AI in the Library and Beyond

In this study, we evaluate the performance of several widely-used GPT detectors using writing samples from native and non-native English writers. Our findings reveal that these detectors consistently misclassify non-native English writing samples as AI-generated, whereas native writing samples are accurately identified. Furthermore, we demonstrate that simple prompting strategies can not only mitigate this bias but also effectively bypass GPT detectors, suggesting that GPT detectors may unintentionally penalize writers with constrained linguistic expressions.

It is nearly impossible to detect, and therefore enforce, the use of AI with any consistency or validity. Rigorous testing has shown that AI detectors are unreliable in most practical scenarios (Sadasivan, 2023). AI detectors have also been shown to be biased against English language learners (Liang, 2023). Another reason prohibitive policies may be problematic is more nuanced. This is to say that rigid restrictions may be detrimental, hindering the development of skills and behaviors needed to engage thoughtfully in an AI-integrated world.