Numlock Sunday: Tara García Mathewson on the AI arms race in education
By Walt Hickey
Welcome to the Numlock Sunday edition.
This week, I spoke to Tara García Mathewson, who wrote “AI Detection Tools Falsely Accuse International Students of Cheating” for The Markup. Here's what I wrote about it:
A new study out of Stanford found that software used to spot plagiarism and AI-generated writing is significantly more likely to incorrectly flag the work of non-native English speakers as AI-generated. Submitting human-written text by native English speakers saw 5.1 percent of submitted work flagged as AI-written, while human-written text by non-native English speakers was flagged 61.3 percent of the time. Part of this is that the way that AI writes — direct, efficient sentences with a somewhat more limited vocabulary — is inherently the way that anyone who has to speak in a new or different language will often write. They tested this in a really clever fashion: When they modified the text from non-native English to include more diverse vocabulary, only 11.6 perce…