ChatGPT-Driven Intelligent Educational Assessment: Algorithm, Challenges, and Ethical Considerations

Abstract

This study empirically examines the impact of graduate students' dependence on AI in academic writing and proposes corresponding regulatory strategies. The purpose of this paper is to address the practical application issues of AI-based writing tools in academic research. At present, AI writing tools have broken through the limitations of traditional writing methods and have significantly influenced academic writing practices, which deserves attention.In this study, empirical analysis is used to explore the influence of AI dependence on academic writing among graduate students, and to propose corresponding strategies, paradigms, and methods in response to issues of educational and writing ethics. The study employs correlation and regression analysis, revealing a significant negative correlation between AI dependence and academic originality (r = -0.482, p < 0.01), indicating that excessive reliance on AI may weaken independent research abilities and logical reasoning. While AI shows a positive effect on linguistic accuracy (r = 0.472, p < 0.01), it has a negative impact on argumentative ability (r = -0.385, p < 0.01).To quantitatively assess the degree of AI dependence, the study innovatively introduces computational algorithms such as Edit Distance, Jaccard Similarity, and Cosine Similarity to measure text similarity. However, as the research deepens, a fundamental consideration arises: the proposal of an AI Usage Index (AUI) model. This model integrates multiple similarity metrics and incorporates a BERT-based semantic similarity component to more precisely evaluate AI dependence. The findings provide a theoretical foundation for developing regulatory frameworks that balance the advantages of AI with academic integrity, with the aim of promoting sustainable and compliant AI development in academia.