AI-AUGMENTED ASSIGNMENT EVALUATION: A FRAMEWORK FOR FAIR AND EFFECTIVE ASSESSMENT IN THE ERA OF LARGE LANGUAGE MODELS

ChatGPT and other Large Language Models such as Gemini have been quickly adopted, and have been affecting academic practice, particularly in assignment writing and grading. Even though these technologies offer a lot of efficiency to the grading process, feedback generating, and documentation, they pose serious pedagogical risks, including the absence of student engagement, excessive reliance on AI-generated content, and the inability to quantify actual learning. I propose AI-Augmented Assessment Evaluation Framework (AAAF) in this paper where automated evaluation will be integrated with human judgment to ensure fairness, transparency, and pedagogical integrity. The framework suggests two additional factors: the Pedagogical Assessment Score (PAS), assessing conceptual knowledge and the level of reasoning, and Pedagogical Risk Score (PRS) assessing AI dependency and authenticity concerns. An integrated assessment model is developed to achieve efficiency and learning results. The proposed framework is validated using a comparative study that would involve 80 undergraduate students that will be divided into three groups: traditional manual evaluation (G1), AI-only evaluation (G2), and AAAEF hybrid evaluation (G3). According to the outcomes of the experiments, in spite of the fact that AI saves a lot of time (44.7 min to 8.1 min), the hybrid AAAEF is better in grade consistency (89.96), retention of learning (86.42), and student satisfaction (87.81) with a safe AI dependency index (27.3). The study provides effective suggestions to educators and policymakers that can reorganize the assessment plans in the evolving AI-centered education environment.