Use of Generative AI in Medical Writing by Non-Native English Researchers

Abstract

Introduction: Generative Artificial Intelligence (AI) has increasingly found its way into scientific medical writing, which can be particularly inappropriate in non-native English-speaking countries. This study aimed to determine the occurrence of AI-generated texts in medical publications originating from the Greater Maghreb countries (Libya, Tunisia, Algeria, Morocco, and Mauritania).

Methods: This was a cross-sectional study that gathered all medical publications indexed on MEDLINE, published in the first week of July 2024, with first author affiliated with Greater Maghreb countries. The rate of AI-generated texts was calculated using the AI detection tool: ZeroGPT^®. Each article was analyzed in its entirety and each section separately (Abstract, Introduction, Methods, Results, and Discussion). Articles or sections were considered “suspects of AI generation” (sAI-g) if the rate was ≥25%. Results were presented as medians associated with their corresponding Inter Quartile Range (IQR).

Results: In all, 48 scientific medical articles were published by first authors from the Greater Maghreb countries. Articles were classified as “sAI-g” in 65% of cases, with a median rate of 36.2%[IQR=11.0%-49.4%]. AI-generated text was detected mainly in three sections: “Methods” (sAI-g=86%, median=59.3%[IQR=28.5%-71.7%]), “Abstract” (sAI-g=69%, median=52.2% [IQR=0.0%-90.2%]) and “Introduction” (sAI-g=58%, median=43.2%,[IQR=0.0%-79.4%]), while the "Discussion" section had the lowest median rate (sAI-g=30%, median=10.4%,[IQR=0.0%-27.9%]). 

Conclusion: Scientific medical articles from the Greater Maghreb countries used Generative AI extensively. This requires, on one hand, advancing medical education and mandating dissertations in English to build capacities for non-native English-speaking researchers, and on the other hand, providing training on the responsible use of AI  ..(abstract truncated at 250 words).