Non-English academics face inequality via AI-generated essays and countermeasure tools

Abstract

The recent introduction of ChatGPT by OpenAI has broadly introduced artificially intelligent (AI) tools as writing aids to the general public. We examined the limits of available detection tools in identifying writing that was fully aided by AI. Short essays (fewer than 400 words) were written by humans in two languages (English, Malay) or were procedurally generated with ChatGPT in four languages (English, Malay, Mandarin, Japanese). Malay, Mandarin, and Japanese essays were translated into English and paraphrased using a paraphrasing tool. All of the essays were then examined with three AI-detection tools available online. All of the English essays generated with the aid of AI were flagged as AI generated, including the AI-translated or paraphrased, non-English, human-written essays. This raises serious concerns for the academic world, especially non-English academics, in clarifying the differences between AI-generated and AI-aided essays in new policies. Blanket policies completely banning all AI tools may put nonnative English speakers at a disadvantage.

Ähnliche Arbeiten