How AI-Powered Grading Systems Are Changing University Exams

The integration of artificial intelligence technologies into evaluating university assignments and exams has brought important debates on human oversight, privacy regulations, and grading reliability to the forefront of the education sector.

AI-driven grading applications are not a new phenomenon, having been commercially developed by educational publishers such as Pearson since 1998. AI models like ChatGPT and Claude predict a grade that can be given by comparing the text patterns in a student's work with existing texts. An instructor can upload an essay and evaluation criteria to an open-access AI tool to request grading and feedback.

Universities enforce strict privacy regulations regarding the use of AI. For example, Deakin University prohibits instructors from uploading student assignments and private data to open-access AI systems. Meanwhile, the University of Melbourne states that approved AIs can be used to support evaluations, but sole reliance on AI cannot be used to determine grades.

Research published in August by the University of Melbourne and Cardiff University revealed that ChatGPT could not precisely mimic human scores when evaluating 50 undergraduate bioscience essays. Researchers found that AI fell short particularly in individual criteria such as content relevance and supportive evidence, and awarded higher grades on average than humans.

It is stated that AI-based grading applications could weaken the relationship between students and teachers, causing students to turn directly to AI instead of teachers. Additionally, some institutions, such as the University of New South Wales, are conducting pilot studies on removing numerical grades and evaluating students based on specific achievements.