Artificial intelligence (AI) agents who publish and debate scientific questions among themselves – without any human intervention: this is the challenge of Agent4Science, an American platform born in 2026. What role can AI play in human scientific progress?

Published on: 04/24/2026 - 1:18 p.m. Modified on: 05/01/2026 - 4:45 p.m.

Is there a single area left where human intelligence is not equaled, or exceeded? Since April 2026, the Agent4Science platform has operated as a scientific social network reserved for artificial intelligence (AI) agents only. Founded by Chenhao Tan of the University of Chicago, it allows more than 150 artificial entities to publish, debate and critique articles – without any human intervention.

Its NeuriCo program designs and conducts experiments autonomously. A twin platform, EinsteinArena, created at Stanford by James Zou, has produced eleven new solutions to listed mathematical problems.

"Two years ago, if you had talked to me about writing a scientific article with artificial intelligence, I would have said: 'It's science fiction'", comments Serge Abiteboul, computer scientist, researcher at the École Normale Supérieure (ENS) Paris and emeritus research director at the National Institute for Research in Digital Sciences and Technologies (Inria). "Today it's becoming reality. And I don't know what it will be in 5 or 10 years."

After having fascinated the general public, the language models – those used on this platform, but also ChatGPT, Claude and other Gemini – amaze and now support the scientific spheres. “They give me new perspectives, which I would not have had by reading an article alone,” confides the architect of the artificial scientists platform.

To display this YouTube content, it is necessary to authorize audience measurement and advertising cookies.

An extension in your browser appears to be blocking the loading of the video player. To be able to watch this content, you must disable or uninstall it.

Doing science from clouds of words

These machines – devoid of any contact with palpable reality – are nevertheless blind. But an abysmal amount of data, combined with a dizzying computing capacity, makes AI a virtuoso of prediction. In fact, Agent4Science agents do science from clouds of words.

Is human intelligence so different? “A scientist is a library that produces a book,” says Philippe Huneman, philosopher of science, research director at the Institute of History and Philosophy of Science and Technology (CNRS / University of Paris I Panthéon-Sorbonne). In addition to books, a scientist potentially produces discoveries capable of changing human destiny: fire, vaccines, soap, electricity, etc.

However, “to discover is to remove a veil that hides something,” recalls Jean-Gabriel Ganascia, computer scientist and AI philosopher at Sorbonne University. And from this point of view, of course the machine can discover. Because it can explore a lot of potential."

Also read Welcome to Moltbook, where AIs talk to each other... or create religions

“A weakening of science”?

And the machine has been discovering for years. In 2024, an American-British trio received the Nobel Prize in Chemistry for having unlocked – thanks to AI tools like AlphaFold2 from DeepMind – the secrets of proteins, opening up unexpected horizons in medical and pharmaceutical research.

These hopes coexist with the anxieties of surpassing humanity. By 2023, an analysis by Goldman Sachs concluded that around two-thirds of American and European jobs would be exposed to some degree of automation by AI.

"In the past, you could understand an equation like E=mc2 by carrying out calculations. Now, you are given terabytes of data, calculated at length by software, which will offer you hypotheses, analyses, statistics. But you are not capable of explaining them as we would have done in the 19th century. In a way, we can see that as a weakening of science", notes Serge Abiteboul.

However, "if we manage to cure a disease which today kills people by the millions thanks to software, I think we can rejoice about it", says the computer scientist.

See also How can artificial intelligence revolutionize medicine?

“We need intelligence”

Diseases, global warming, resource depletion, overpopulation and resulting conflicts: certainly, some of the challenges of the 21st century are linked to physical, chemical and biological questions.

And technologies like AI “solve problems that we were unable to understand 100 years ago. The use of AI in science is exciting as a computer scientist and scientist, but also as a citizen, because it shapes the world of tomorrow,” believes Serge Abiteboul.

Despite our lack of perspective on these technologies, "we need intelligence much more than being afraid of artificial intelligences, because they are complementary to ours", agrees Jean-Claude Heudin, AI researcher.

However, "a certain number of our problems are due to the forms of society that we have. No need for artificial intelligence to understand that we have a small problem with the distribution of wealth, or with overconsumption", smiles Philippe Huneman.

"The problem with the idea that artificial intelligence is going to find solutions is that the solutions are there. It's just that they come at a cost in terms of economic and political change."

Keywords associated with the article

The content you are trying to access does not exist or is no longer available.