AI-assisted grading: feedback on a full-scale test at EDHEC

In the education sector, artificial intelligence (AI) is raising great hopes: greater accessibility to knowledge without any language barriers, individualized teaching methods adapted to the needs of each learner, instant, personalized student feedback irrespective of time or place.

There are also many fears: of the promotion of ‘lazy’ students, who will make AI work for them instead of using their brains, and teachers without any added value, who will eventually become redundant.

While much academic work focuses on the use of AI by students, the work carried out at EDHEC takes the opposite approach and focuses on its use by professors.

Grading at a crossroads

Among the experiments currently underway at EDHEC (virtual teaching assistant, native implementation of AI tools in the learning management system, etc.), an in-depth test has been carried out on a very specific application of AI in education, namely the grading of exam papers, which lends itself nicely to automation given the time-consuming and repetitive nature of this task.

This project was made possible with the close collaboration of the PiLab, whose Director, Emmanuelle Houet, in a recent interview (2) outlines the laboratory's contribution to this project and, more broadly, its missions.

In line with the ‘learning by doing’ culture, the implementation of AI-assisted correction of exam papers has given rise to a research project (1), operational applications and has served as a basis for new educational objectives.

A multi-faceted research project

The first project, which was exploratory in nature, was designed to share the experience of a teacher on her ‘journey’ of getting to grips with AI for the purpose of assisting with the correction of exam papers. This work drew on existing research, in particular the work of Molenaar (3) on the hybridization of humans and technology in education and that of Raisch and Krakowski (4) on the automation-augmentation paradox.

In terms of methodology, the task of correction was broken down into different stages to clearly identify where AI would have the most added value.

Student assessment must be approached as a whole in seven successive stages: 1) creation of an exam subject; 2) formalisation of marking criteria; 3) reading of the document produced by the student; 4) identification of key elements; 5) assessment of the relevance of these elements (marking activity proper); 6) production of observations for the student; and 7) verification of the consistency of the assessment process as a whole.

Depending on the teacher, the added value of AI can be sought at one or other of these stages. In this case, AI was used at four of them, on exam papers that had already been marked by professors beforehand and used ex-post to test the ability of this technology to reproduce human marking.

The real contribution of this research (1) lies in the formalization of the so-called ‘8C’ model, which summarizes the ins and outs of an AI-assisted correction process.

Indeed, to embark on the adventure, the teacher must demonstrate: 1) curiosity because nothing forces them to do so; 2) confidence in both the tools and their institution to initiate the process. Then, technically, they will have to face technical challenges that will be invitations to turn back: 3) the composition and format of the copies, paper or digital, 4) the clarification of teacher expectations such as the aptitude for the technique of answering quickly and thespeed in getting to grips with the IA, 5) the comparison between the different tools and the assistance required in getting to grips with them, and 6) compatibility with existing tools, particularly the Learning Management System (LLM). If they have followed the process so far, they will then have to carry out 7)quality control of the overall result and analyse and also consider  8) the cost, in terms of technology but also in terms of time, to see if they want to repeat the experience.

Ultimately, the ‘8Cs’ model proposed by the authors covers this entire process: 1) Curiosity/creativity; 2) Confidence; 3) Composition of the copies; 4) Clarification of expectations; 5) Comparison; 6) Compatibility; 7) Control; and 8) Cost.

The conclusion of this auto-ethnographic research is that, in any case, the teacher will remain responsible for marking and that the future is probably not about to see the task of marking being delegated to an AI tool, but about learning to work with it to become an ‘augmented professor’, capable of working hand in hand with AI.

The second part of the research project on AI-assisted exam correction focuses on the social acceptability of such an approach and its ethically acceptable uses. Based on a questionnaire distributed within the EDHEC Business School, PiLab sought to collect the perception of everyone, whether professor, students or programme managers, regarding the implementation of AI-assisted corrections.

The results are currently being analysed, but one key finding is already emerging from the questionnaire: the students, although regular users of AI, were not very enthusiastic about the idea of their professor doing the same!

What about dissemination?

Beyond this article, the initial conclusions of this research have given rise to several dissemination initiatives. Outside the school, professors have presented their work, for example, at the EURAM conference in Bath, England, last June, or during the Tax Research Network's pedagogical day in Cardiff, Wales (‘Can ChatGPT relieve me from grading my tax exams?’).

The programme is also being disseminated within the school: word of mouth has led other EDHEC professors to get involved and to test AI-assisted marking in their courses.

The originality of the training programme

From an educational and student-focused perspective, it was decided to train all first-year students on the Grande Ecole programme in the legal issues of AI. This will enable them to better understand these issues as students are invited to learn about the AI Act in the form of a case study, focusing specifically on AI-assisted marking.

This piece of legislation is particularly dense, and exploring it, under the guise of a case study, which is very student-relevant,, allows students to avoid getting lost in the meanderings of the many hypotheses and to stay focused.

This case study is timely: the students' mid-semester exam, which took place at the beginning of February, was pre-corrected by an AI tool. Through this case study, EDHEC Business School is taking its obligation of transparency towards students to heart, and for almost two hours, the students analysed all the legal and ethical issues, because to date, the part of the AI Act relating to the correction of exams assisted by AI is not yet applicable!

The strength of this project lies in its cross-disciplinary nature: all those involved in the school are made aware, trained and supported in their exploration of AI through the particular use case of marking exam papers. This seemingly narrow AI focus actually allows everyone to move beyond a purely utopian or dystopian approach to AI, and enter the reality of the challenges of generative AI in education.

References

(1) Daly, P., & Deglaire, E. (2024). AI-enabled correction: A professor’s journey. Innovations in Education and Teaching International, 1-17.  https://doi.org/10.1080/14703297.2024.2390486

(2) 4 questions for Emmanuelle Houet (EDHEC PiLab) on educational innovation and the role of artificial intelligence (March 2025) EDHEC Vox - https://www.edhec.edu/en/research-and-faculty/edhec-vox/4-questions-emmanuelle-houet-edhec-pilab-educational-innovation-role-artificial-intelligence

(3) Molenaar, I. (2022). Towards hybrid human‐AI learning technologies. European Journal of Education, 57(4), 632–645. https://doi.org/10.1111/ejed.12527

(4) Raisch, S., & Krakowski, S. (2021). Artificial intelligence and management: The automation–augmentation paradox. Academy of Management Review, 46(1), 192–210. https://doi.org/10.5465/amr.2018.0072

Pexels.com - Markus Winkler (2025)