A study analyzing over one million scientific papers found AI-generated content is pervasive. The research utilized statistical models to identify unique patterns and word usage differentiating AI text from human authorship. Significant variations exist among fields, with computer science showing the highest AI usage at 22.5% of sentences, followed by 9.8% in physics and 7.8% in mathematics. Shorter papers and those in crowded research fields exhibited more AI characteristics. Journals have begun mandating disclosure of generative AI tool usage, highlighting AI's role in editing rather than authorship.
The study found that AI is pervasive across research topics, with significant usage in computer science (22.5% of sentences) compared to physics (9.8%) and mathematics (7.8%).
Researchers identified signature traces and patterns in AI-generated texts, such as uncommon word frequencies, to distinguish between AI and human authorship.
Collection
[
|
...
]