The ChatGPT conundrum: Human-generated scientific manuscripts misidentified as AI creations by AI text detection tool

• Published in: Journal of pathology informatics Vol. 14; p. 100342
• Main Authors: Rashidi, Hooman H., Fennell, Brandon D., Albahra, Samer, Hu, Bo, Gorbett, Tom
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.01.2023; Elsevier
• Subjects: Chat-GPT; Generative AI; GPT; Human-generated; Large Language Model; LLM; Machine learning; Original; Text detection; Generative AI; Chat-GPT; Large Language Model; Text detection; Machine learning; GPT; Human-generated; LLM
• ISSN: 2153-3539; 2229-5089; 2153-3539
• DOI: 10.1016/j.jpi.2023.100342

AI Chat Bots such as ChatGPT are revolutionizing our AI capabilities, especially in text generation, to help expedite many tasks, but they introduce new dilemmas. The detection of AI-generated text has become a subject of great debate considering the AI text detector’s known and unexpected limitations. Thus far, much research in this area has focused on the detection of AI-generated text; however, the goal of this study was to evaluate the opposite scenario, an AI-text detection tool's ability to discriminate human-generated text. Thousands of abstracts from several of the most well-known scientific journals were used to test the predictive capabilities of these detection tools, assessing abstracts from 1980 to 2023. We found that the AI text detector erroneously identified up to 8% of the known real abstracts as AI-generated text. This further highlights the current limitations of such detection tools and argues for novel detectors or combined approaches that can address this shortcoming and minimize its unanticipated consequences as we navigate this new AI landscape. •Little research has been focused on an AI-text detection tool's ability to discriminate human-generated text.•Abstracts from esteemed scientific journals were used to test the predictive capabilities of these detection tools, assessing abstracts from 1980 to 2023.•An AI text detector erroneously identified up to 8% of the known real abstracts from prior to the existence of LLMs as AI-generated text with a high level of confidence.•This work highlights the yet unexplored risk of mislabeling human-generated work as AI-generated and begins the discussion on the potential consequences of such a mistake.