The article AI is watering down science – with more and more articles appearing first in the online magazine BASIC thinking. With our newsletter UPDATE you can start the day well informed every morning.
Large language models like ChatGPT, Gemini and Co. are changing science. Researchers at Cornell University show that the use of AI significantly increases the number of scientific publications. But the mass of articles raises questions about the quality of the research.
Since the release of ChatGPT in 2022, the topic of artificial intelligence has gained significant momentum in almost all areas. Large language models now also play a crucial role in science.
More and more scientific publications are being created with the help of ChatGPT and Co. AI language models can help scientists formulate, structure and revise texts. This also makes it easier for researchers whose native language is not English to get started.
At the same time, however, there is a risk that more scientific publications will be produced more quickly and the quality will suffer. Researchers at Cornell University in the USA examined exactly this question and their results in the scientific journal Science published.
What role does AI play in science?
For their study, Cornell University researchers collected more than two million articles published on three online preprint websites between January 2018 and June 2024. The arXiv, bioRxiv and Social Science Research Network (SSRN) sites specialize in the fields of physics, life sciences and social sciences.
Scientific articles that have not yet been peer-reviewed are published on these pages. This is a common procedure in science in which specialist colleagues check scientific papers before they are published. This ensures quality, comprehensibility and technical accuracy.
The researchers led by Assistant Professor Yian Yin have developed an AI detector for their study. To do this, they probably compared articles written by humans from before 2023 with AI-generated texts. The aim of this comparison was to develop an AI model that can distinguish AI-generated texts.
The introduction of AI language models showed an enormous boost in productivity when evaluating publications. On the website arXiv, for example, around a third more articles were published by scientists who used AI than by scientists without AI support. For the websites bioRxiv and SSRN the increase was even more than 50 percent.
“It is a very widespread pattern in various scientific fields – from physical and computer sciences to life sciences and social sciences,” explains Yian Yin. “There is a major change in our current ecosystem that requires very serious consideration, particularly for those making decisions about what science we should support and fund.”
Non-native speakers benefit the most
The study results also show that scientists whose native language is not English benefit most from large language models. Accordingly, during the period examined, researchers from Asian institutions published between 43 and around 90 percent more articles than similar scientists who did not use the technology.
According to Yin, this development could even lead to a global shift in the regions with the greatest scientific productivity. Because of the use of AI language models, areas that were previously disadvantaged by the language barrier can now also participate.
“People who use LLMs have access to more diverse knowledge, which could lead to more creative ideas,” explains lead author Keigo Kusumegi, a doctoral student in information science. However, this does not mean that the quality of the publications will improve.
This shows the acceptance rate of the articles examined. This is because the articles that were likely written by LLMs were less likely to be accepted by academic journals. According to the researchers, this indicates that the reviewers classified many of these papers as of little scientific value, despite the convincing language.
According to Yin, this discrepancy between writing quality and scientific quality can lead to significant problems for editors and reviewers. Because of the increasing mass, it is becoming increasingly difficult to identify valuable submitted works. It is also no longer possible to evaluate scientists based on their pure productivity.
Also interesting:
- AI dialogues: Why artificial intelligence seems smarter than it is
- This is how you find out everything ChatGPT knows about you
- AI hype: Why we should slow down
- ChatGPT: Activate the reminder function – this is how it works
The post AI is diluting science – with more and more articles appearing first on BASIC thinking. Follow us too Google News and Flipboard or subscribe to our newsletter UPDATE.
As a Tech Industry expert, I have seen the rise of articles and discussions surrounding AI and its impact on science. While AI has undoubtedly revolutionized many aspects of scientific research, there is a concern that its increasing presence may be diluting the rigor and integrity of scientific processes.
One of the main issues is the reliance on AI as a “black box” solution, where researchers input data and receive outputs without fully understanding the underlying mechanisms at play. This can lead to a lack of transparency and reproducibility in scientific studies, as well as potential biases and errors in the AI algorithms themselves.
Additionally, the proliferation of AI-generated content, such as research papers and articles, raises questions about the authenticity and credibility of these sources. There is a risk that AI-generated content may lack the critical thinking and creativity that are essential components of scientific discovery.
As a Tech Industry expert, I believe it is crucial for scientists and researchers to be mindful of these challenges and to approach the integration of AI into their work with caution and skepticism. While AI has the potential to greatly enhance scientific progress, it is important to maintain the integrity and rigor of the scientific method in order to ensure the credibility and reliability of our findings.
Credits