AI detectors are almost never wrong, and up to 30% of STEM dissertation abstracts use AI

Afinetheorem · x · 2026-11-05

University of Toronto Rotman professors Kevin A. Bryan and Dror Shvadron argue in The Globe and Mail that AI detectors like Pangram are almost never wrong, using the case of prize-winning Haitian-Québécois novelist Thélyson Orélien — whose publisher Boréal cut ties over suspected AI writing — as a launching point.

Key points:

A new study finds up to 30% of recent STEM PhD dissertation abstracts were written with AI assistance;
Other examples include work by Harvard professor Ricardo Hausmann, a Granta prize-winning story, an Akutagawa Prize novel by Rie Kudan, and a NYT book review;
The authors argue AI use in creative tasks is increasingly common, whether admitted or not, and excuses denying AI authorship are always wrong.

Original post →

AI detectors are almost never wrong, and up to 30% of STEM dissertation abstracts use AI

More from AGI Musings