AI use in science narrows topical diversity and accelerates individual careers, study finds

A large-scale analysis of 41.3 million academic papers across six natural sciences found that researchers who incorporate AI tools into their work publish roughly three times as many papers and receive nearly five times as many citations as those who do not. The study, led by sociologist James Evans at the University of Chicago and published 14 January 2026 in Nature, used a natural language processing model to identify AI-augmented research in biology, chemistry, physics, medicine, materials science, and geology from 1980 to 2025.

The researchers mapped AI-heavy papers in a high-dimensional knowledge space and found they occupy a smaller intellectual footprint, cluster more tightly around popular, data-rich problems, and generate weaker networks of follow-on engagement between studies. This pattern persisted across decades of AI development, including early machine learning, the rise of deep learning, and the current wave of generative AI.

The findings highlight a tension between personal career advancement and collective scientific progress. Evans and colleagues argue that AI tools reward speed and scale but not surprise, potentially funneling researchers into the same crowded corners of inquiry. "You have this conflict between individual incentives and science as a whole," Evans said. Some experts, such as physicist Luís Nunes Amaral of Northwestern University, described the trend as "very problematic," warning of a feedback loop of conformity and declining originality.

The study excluded fields focused on developing AI methods, such as computer science and mathematics, to isolate the effect of AI use within the natural sciences. The dataset spanned 45 years and compared approximately 311,000 AI-augmented papers with millions of non-AI papers, tracking individual career trajectories, citation accumulation, and intellectual clustering across fields.