Research on Ethical Risks of Generative AI Applications in Higher Education

Engineering notes

Engineering notes will be added by the aipentium editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为大语言模型、生成式AI、ChatGPT相关技术、计算机视觉、深度学习等高价值论文补充中文说明。

Original abstract

This study constructs a dual-track academic risk detection system that integrates deep learning and graph theory algorithms, based on a heterogeneous dataset of documents including student assignments and academic journal papers. Driven by rapid advances in large language models, generative AI has been widely applied in higher education. While enhancing educational efficiency, it incurs severe academic ethical risks. Conventional plagiarism detection methods fail to identify AI-generated and semantically revised plagiarized content. The study establishes a multi-dimensional quantification model for AI generation probability, which integrates perplexity, burstiness, lexical features, and syntactic features to evaluate the ratio of AI-generated content in texts. Then, the paper adopts a Sentence - BERT model to map texts to high - dimensional semantic vectors, constructs a semantic similarity matrix, and builds an academic collusion network to identify covert plagiarism and improper citation behaviors. The empirical results show that about 15% of the documents in the dataset exhibit significant AI generation characteristics, and the language complexity of student assignments is significantly lower than that of academic papers, indicating a high risk of over-reliance on AI tools. Meanwhile, the semantic detection system accurately identifies multiple cases of academic misconduct, including direct plagiarism with a semantic similarity of 0.986 between documents, and reveals typical cheating patterns such as twin-star mutual plagiarism, dependent plagiarism, and free-riding group collusion. This study verifies the feasibility of localized large models for academic integrity evaluation under the premise of privacy protection, and proposes targeted governance paths for ethical risks of generative AI in higher education, providing technical and theoretical support for the construction of academic integrity systems in the era of human-machine collaboration.

Links and sources

Comments