COURSEWORK ASSESSMENT
rubrics. Its integration allowed us to track not only accuracy but also workflow efficiency, student reactions and the quality of AI‑generated feedback. Training AI to reduce grading errors One of the most interesting outcomes of this project has been the way it has prompted us, as educators, to rethink our approach to assessment. In pedagogy, we often distinguish between summative assessment – ie the final grade – and formative assessment, ie the feedback that supports student learning and development. While AI is clearly effective at standardised, rubric-based summative grading, it also shows real promise in the realm of formative assessment. When prompted effectively, AI can provide detailed, structured feedback tailored to the student’s work. In fact, the act of designing those prompts has made us more deliberate in how we think about evaluation and feedback more broadly. To get the most out of AI, we had to train it to align with our expectations. Initially, the feedback was either too vague or overly enthusiastic. In some cases, the AI even ‘hallucinated’ content, praising students for arguments they hadn’t made or referencing sections that didn’t exist. To counter this, we introduced stricter parameters that required the tool to justify every comment with a direct quote from the student’s paper. This significantly reduced hallucinations and led to more focused, relevant feedback. Scaling AI in assessment: the 80/20 model That said, not all papers can or should be evaluated solely by AI, which is why we developed what we now refer to as the 80/20 model of AI grading. In our experience, 80 per cent of student submissions follow predictable patterns and can be reliably assessed using standard rubrics, which makes them well-suited to automation. The remaining 20 per cent – those that present more creative or unconventional reasoning – still require human judgment. For example, in law assignments, some students applied legal principles that weren’t explicitly expected but were still valid and well-argued. These nuanced responses needed to be reviewed manually to ensure that original thinking was recognised and rewarded.
This division of labour is particularly valuable in large cohorts, where professors often struggle to give each student the attention they deserve. AI-assisted grading frees up time, allowing educators to focus on more meaningful interactions, whether through coaching, personalised feedback, or deeper classroom engagement. When used at scale, particularly when assessment formats remain consistent over time, the benefits of this approach compound. The time-saving aspect should not be underestimated. For a professor managing more than 100 students, the grading workload can be substantial, often taking days to complete. With AI handling the more routine assessments, those hours can be redirected toward one‑on-one mentoring or curriculum development. Moreover, this hybrid model enables educators to intervene only when necessary, utilising their expertise where it has the most significant impact. It is understandable that accuracy remains a key concern, but our findings suggest that AI and human graders are often remarkably aligned. In one of our testing rounds, 12 out of 15 papers received grades from AI that deviated less than 1.5 points from human scores. Importantly, AI also flagged a few cases where a student’s reasoning was stronger than what had been recognised by the original grader. Upon reviewing these cases, we found AI’s interpretation to be valid, suggesting that, beyond consistency, AI might even serve as a form of quality control mechanism. Analysing student feedback Where we’ve encountered more resistance is on the human side of the equation. While faculty members have been generally open to experimenting with AI tools, students have expressed hesitation. For many, grading represents more than a score; it’s a form of academic recognition and personal validation. There is a strong sense that their work should be seen and understood by a person, not just processed by a machine. This response underscores a critical truth: assessment is not only technical but also relational. That human dimension must remain central. AI can support a professor’s work, but it cannot replace the pedagogical relationship that underpins effective learning. Students want to know their work matters and that someone, not something, is paying attention. Looking ahead, we believe AI will play an increasingly significant role in academic assessment, but its integration must be thoughtful and transparent. We need more experimentation, more research and more open dialogue. There is no plug-and- play solution here. However, by integrating these tools into our teaching practice and studying them in real-world settings, we can begin to develop an approach that strikes a balance between innovation and integrity. Grading may be changing, but the core values of education are not. As educators, it’s our responsibility to lead this transformation with curiosity, care and a willingness to challenge our assumptions. The red pen may not be obsolete yet, but it may soon be forced to share desk space with AI.
BIOGRAPHIES
Peter Daly is professor of management and former director of the MSc in Management and Leadership at EDHEC Business School
Emmanuelle Deglaire is associate professor of law and taxation and a member of the EDHEC Augmented Law Institute
Ambition • ISSUE 6 • 2025 35
Made with FlippingBook - Share PDF online