Fact of the Week: At Least 10 Percent of Scientific Research Is Being Co-Authored by AI

Source: Kobak, Dmitry et al. "Delving into ChatGPT usage in academic writing through excess vocabulary. " (working paper, Hertie Institute for AI in Brain Health, University of Tübingen, Germany, 2024).

Commentary: In a recent working paper published by Dmitry Kobak et al., researchers tracked the usage of excess words to assess the prevalence with which large language models (LLMs), a form of AI, were being used to write scientific articles. Using a sample of over 14 million biomedical abstracts extracted from PubMed, an online database of biomedical and life sciences articles, researchers assessed the frequency that unique style words often used by ChatGPT and other LLMs were used. These style words, such as "intricacies", "meticulous", and "delving", were found to have increased in frequency at an unparalleled rate compared to historic norms, indicating the usage of LLMs. The authors found that at least 10 percent of biomedical abstracts published in 2024 were found to have been written with an LLM.