Distinguishing between Student-Authored and ChatGPTGenerated Texts: A Preliminary Exploration of Human Evaluation Techniques

Abstract

The emergence of ChatGPT has opened up numerous possibilities as a supportive tool in the realms of education and research. However, the potential for students to engage in plagiarism facilitated by ChatGPT poses a significant challenge for university faculty members, particularly those possessing limited algorithm literacy and working in low-resourced educational contexts. The study draws upon the initial experiences and reflections of one of the authors and is accomplished by the collaboration of all the contributing authors. ChatGPT and forum posts within the learning environment served as the research tools. Employing the Turing Test, seven key human-detection techniques for deciphering ChatGPT-generated texts have been proposed in this study, including detecting discourse particles, conversational indicators, the degree of grammatical flawlessness and clarity, formulaic genre structure, numbered and sub-sectioned body paragraphs, use of transitional words and self-acknowledgment of ChatGPT’s non-human nature. This study contributes to the emerging body of literature on ChatGPT by enhancing quality education, reinforcing academic integrity and catalysing further research endeavours in the development towards human-detection strategies to combat ChatGPT-facilitated plagiarism.

Keywords

• ChatGPT
• Plagiarism
• AI
• Virtual Learning
• Human-detection