The adoption of artificial intelligence (AI) tools for data analysis and outcome modeling has significantly impacted young scientists’ career prospects, boosting their chances of achieving influential positions. However, a new study suggests this individual benefit may come at a cost to the broader scientific landscape.
Researchers from the University of Chicago and Tsinghua University in China analyzed nearly 68 million research papers across six scientific disciplines (excluding computer science). They found that papers incorporating AI techniques received more citations but also exhibited a narrower topical focus and increased repetitiveness. Essentially, the more scientists utilize AI, the more they concentrate on problems addressable with existing large datasets, potentially neglecting fundamental questions that could spark entirely new research areas.
“I was surprised by the dramatic scale of the finding; AI dramatically increases individuals’ capacity to thrive and advance within the system,” said James Evans, co-author of the pre-print paper and director of the Knowledge Lab at the University of Chicago. “This suggests a massive incentive for individuals to incorporate these systems into their work… it’s the difference between thriving and simply surviving in a competitive research field.”
However, as this incentive fosters growing dependence on machine learning, neural networks, and transformer models, “the entire system of science conducted through AI is shrinking,” he cautioned.
AI’s Influence on Scientific Productivity and Career Progression
The study, covering papers published from 1980 to 2024 in biology, medicine, chemistry, physics, materials science, and geology, revealed that scientists employing AI tools published an average of 67 percent more papers annually, with these papers receiving over three times the citations compared to those without AI integration.
Examining the career paths of 3.5 million scientists, categorized as either junior (not yet leading a research team) or established, the researchers discovered that junior scientists using AI were 32 percent more likely to lead research teams, progressing much faster than their non-AI counterparts, who were more prone to leaving academia.
The Concentration and “Lonely Crowds” of AI-Driven Research
Using AI models to categorize topics and analyze citation patterns, the authors found that AI researchers “shrunk” their topical coverage by 5 percent compared to non-AI researchers across all six disciplines.
AI-enabled research also exhibited a dominance of “superstar” papers. Approximately 80 percent of citations within AI research went to the top 20 percent of cited papers, and 95 percent went to the top 50 percent. This suggests that roughly half of AI-assisted research rarely receives subsequent citations.
Furthermore, the study found AI research stimulated 24 percent less follow-on engagement than non-AI research, as measured by reciprocal citations between papers.
“These findings suggest AI in science has become concentrated around specific hot topics, forming ‘lonely crowds’ with reduced interaction among papers,” the authors wrote. “This concentration leads to more overlapping ideas and redundant innovations, linked to a contraction in knowledge breadth and diversity across science.”
Balancing AI’s Benefits with the Need for Novel Exploration
Evans, who specializes in studying research practices and learning, acknowledges the benefits of the internet and AI in science. However, his study’s findings suggest a need for government funding bodies, corporations, and academic institutions to adjust incentive structures for scientists. This would encourage work focused on groundbreaking discoveries rather than solely on specific tool utilization, laying a foundation for future researchers.
“There’s a poverty of imagination,” he observed. “We must temper the complete shift of resources towards AI-related research to preserve alternative, existing approaches.”
Evans previously published a 2008 paper in Science showing how the shift to digital publishing altered citation practices, favoring fewer papers, a smaller range of journals, and newer research. He believes a similar dynamic may be at play with the increasing reliance on AI.
The Call for a More Balanced Approach
The implications of this research are crucial for the future of scientific inquiry. While AI offers valuable tools, it’s essential to ensure its use doesn’t stifle exploration and the pursuit of fundamental questions that could lead to transformative scientific breakthroughs. A balanced approach, supporting both AI-driven research and exploratory investigations, is necessary for a vibrant and diverse scientific landscape.
