Stanford study finds AI detectors misclassify non-native English student essays
A new Stanford study found that popular AI detectors are flagging a lot of non-native English speakers' essays as AI-generated, even when they're written without AI.
Researchers looked at 91 real student essays and saw that more than 61% of the essays were classified as AI-generated, on average, by seven detectors, while 89 of the 91 essays were flagged by at least one detector.
Meanwhile, American students' essays were usually classified correctly.
Researchers' ChatGPT rewrites cut false positives
When researchers rewrote the same essays with more advanced vocabulary using ChatGPT, the false alarms dropped to just 12%.
This shows current AI detectors aren't always fair and could hurt students who don't use complex English.
The takeaway? Schools should be careful about relying on these tools to judge academic honesty.