Artificial intelligence is acting as rocket fuel for individual scientific careers, but a major new study warns it is coming at a heavy cost: the collective narrowing of human knowledge.
Research from the University of Chicago, published in Nature, reveals a stark paradox at the heart of the AI revolution. While AI tools allow individual researchers to become super-productive “stars,” their widespread adoption is causing the overall scope of scientific inquiry to contract.
The study, which analysed a massive dataset of 41.3 million research papers, found that the benefits for early adopters are staggering.
Scientists who utilise AI tools publish 3.02 times as many papers and receive 4.85 times as many citations as their peers. They also attain leadership positions in their fields an average of 1.4 years earlier.
The ‘lonely crowd’
However, this individual efficiency is creating a “methodological monoculture.” The researchers found that as AI use increased, the diversity of scientific topics studied declined by 4.63 per cent. Furthermore, engagement between scientists — the cross-pollination of ideas — dropped by 22 per cent.
James Evans, the Max Palevsky Professor of Sociology & Data Science and co-author of the study, describes this phenomenon as the creation of “lonely crowds.”
The problem, according to the researchers, is that AI models are designed for optimisation, not exploration. They thrive in “data-rich” environments where benchmarks are clear.
As a result, scientists are migrating en masse toward fields where data is already abundant, effectively abandoning “data-poor” or messy areas of research that might hold the next great breakthrough. Everyone is mining the same vein of gold because they have better pickaxes, leaving the rest of the map unexplored.
Premature convergence
The study warns that, without diverse approaches, science risks “premature convergence,” in which researchers rush to agree on established paradigms rather than testing genuinely novel ideas.
“To preserve collective exploration in an era of AI use, we will need to reimagine AI systems that expand not only cognitive capacity but also sensory and experimental capacity,” the authors wrote.
They argue that to save science from shrinking, policy changes are needed to incentivise the gathering of new data from inaccessible domains, rather than rewarding algorithms that merely squeeze more efficiency out of the data we already have.
