Comparison of Generative Artificial Intelligence Tools in the Assessment of Student Assignments

Abstract

Grading student assignments, especially in large classes, is a time-consuming task, making the automation of the assessment process highly desirable. With the rise of Generative Artificial Intelligence (GAI), tools such as ChatGPT, Claude AI, and DeepSeek have emerged as possible solutions for automating this process. This study examines the effectiveness of the three GAI tools in grading three types of student assignments: graphical data models, code quality assignment, and text-based assignments. We conducted experiments across three undergraduate courses, comparing AI-generated grades with those manually assigned by teachers. The results showed that while GAI tools performed well in grading structured assignments like graphical models and code assessments, they struggled with the subjective nature of descriptive text evaluations. Claude AI performed best in evaluating graphical models, while DeepSeek outperformed ChatGPT in assessing code quality. However, deviations were observed in grading descriptive assignments, raising concerns about the reliability of GAI tools for tasks that require deep contextual understanding and creativity. The study highlights the potential of GAI to reduce grading time and provide constructive feedback, but also emphasizes the need for careful integration of these tools into educational practices. Ethical concerns, particularly regarding data privacy, suggest that offline or institutionally hosted GAI models could provide a viable solution.

Ähnliche Arbeiten