NBER: AI-written PhD abstracts predict leaving academia

AI-generated text appeared in 29.4% of US STEM PhD dissertation abstracts by May 2026, up from 1.7% in 2023, Nature reported on a working paper that ran more than 226,000 dissertations through the AI-detection service Pangram. Mathematical and computer sciences topped the field at 36%.

The split by language background was stark. About 41% of dissertations from graduates of non-English-speaking countries carried AI-generated abstracts, versus 23% from English-speaking countries.

What makes the paper unusual is the career data. Among US undergraduates, PhD graduates whose abstracts contained AI-generated text were 17% less likely to work in academia and 16% more likely to go to industry. The correlation weakens for graduates who did their undergraduate work outside the US.

"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" needed for scientific careers, asked Daniel Gross, an economist at NBER. Ang Liu, an engineer at the University of New South Wales, said AI poses challenges when research enters areas with incomplete knowledge, where rigorous training becomes critical.

Two researchers on our Who's Who radar passed the paper around this week. The study tested abstracts, not full dissertations, and did not distinguish between wholly and partly AI-generated text.