Artificial Intelligence Potentially Reducing Scientists' Creativity Levels
Employing AI tools to dissect data and foresee outcomes significantly boosts the job opportunities of young researchers in various scientific fields, increasing their chances of ascending to influential positions, as per a recent study. However, this advantage for individual scientists appears to come at a price for the wider scientific community.
Researchers from the University of Chicago and Tsinghua University, China, examined nearly 68 million research papers across six scientific disciplines (excluding computer science) and discovered that papers incorporating AI techniques received more citations but also delved into a limited array of topics and exhibited repetitive patterns. Essentially, the more scientists utilize AI, the more they gravitate towards addressing the same set of issues that can be resolved using existing large datasets, and the less they delve into fundamental questions that could spawn entirely new domains of research.
"I was amazed by the magnitude of this finding," said James Evans, a co-author of the pre-print paper and director of the Knowledge Lab at the University of Chicago. "This indicates a significant incentive for individuals to adopt these systems in their work... survival in a competitive research field depends on it."
As this incentive fosters an increasing reliance on machine learning, neural networks, and transformer models, "the entire AI-driven scientific system is shrinking," he noted.
The study examined papers published between 1980 and 2024 in the disciplines of biology, medicine, chemistry, physics, materials science, and geology. It revealed that scientists utilizing AI to conduct their research published 67% more papers annually, on average, and their papers were cited over three times as frequently as those who did not use AI.
Evans and his co-authors then scrutinized the career trajectories of 3.5 million scientists and categorized them as either aspiring researchers or established scientists. They found that junior scientists utilizing AI were 32% more likely to lead a research team and progressed to this stage of their career more quickly, compared to their non-AI peers, who were more likely to leave academia.
The researchers then employed AI models to categorize the topics covered by AI-assisted versus non-AI research and to examine the interconnections between the different types of papers. They discovered that across all six scientific fields, scholars utilizing AI narrowed the scope of topics they covered by 5%, compared to those who did not use AI.
The realm of AI-driven research was also dominated by "star" papers. Approximately 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, indicating that half of AI-assisted research is seldom, if ever, cited again.
Similarly, Evans and his co-authors - Fengli Xu, Yong Li, and Qianyue Hao - found that AI research spurred 24% less follow-on engagement, in the form of papers citing each other and the original paper, than non-AI research.
"These findings suggest that AI in science has become more concentrated around specific trendy topics that create 'lonely crowds' with reduced interaction between papers," they wrote. "This concentration leads to more overlapping ideas and redundant innovations linked to a contraction in knowledge extent and diversity across science."
Evans, a researcher specializing in studying how people learn and conduct research, said that the contracting impact of AI on scientific research is akin to what happened as the internet emerged and academic journals went online. In 2008, he published a paper in the journal Science demonstrating that as publishers went digital, the types of studies researchers cited changed. They cited fewer papers, from a smaller pool of journals, and favored newer research.
As a prolific user of AI techniques himself, Evans stated that he is not anti-technology; the internet and AI both offer clear advantages to science. However, the findings of his latest study indicate that funding bodies, corporations, and academic institutions need to adjust the incentive systems for scientists to encourage work that is less focused on using specific tools and more focused on breaking new ground for future generations of researchers to build upon.
"There's a lack of creativity," he said. "We need to slow down the complete replacement of resources to AI-related research to preserve some of these alternative, existing approaches."
The increase in the use of artificial intelligence in scientific research has led to a concentrating effect, where scholars focus on a limited number of topics and exhibit repetitive patterns, as shown by the study. This trend potentially reduces the exploration of fundamental questions and the discovery of entirely new research domains, indicating a potential shrinking of the AI-driven scientific system.
Furthermore, the study revealed that in six scientific disciplines, researchers utilizing AI published more papers annually and saw their work cited more frequently. However, this reliance on AI tools for research has resulted in a narrowing of topics and a decrease in follow-on engagement, suggesting a reduction in the diversity and extent of knowledge in these fields.
