Evaluating the Accuracy and Viability of AI‑Assisted  G  Grading in Multimodal University Assessments

 The integration of Artificial Intelligence (AI) in educational assessments has become a significant demand from both students and teachers, driven by the promise of responsive and scalable feedback. While it is technically possible for AI systems to generate both formative comments and summative grades, practical implementation reveals critical challenges surrounding consistency, fairness, and rubric interpretation. Human grading is inherently subjective with grading fatigue; educators strive to minimise bias, yet complete objectivity remains unattainable. Conversely, while AI may appear objective, it is susceptible to manipulation, and often interprets grading rubrics differently than human teachers. Based on a comprehensive review conducted six months ago, these discrepancies limit current AI grading capabilities primarily to formative feedback, rather than accurate summative scoring.To address these ongoing challenges and explore future possibilities, this project examines the accuracy and practical viability of two specific AI platforms: an in-house AI grading tool developed by the Computing Services Centre (CSC),and Moodie.ai, developed by HK Tech 300 startups. Utilising hundreds of multimodal assessment records from the SG8001 (Teaching Students: First Steps) course as empirical test cases, this study will evaluate the overall accuracy of AI grading and rigorously check the consistency of both numerical scores and qualitative comments. Beyond accuracy, the project aims to evaluate the feasibility of officially integrating these AI tools into the course curriculum. Furthermore, the research will focus on improving the user experience and interface design, identifying vulnerabilities to prevent malicious use, and proposing and evaluating innovative applications for AI-assisted pedagogical assessment.