Key points are not available for this paper at this time.
Artificial Intelligence (AI) language models continue to expand in both access and capability. As these models have evolved, the number of academic journals in medicine and healthcare which have explored policies regarding AI-generated text has increased. The implementation of such policies requires accurate AI detection tools. Inaccurate detectors risk unnecessary penalties for human authors and/or may compromise the effective enforcement of guidelines against AI-generated content. Yet, the accuracy of AI text detection tools in identifying human-written versus AI-generated content has been found to vary across published studies. This experimental study used a sample of behavioral health publications and found problematic false positive and false negative rates from both free and paid AI detection tools. The study assessed 100 research articles from 2016-2018 in behavioral health and psychiatry journals and 200 texts produced by AI chatbots (100 by "ChatGPT" and 100 by "Claude"). The free AI detector showed a median of 27.2% for the proportion of academic text identified as AI-generated, while commercial software Originality.AI demonstrated better performance but still had limitations, especially in detecting texts generated by Claude. These error rates raise doubts about relying on AI detectors to enforce strict policies around AI text generation in behavioral health publications.
Popkov et al. (Fri,) studied this question.