Using artificial intelligence (AI) increases scientists’ productivity and impact but collectively leads to a shrinking of research focus. That is according to an analysis of more than 41 million research papers by scientist in China and the US, which finds that scientists who produce AI-augmented research also progress faster in their careers than their colleagues who do not (Nature 649 1237).
The study was carried out by James Evans, a sociologist at the University of Chicago, and his colleagues who analysed 41.3 million papers listed in the OpenAlex dataset published between 1980 and 2025. They looked at papers in physics and five other disciplines – biology, chemistry, geology, materials science and medicine.
Using an AI language model to identify AI-assisted work, the team picked out almost 310 000 AI-augmented papers from the dataset. They found that AI-supported publications receive more citations than no-AI-assisted papers, while also being more impactful across multiple indicators and having a higher prevalence in high-impact journals.
Individual researchers who adopt AI publish, on average, three times as many papers and get almost five times as many citations as those not using AI. In physics, researchers who use AI tools garner 183 citations every year, on average, while those who do not use AI get only 51 annually.
AI also boosts career trajectories. Based on an analysis of more than two million scientists in the dataset, the study finds that junior researchers who adopt AI are more likely to become established scientists. They also gain project leadership roles almost one-and-a-half years earlier, on average, than those who do not use AI.
Fundamental questions
But when the researchers examined the knowledge spread of a random sample of 10 000 papers, half of which used AI, they found that AI-produced work shrinks the range of topics covered by almost 5%. The finding is consistent across all six disciplines. Furthermore, AI papers are more clustered than non-AI papers, suggesting a tendency to concentrate on specific problems.
Knowledge grows step-by-step despite the exponential growth of papers, finds study
AI tools, in other words, appear to funnel research towards areas rich in data and help to automate established fields rather than exploring new topics. Evans and colleagues think this AI-induced convergence could drive science away from foundational questions and towards data-rich operational topics.
AI could, however, help combat this trend. “We need to reimagine AI systems that expand not only cognitive capacity but also sensory and experimental capacity,” they say. “[This could] enable and incentivize scientists to search, select and gather new types of data from previously inaccessible domains rather than merely optimizing analysis of standing data.”
Meanwhile, a new report by the AI company OpenAI has found that messages on advanced topics in science and mathematics on ChatGPT over the last year have grown by nearly 50%, to almost 8.4 million per week. The firm says its generative AI chatbot is being used to advance research across scientific fields from experiment planning and literature synthesis to mathematical reasoning and data analysis.
