Adopting artificial intelligence tools to analyze data and model results is having a huge impact on the career prospects of young scientists, greatly increasing their chances of rising to influential positions in their fields, according to new research. But that benefit for individual researchers appears to come at a broader cost to science.
Researchers at the University of Chicago and Tsinghua University, China, analyzed nearly 68 million research papers in six scientific fields (excluding computer science) and found that papers involving AI techniques were cited more often but also focused on a smaller topic and were more frequently cited. In fact, the more scientists use AI, the more they focus on the same set of problems that can be answered with large, existing datasets and gradually explore fundamental questions that can lead to new areas of research.
“I was amazed at the amazing level of discovery, [AI] it greatly increases the ability of people to stay and thrive within the system,” said James Evans, co-author of the previously published paper and director of the Knowledge Lab at the University of Chicago. “This suggests that there is a strong incentive for people to use these types of systems in their work … it’s between thriving and not surviving in a competitive research field.”
As that incentive leads to an increased reliance on machine learning, neural networks, and transformer models, “the entire scientific system created by AI is shrinking,” he said.
The study examined 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 percent more papers per year, on average, and their papers were cited more than three times more than those who did not use AI.
Evans and his co-authors then examined the career paths of 3.5 million scientists and categorized them as junior scientists, those who had not led a research team, or founding scientists, those who had. They found that young AI scientists were 32 percent more likely to go on to lead a research team—and progress to that stage of their career much faster—compared to their non-AI counterparts, who were more likely to drop out altogether.
Next, the authors used AI models to separate the topics covered by AI-assisted and non-AI-assisted research and examine how different types of papers correlated and sparked new lines of inquiry.
They found that, in all six scientific fields, researchers who used AI “shrunk” their discoveries by 5 percent, compared to researchers who did not use AI.
The field of AI-enabled research was also dominated by “star” papers. About 80 percent of all citations in that category went to the top 20 percent of most-cited papers and 95 percent of all citations went to the top 50 percent of most-cited papers, meaning that about half of the AI-assisted research was rarely if ever cited. again.
Similarly, Evans and his co-authors—Fengli Xu, Yong Li, and Qianyue Hao—found that AI research stimulated 24 percent more follow-up than non-AI research in the form of papers citing each other and originals. paper.
“These cumulative findings suggest that AI in science has become increasingly focused on certain hot topics that become a ‘lonely crowd’ with reduced communication between papers,” they wrote. “This focus leads to too many overlapping views and unwanted innovation that is linked to a decrease in knowledge and diversity throughout science.”
Evans, who specializes in how people learn and do research, said the effect of contracting in scientific research is similar to what happened with the advent of the Internet and academic journals going online. In 2008, he published a paper in the journal Science showing that as publishers went digital the types of studies cited by researchers changed. They have cited a few papers, from a small group of journals, and are interested in new research.
As an avid user of AI techniques himself, Evans said he is not anti-technology; the internet and AI both have obvious advantages in science. But the findings of his latest research suggest that government funding agencies, organizations, and academic institutions need to consider incentive programs for scientists to encourage work that is less focused on using specific tools and more focused on breaking new ground for future generations. of the researchers on which it was built.
“There is a poverty of thought,” he said. “We need to reduce the total diversion of resources to AI-related research in order to preserve some of these existing methods.”
Source link
