Computer Science Papers Lead in AI Use for Writing, Study Finds
Researchers have identified a significant increase in the use of large language models (LLMs) like ChatGPT in scientific writing, particularly in computer science. An analysis of over one million pre-print and published papers between 2020 and 2024 revealed that computer science research exhibits the highest and fastest adoption of AI tools, with usage rates reaching up to 22%.
Study Overview
The research team examined over 1.1 million papers from pre-print archives arXiv and bioRxiv, as well as published articles in Nature journals. By identifying terminology commonly associated with LLMs, they estimated the extent to which AI systems were involved in modifying research content.
Key Findings
- Fastest Growth in Computer Science: The largest increase in AI use was observed in computer science papers, with up to 22% showing signs of LLM involvement.
- Regional Differences: Pre-prints from authors in regions with fewer English speakers, such as China and continental Europe, showed higher AI reliance.
- Lower Usage in Mathematics and Nature Journals: Papers related to mathematics and those published in Nature journals demonstrated less evidence of AI content modification.
- Shorter Papers and Frequent Pre-prints: There was a higher rate of AI use in shorter papers and among authors who post pre-prints more often, suggesting AI helps researchers manage higher writing volumes.
Implications for Research Culture
The study suggests that competitive pressure and the need to publish quickly may be driving scientists to adopt AI writing tools. However, disclosure of AI use remains rare. An inspection of 200 randomly selected computer science pre-prints from February 2024 showed only two papers explicitly mentioned using LLMs during writing.
This low disclosure rate points to unclear policies or other reasons behind authors’ reluctance to reveal AI assistance. Future research could explore these motivations further, helping develop clearer guidelines for AI transparency in academic writing.
Broader Impact on Academic Writing
Another recent study analyzing over 15 million biomedical papers from 2010 to 2024 found a notable shift in academic vocabulary linked to AI-generated content. According to this research, at least 13% of 2024 research abstracts likely benefited from LLMs, identified by increased use of stylistic language favored by AI.
As AI tools continue to influence scientific writing, understanding their role and encouraging responsible disclosure will be essential for maintaining research integrity.
For those interested in learning about AI applications in writing and research, Complete AI Training offers a range of courses on large language models and AI-driven content creation.
Your membership also unlocks:
