The Rise of AI and the Fall of Collectivism: How AI is Impacting Scientific Research
By Isa Arellano
Collaboration in science
Students on both the undergraduate and graduate levels may recall the gruelling process of scouring for secondary sources for their final essays and shudder at the thought. The countless hours of combing through Google Scholar or JSTOR to develop a thesis is a tedious but necessary ordeal. For scientists in academia, this research is essential for their job. It isn’t uncommon for their papers to have a bibliography spanning several pages in tiny font to properly cite the dozens of research others have done that contributed to their own work.
Contrary to most pop culture depictions of science projects as an individualistic endeavour undertaken by one or two eccentric researchers, major breakthroughs in science require the collective effort and work of hundreds, if not thousands, of scientists around the world. Advancements in science come from building off the work of others, and researchers nowadays seldom work alone for their milestone works. In fact, collaboration in the field of science results in bigger and better results. When scientists work together, they combine their unique perspectives, methodologies, and expertise to develop a greater understanding of complex science. Often this results in an accelerated research process, higher quality research, and builds connection between different scientific disciplines. [1]
Historically, technology has been a great aid to the collaborative spirit of the scientific community. With the invention of the internet and its accompanying platforms of communication and dissemination of information, scientists around the world are able to work with the greatest minds in the field without having to deal with geographic barriers. As a result, modern science has become global on a scale that could scarcely be imagined in the days of Sir Isaac Newton.
A recent article in the science journal Nature, however, indicates that the nature of collaboration and scientific diversity may be changing with the introduction of AI.
How AI is impacting science research
Contrary to the popular—though controversial—use of AI as a generative tool for some students, AI in science research is typically used for optimisation or predictive purposes, though the advancement of generative AI has inevitably led to its use in research as well. The most popular models of AI used in research papers are convolutional neural networks (CNNs), support vector machines (SVMs), and the random forest (RF) model. [2] The usefulness of these AI models in research becomes apparent when you understand what they do. CNNs are used to optimise image classification and object recognition [3] while SVMs are commonly used in classification issues to optimise the distinguishment between two classes. [4] In contrast, RF models are predictive algorithms used to generate a single result based off of multiple variable decisions. [5]
For scientists, the abilities of AI algorithms such as these allow for the automation of intensely time-consuming tasks. Instead of having to sort through large amounts of data to compile them into different categories on their own, scientists can use these algorithms to sort through data that would have taken them hours to do by hand. The increased speed of this type of data analysis also means they can use much larger data sets in their research, which allows for more accurate results.
When AI is used as a research tool like this, there seems to be little drawback. But when we stop focusing on what AI can do for individual researchers and look at the bigger picture of science as a whole, things become much more complicated.
AI as a tool for individual success
Published just this month on January 14th, an article by scientists from the Beijing National Research Center for Information Science and Technology and University of Chicago comes to a surprising conclusion about the latest technological advancement: AI constricts collective science.
Analysing over 40,000,000 research papers in six different scientific fields (biology, chemistry, physics, medicine, geology, and material science) from the OpenAlex research dataset, authors Qianyue Hao, Fengli Xu, Yong Li, and James Evans used a language model to identify research papers that used AI and verified them with humans. They then compared the data from these papers to ones that did not use AI. What they found was that authors of papers that used AI typically had accelerated careers, greater visibility, and more productivity.
Comparing statistics for these papers, AI papers had 98.70% more annual citations compared to their non-AI counterparts and their authors published three times more annually than their non-AI adopting counterparts on average. Research authors who adopt AI also had a higher rate of career success. Measuring the number of junior scientists who became established scientists (who they defined as scientists who have led one or more research projects), they found that AI-adopters had a rate of 45% of becoming an established scientist—13.64% more than their non-AI peers—and that they did so at an average of 1.37 years faster. [2]
All for one, less for all
If scientists who adopt AI for research are more productive and more successful, why does AI restrict science research as a whole?
The answer comes from the way AI is trained on data. The more data it is fed, the better it gets at analysing it. The issue that arises is when the AI is trained, it’s results will naturally be biased towards areas of research that already have lots of data and research because that’s the type of data it was trained to be good at understanding. Because of this, research that uses AI is skewed towards core areas of science that already have lots of existing data rather than new or unexplored topics that are more likely to lead to novel scientific breakthroughs.
In fact, signs of engagement with research papers that used AI were 22% less than those that did not use AI, indicating that “AI papers tend to only concentrate on the original paper, rather than forming dense interactions among each other, which is the characteristic of emerging fields” [2]. The authors of the paper suggest that this lack of engagement is caused by AI’s propensity for popular research topics within fields. These heavily researched topics do not create as much engagement because there is so much research, it is difficult to develop meaningful new discoveries.
The combination of less engagement with AI papers with their tendency to focus on well-studied topics results in overlapping research and an overall reduction in the diversity of scientific studies.
While AI certainly helps individual scientists with their research, it does so in a way that doesn’t encourage research in developing fields where breakthroughs and new research would be more impactful for humanity’s collective scientific knowledge. As AI becomes increasingly prominent in every aspect of our lives, it’s important to be conscious of its effects—whether positive or negative—on the information that will guide the future.
Sources:
[1] Quincy. 2025. “The Importance of Collaboration in Modern Science.” Research Studies Press. April 23, 2025. https://research-studies-press.co.uk/2025/04/23/the-importance-of-collaboration-in-modern-science/.
[2] Hao, Qianyue, Fengli Xu, Yong Li, and James Evans. 2026. “Artificial Intelligence Tools Expand Scientists’ Impact but Contract Science’s Focus.” Nature, January. https://doi.org/10.1038/s41586-025-09922-y.
[3] IBM. 2021. “Convolutional Neural Networks.” Ibm.com. October 6, 2021. https://www.ibm.com/think/topics/convolutional-neural-networks.
[4] Eda Kavlakoglu. 2023. “Support Vector Machine.” IBM. December 12, 2023. https://www.ibm.com/think/topics/support-vector-machine.
[5] Eda Kavlakoglu. 2021. “What Is Random Forest?” Ibm.com. IBM. October 20, 2021. https://www.ibm.com/think/topics/random-forest.
