Purpose. To evaluate the effectiveness of existing AI-generated text detection tools in identifying machine-generated content within students’ qualifying papers. The research aims to analyze the accuracy, reliability, and limitations of these detection methods, particularly for texts written in Ukrainian, and to explore potential improvements in AI content recognition. Methodology. The study employs a combination of statistical analysis and experimental evaluation of AI-detection tools. A dataset of AI-generated and human-written academic texts is used to assess detection accuracy, with additional experiments conducted to analyze the influence of linguistic structures, translation effects, and text formalization. The research also incorporates computational techniques to examine error rates and determine patterns in AI-generated content recognition. Findings. The study results provide insights into the performance of current AI-detection systems, highlighting their strengths and weaknesses. The analysis reveals significant challenges in detecting AI-generated text, particularly in non-English languages such as Ukrainian. Additionally, the study identifies the impact of translation on detection accuracy and discusses the effectiveness of different linguistic and statistical approaches. Originality. This research contributes to the ongoing discourse on academic integrity by addressing the limitations of AI-detection tools for non-English academic texts. Unlike previous studies focusing primarily on English-language AI-generated content, this study provides a unique perspective on the challenges of detecting AI-generated text in Ukrainian, offering novel insights into the adaptation of detection models for diverse linguistic contexts. Practical value. The findings of this study are valuable for educators, researchers, and policymakers concerned with maintaining academic integrity in the era of generative AI. By identifying weaknesses in current detection systems and suggesting possible improvements, the research provides a foundation for developing more robust AI-detection methodologies that can effectively apply to academic texts in multiple languages.
ПРИХОДЧЕНКО et al. (Wed,) studied this question.