Recently, the application of artificial intelligence (AI) in academic writing has become increasingly popular, especially in the biomedical field. According to a recent study published in Nature, among the 1.5 million biomedical research abstracts published on PubMed in 2024, more than 200,000 papers showed features of AI-generated text. These features are often elaborate verbs and adjectives. Although they do not affect the core content of the paper, they are sufficient to reveal the "AI assistant" traces of the authors.
The study shows that the proportion of AI-assisted writing varies significantly across different countries and disciplines. In some countries and disciplines, the use of AI in writing exceeds 20%. In non-English speaking countries such as China and South Korea, the usage rate of large language models (LLMs) is as high as 15% due to language barriers. This has also led to a sharp increase in AI usage rates in some open-access, low-barrier journals, reaching up to 24%.
Notably, as people gradually recognize the characteristics of AI writing, many authors have started to deliberately avoid obvious AI traces. They hope to minimize the risk of being detected while working with AI. For example, the frequency of certain typical AI vocabulary has begun to decline after 2024, while the frequency of more general vocabulary has increased.
The research team downloaded 14 million abstracts from PubMed and, through analysis of vocabulary usage frequency, concluded that the influence of AI in biomedical papers is continuously deepening. They found that some words, such as "coronavirus," were overused before 2024, while after that, a large number of style words unrelated to the research content appeared, such as "intricate" and "notably." These words are mostly verbs and adjectives.
Facing the involvement of AI, researchers call for further exploration of how to regulate the application of AI in academic writing to ensure the rigor and fairness of scientific research. In the future, researchers hope to analyze more texts to understand the actual impact of AI on academic literature, so as to better guide the use of AI and ensure transparency and credibility in the academic community.