Investigates the Accuracy, Efficiency, and Potential Bias of AI-Driven Automated Grading Systems

Abstract

Automated grading systems have become a part of the modern education evaluation process and are powered by Artificial Intelligence (AI) with the promise of efficiency, consistency, and scale. This paper examines the validity, effectiveness, and possible biasness of AI-based grading systems in learning institutions. With the growing use of machine learning and natural language processing algorithms by institutions to measure the performance of students, the question has arisen about the reliability, transparency, and fairness of automated evaluations. The study examines the relative performances of AI in grading systems and human ratings; the study finds that there are discrepancies in performance based on gender, ethnicity, and language. The research design based on mixed-method is that which integrates quantitative data analysis of the results of grading 500 student essays with qualitative data analysis of the teacher and student interviews. To estimate whether AI scoring systems are systemically biased or nonadherent to human grading criterion, statistical tests (regression analysis, ANOVA, and t-tests) are utilized. Results indicate that AI grading results in a time efficiency of up to 70 per cent, although accuracy in different fields is greatly different, and the issue of fairness remains to be debated, particularly in the case of non-native speakers of English. The paper highlights the need to have algorithmic transparency, ethical auditability and hybrid assessment models which incorporate human control. Finally, the paper will be used to contribute to the current debate on the responsible use of AI in education by providing empirical evidence and suggestions of the creation of a fair automated assessment system.

Keywords: Artificial Intelligence, Automated Grading, Algorithmic Bias, Efficiency, Accuracy, Educational Assessment, AI Fairness, Natural Language Processing, Machine Learning.