NBER: AI-written STEM dissertations tied to industry exit

In the first five months of 2026, 29.4% of US STEM doctoral dissertation abstracts contained AI-generated text, up from 1.7% in 2023 and 18.8% in 2025, according to a working paper covered by Nature.

Daniel P. Gross, Dror Shvadron and Hansen Zhang ran AI-text detection across more than 226,000 STEM dissertations completed between January 2019 and May 2026 at US universities. The study is NBER Working Paper 35859. The pattern is uneven: mathematical and computer sciences lead at roughly 36%, and graduates from non-English-speaking countries had AI-flagged abstracts 41% of the time versus 23% for graduates from English-speaking countries.

The career finding is the headline. Among PhDs who did their undergraduate degrees in the US, those with AI-generated abstracts were 17% less likely to pursue academic employment and 16% more likely to go to industry. The correlation was less pronounced for international undergraduate holders.

'If our students are only doing half the work and they're dishing the rest to AI, do they really have the depth of understanding to be good scientists?' Gross, an economist at NBER, asked. The paper's own conclusion lands softer but points the same way: AI 'thus appears to also be displacing a potentially important portion of the work through which scientific expertise is developed.'

Ang Liu, an engineer at the University of New South Wales, told Nature that AI performs well where prior knowledge is substantial, but when research moves into unknown territory 'rigorous research training becomes even more important.' The paper does not establish causation. Whether AI-leaning students were already headed for industry, or whether leaning on AI reshaped where they could go, these abstracts cannot settle.