Study Finds AI Detectors Often Misclassify Careful Non-Native Writing as Machine-Generated
A 2023 Stanford study tested seven commercial AI detectors on 179 essays, including 91 written by non-native English speakers under exam conditions. The detectors flagged 89 of these non-native essays as machine-generated, showing a 61.3% false-positive rate, compared to about 5% for native speakers. This occurs because detectors rely on 'perplexity,' which measures word predictability; careful, clear writing by non-native speakers or editors resembles AI-generated text statistically, causing misclassification.
First-hand measurement across 2 sources
We measured how 2 outlets covered this story. No outlet gave this story a measurable political slant — there is no left–right reading to report. Overall sentiment is neutral (46/100). Lens Score 27/100.
Outlets measured: timesnow, timesnow. See how each one headlined and framed the same story in the source comparison below.
AI Analysis
Sentiment was consistent across outlets (45–48/100), indicating broadly factual reporting rather than editorialising.
Coverage timeline
timesnow broke this story on 14 Sept, 04:01 pm. Other outlets followed.
- 1