Artificial intelligence has dramatically accelerated individual scientific productivity, but at a growing cost to the broader scientific enterprise. A landmark study published in Nature reveals that scientists who use AI tools—spanning from early machine learning to modern large language models—publish significantly more papers, earn far more citations, and advance faster in their careers than those who do not. On average, AI adopters publish 3.02 times more papers and receive 4.84 times more citations over their lifetimes. Junior researchers using AI are also more likely to stay in academia and reach leadership positions nearly 1.5 years earlier. However, the study uncovers a troubling trade-off: while AI benefits individuals, it may be undermining the collective progress of science. The research, which analyzed over 41 million scientific papers from 1980 to 2025 across biology, medicine, chemistry, physics, materials science, and geology, found that AI-driven research covers 4.6% less scientific territory than traditional work. This narrowing reflects a growing tendency for scientists to focus on a small set of popular, well-documented problems—often those with large, accessible datasets—creating a feedback loop where AI tools are most effective in already crowded areas. The study also found that AI papers generate 22% less cross-citation activity than non-AI papers. Instead of building diverse, interconnected knowledge networks, AI-driven research tends to cluster around a few high-profile “superstar” papers—such as AlphaFold in protein folding—leading to a concentration of effort and attention. This pattern, researchers say, limits the exploration of new frontiers and reduces the collaborative, cumulative nature of science. To identify AI use across such a vast dataset, the team trained a language model to scan titles and abstracts, flagging approximately 310,000 AI-related papers. Human reviewers confirmed the model’s accuracy was comparable to that of expert humans. The findings suggest that AI’s impact is not just about speed or efficiency—it’s reshaping scientific incentives and priorities. Experts warn that this trend could lead to a homogenized, less innovative scientific landscape. “Science is a collective endeavor,” says sociocultural anthropologist Lisa Messeri. “If AI benefits individuals at the cost of the system, we need a deep reckoning.” Yian Yin, a computational social scientist at Cornell, calls the results “amazing” in scale but alarming in implication. The good news, researchers say, is that the trend may be reversible. One path forward is to expand and improve data availability in underexplored scientific domains, making AI tools more useful across a wider range of research. As Zhicheng Lin of Yonsei University notes, the goal should not be to abandon data-driven science, but to make data more abundant and accessible in new areas. Looking ahead, the next generation of AI systems—moving beyond data analysis to autonomous scientific agents capable of hypothesis generation and experimentation—could reinvigorate discovery. Until then, the scientific community must reevaluate how it funds, rewards, and structures research to ensure AI enhances the collective pursuit of knowledge, not just individual success. As James Evans of the University of Chicago puts it: “We don’t have to accept a future where AI just intensifies old questions. We can build one where it opens new ones.”