What happened
Chemical & Engineering News reports that just over 29% of science, technology, engineering, and mathematics PhD dissertations submitted in the United States through May contained artificial intelligence–generated writing. The outlet says the figure was 1.7% in 2023, the first full year ChatGPT was publicly available. The figures come from a recently published National Bureau of Economic Research working paper.
According to Chemical & Engineering News, slightly more than 29% of US STEM PhD dissertations submitted through May this year contained AI-generated writing. The outlet contrasts that with 1.7% in 2023. C&EN identifies the source as a recently published National Bureau of Economic Research working paper and names Daniel Gross, a Duke University professor of business administration, as one of its authors.
The supplied report does not provide the working paper’s title, methodology, sample size, definition of “AI-generated writing,” or details about how dissertations were selected and analyzed. It also does not say whether the figure represents any AI-assisted passage or a larger share of a dissertation’s text. These details remain unconfirmed from the supplied material.
Why it matters
The reported increase suggests that is becoming a significant part of scientific writing and doctoral education, with implications for research training, authorship norms, evaluation, and academic integrity. It may also affect how universities assess students’ writing and research skills. However, the supplied report does not establish whether AI assistance improved or harmed dissertation quality, how much text was generated, or whether institutions permitted the use.
If the reported trend is representative, universities may need clearer rules distinguishing proofreading and language assistance from substantive authorship. Supervisors and examiners may also need methods for evaluating a dissertation’s intellectual contribution when AI tools helped produce some of its prose.
The result is consequential because it concerns completed doctoral research across STEM fields rather than casual experimentation with chatbots. Still, the source does not independently verify the working paper’s estimate, and it provides no information about sample size, institutions, detection method, or the paper’s peer-review status.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
A route planner searches possible journeys using explicit rules. What does this illustrate about AI?
What to watch next
The key next questions are how the working paper measured AI-generated writing, whether the estimate varies by discipline or institution, and whether later data show the trend continuing. Universities’ policies and evidence about the effect of AI assistance on scientific reasoning and writing quality will also matter.
C&EN quotes Gross as expecting use in scientific writing to continue increasing. That is an author’s expectation, not an independently established forecast. Future reporting should test whether dissertation use continues to rise and whether patterns differ across STEM disciplines, universities, degree programs, or types of writing assistance.
No product access, availability, or pricing is relevant to this finding. The source does not state whether the working paper, underlying data, or analysis code is publicly available. It also does not report institutional responses, academic-integrity findings, or measured effects on dissertation quality.