NBER: AI-written dissertations push PhDs toward industry

PhD students who used AI to write their dissertation abstracts were measurably less likely to stay in academia, according to a working paper that analyzed more than 226,000 US STEM dissertations filed between January 2019 and May 2026. Nature reported the finding from NBER paper 35859 by economist Daniel P. Gross with co-authors Dror Shvadron and Hansen Zhang, who ran dissertation abstracts through the Pangram AI-detection service.

The share of abstracts flagged as AI-written climbed from 1.7% in 2023 to 18.8% in 2025, reaching 29.4% in the first five months of 2026. Mathematical and computer sciences led the fields at 36%, engineering at 35%, physical sciences at 22%. Graduates from non-English-speaking countries used AI in 41% of their abstracts, against 23% from English-speaking countries.

The career signal is strongest for US-origin graduates: those whose dissertations were flagged were 17% less likely to pursue academic employment and 16% more likely to go into industry than peers in the same program and cohort. The pattern was weaker for international graduates. The authors write that 'AI appears to also be displacing a potentially important portion of the work through which scientific expertise is developed.'

Gross framed the stake bluntly: '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?' Ang Liu, an engineer at the University of New South Wales, told Nature that for frontier research 'rigorous research training becomes even more important.'

NBER paper 35859 is unreviewed, and the authors report correlation, not cause. The Pangram method flags only the abstract, not the underlying research.