained. Moreover, the work passed an initial scientific peer review for a workshop within a major machine-learning conference. It received high scores and outperformed 55% of papers written by humans.
Of course, one could say that in today’s world the quality of most scientific work leaves much to be desired, since many so-called scientists are engaged not in science in its true sense, but in securing and spending grants. The entire system of scientific inquiry is heavily distorted by serving the interests of business corporations or politicians. Unfortunately, all of this is true. And it is precisely these “grant-driven” scientific studies, which move humanity nowhere, that artificial intelligence is capable of performing far better than people who call themselves scientists.
Recently, the leaders in AI development have been actively creating models capable of analyzing and writing scientific papers. Although their results are still limited and rarely truly innovative, the consequences of the ability to produce research articles quickly and cheaply are already rippling through the scientific community. A natural question arises: what will happen to the tiny layer of genuine heroes of science who are still capable of real innovation, not merely on paper? Will they drown in these waves of automatically mass-produced scientific works?
And when American tech magnate Peter Thiel argues that science has not produced breakthrough discoveries over the last 50 years and therefore it is time to replace scientists with artificial intelligence, he is speaking precisely about the dead end of the current mechanism for producing scientific knowledge, presenting it as a dead end of human thought and humanity as a whole.
For example, one of the leaders in the AI market, OpenAI, recently introduced a specialized model for biology and drug discovery called GPT-Rosalind. Its competitor Google is working on powerful AI tools, including deep-data-analysis services known as Gemini Deep Research. Anthropic is conducting experiments in which several copies of its Claude model jointly conduct scientific research.
The scientific community is publishing an increasing number of papers created with the assistance of AI. The old system, based on peer review of articles published in scientific journals, is no longer capable of processing such a quantity of material or even distinguishing valuable work from the banal. According to estimates from the French National Center for Scientific Research, by 2025 there were around five million unpublished papers worldwide awaiting peer review.
It is already known that artificial intelligence systems, even the most advanced ones, tend to adapt themselves to user expectations. In the scientific world this is called fitting the result to the desired outcome. But until recently such fitting was not so widespread. Although a scientist may not know that results have been tailored by the system to fit their theory, the burden on reviewers does not decrease because of this. This means that at the next stage AI will begin reviewing scientific papers on a massive scale as well, and the circle will close.
The use of AI creates the risk of appropriating other people’s ideas without attribution. In addition, there is the problem of verifying results, because AI still suffers from “hallucinations.” For example, some references in an AI-generated article may simply not exist. Only a human being can establish this.
Until recently, the value of a scientist was roughly measured by the number of scientific publications they produced, as well as the frequency with which those works were cited by other researchers. This criterion was, to put it mildly, imperfect, since it was vulnerable to falsification and manipulation, but in the age of AI it is simply inadequate. Today, with little effort, one can quickly produce an article that will be published in respected scientific journals.
No one has yet devised a way to account for AI-inflated publication counts when making hiring and promotion decisions. Nor is it clear what will happen to young researchers if machines perform scientific tasks in their place — whether they will still be able to develop into truly serious scientists.
However, the greatest problem may be that the convenience of AI tools could steer global science down the path of least resistance. Human beings are weak, and children are being taught to use AI almost from the first grade, depriving them of the opportunity to learn how to think independently. And without that ability, no scientist can exist.
Nature cites data from researchers at Peking University: the introduction of AI sharply increases scientists’ productivity, but at the same time narrows the range of topics they study. Models are trained on gigantic volumes of already accumulated data. They can identify patterns where there is a great deal of data — for example, in bioinformatics, climatology, or solid-state physics. This situation shifts science away from the search for and discovery of the new toward endless recombination and permutation of already known knowledge — a kind of glass bead game.
As a result, humanity will receive countless brilliant refinements to already existing theories. But no scientific revolution comparable to the discovery of quantum mechanics or the theory of relativity should be expected, contrary to the hopes of AI enthusiast Peter Thiel.
This is a translation of an excerpt from the article No chance of stability by Olga Levandovksaya, Dmitry Vetchinkin, Maksim Karev, first published in The Essence of Time newspaper, issue 669.
