Artificial intelligence may make scientists less creative

Adopting artificial intelligence tools to analyze data and model results has a significant impact on the career prospects of young scientists, dramatically increasing their chances of rising to influential positions in their fields, a new study shows. But this boon for individual researchers seems to come at the expense of science more broadly.

Researchers at the University of Chicago and Tsinghua University in China analyzed nearly 68 million papers in six scientific disciplines (not including computer science) and found that papers involving AI techniques were cited more often but also focused on a narrower set of topics. And topics. It was more frequent. In essence, the more scientists use AI, the more they focus on the same set of problems that can be answered with large existing data sets, and the less they care about exploring fundamental questions that can lead to entirely new areas of study.

“I was surprised by the sheer magnitude of this finding, as AI dramatically increases people’s ability to survive and advance within the system,” said study co-author James Evans. Prepress paper And director of the Knowledge Laboratory at the University of Chicago. “This suggests that there is enormous incentive for individuals to assimilate these types of systems into their work. It is between thriving and not surviving in a competitive field of research.”

As this incentive leads to increased reliance on machine learning, neural networks and transformer models, “the entire system of science that AI does is shrinking,” he said.

The study examined research papers published from 1980 to 2024 in the fields of biology, medicine, chemistry, physics, materials science, and geology. It found that scientists who used AI tools to conduct their research published 67% more papers per year, on average, and their papers were cited more than three times as much as those who did not use AI.

Next, Evans and his colleagues examined the career paths of 3.5 million scientists, categorizing them into either junior scientists, that is, those who did not lead a research team, or established scientists, that is, those who did lead a research team. They found that junior scientists who used AI were 32% more likely to go on to lead a research team – and advanced to that stage of their career much faster – compared to their non-AI counterparts, who were more likely to leave academia altogether.

The authors then used AI models to classify topics covered by AI-powered versus non-AI-powered research, and to examine how different types of papers cited each other, and whether they stimulated new lines of research.

They found that across all six scientific fields, researchers using AI “reduced” the subject area they covered by 5 percent, compared to researchers who did not use AI.

The field of AI-based research has also been dominated by “brilliant” research papers. Nearly 80% of all citations within this category went to the top 20% of most cited papers and 95% of all citations went to the top 50% of most cited papers, meaning that about half of AI-powered research was rarely cited again. Ever.

Likewise, Evans and his co-authors—Fengli Xu, Yong Li, and Qianyu Hao—found that AI research stimulated 24% less follow-up engagement than non-AI research in the form of papers that cited each other as well as the original research. paper.

“These combined results suggest that AI in science is becoming more focused around specific hot topics that have become lonely crowds as interaction between papers declines,” they wrote. “This focus leads to more overlapping ideas and redundant innovations associated with a shrinking scope of knowledge and diversity across the sciences.”

Evans, who specializes in studying how people learn and conduct research, said the impact of contracting on scientific research is similar to what happened with the advent of the Internet and the spread of academic journals online. In 2008, he published a book paper In the journal Science, it was found that as publishers went digital, the types of studies reported by researchers changed. They cited fewer papers, from a smaller set of journals, and preferred more recent research.

As an avid user of AI technologies, Evans said he is not anti-technology; Both the Internet and artificial intelligence have clear benefits for science. But the results of his latest study suggest that government funding bodies, companies, and academic institutions need to adjust incentive systems for scientists in order to encourage work that is less focused on using specific tools and more focused on opening new horizons for future generations. For researchers to build on.

“There is a poverty of imagination,” he said. “We need to slow down this complete replacement of resources for AI-related research to preserve some of these existing alternative approaches.”