Faced with the massive use of chatgpt by students, the issue is no longer detection but learning to use critical and creative use of artificial intelligence tools.

In teaching, but not that, cheating has always existed. But with generative artificial intelligence, we have changed dimension for a few months, the conferences and educational seminars are linked. The generative AI is invited in all the debates, in all the rooms, on all the slides. And very quickly, a concern arises, always the same: how to detect the cheating of students with Chatgpt?

The expected answer: miracle software. An algorithmic defense line. As if the real challenge was whether the duty had been written by a student or by a machine.

AI in teaching: educational fracture?

The figures speak. Or rather shout. More than 90% of students regularly use AI as part of their studies. In terms of teachers, we are rather in the range of 10 to 20 % of real uses. Alain Goudey, deputy director general at Neoma, underlines this in “AI at the service of marketing” (Dunod): “Although 100 % of participants know Chatgpt, only 10 % use it daily.” An observation taken from his exchanges with many teachers during his conferences.

The reality is that students better master the tools than those who note them. And you have to be lucid: a student with some prompt bases can produce a perfectly undetectable copy, sprinkled with some voluntary faults. Add the online tutorials to that to bypass detectors, the tools that reformulate a text generated to make it “human-compatible”, and the algorithms are completely duped. Seeking to “detect” is like looking for reliable information on Chatgpt: it’s random, and the illusion is (often) perfect.

The copies that shine (or not)

As an intervening in business school, I corrected memories and I start to distinguish two main categories of production. The brilliant, well written, clear, structured copies, with solid references. A sometimes impressive level. Better than what I could have produced myself at their age. Is it helped by AI? Probably. Is this a problem? No, as long as the content is thought, piloted, nourished by a human.

And then the others. Copies with impeccable grammar, but artificial style. Extension sentences. Omnipresent adverbs. Ready -made expressions. The absurd rehearsals of “crucial” and “essential” (the record: 44 in 32 pages). And above all: bibliographies … issues. Or rather: Webliographies.

Sources from dark blogs, articles without author, pages that do not cite any serious reference, and sometimes even … Shadow links. But where are the books? No need for software to understand: the text is not the fruit of a reflection of human neurons, but a copygpt.

So what do I do, as an evaluator? On the form, the copy is clean. Basically, it is hollow, even doubtful. But should I, for each duty, spend hours checking each source? To analyze the consistency of each argument? To cross each quote? It’s unrealistic. So what do I do?

Experience differently with AI

For some homework, I tested a hybrid approach: human × IA. I divide the class into two teams: one writes an indictment against the use of AI in communication and marketing, the other builds a plea in its favor. The whole thing ends with a contradictory debate, where everyone must defend their position.

Before I start, I provide them with a text generated by Chatgpt, produced in a few minutes from my prompt. An average-up copy, somewhere between serious application and stylish laziness. And I throw them: “This is what AI can do. It’s up to you to do better, with the help of AI.” Their mission? Identify the faults. Better offer. Add value. Bring their vision, their critical thinking. I do not need to know the medieval opinion of Chatgpt, I already know him. Make me think, even if I don’t agree with you. There, it will be successful.

Result: a real intellectual dynamic with AI. Some students are revealed. Others include: “Our role is not limited to correcting Chatgpt”, explains Emeline Vaquero, a master’s degree in communication manager. And to add: “This is an exercise that we all have been particularly appreciated since it asked us to express ourselves on a subject that concerns us all the more by our business of communicators. This debate offered us a moment of reflection on the advantages and issues of AI while showing open -mindedness to welcome opposing opinion.”

Patrick C., educational manager of Sup’De Com Nantes and member of Jury explains to me: “The proposal generated by Chatgpt was of a completely correct level. But what really surprised me is that the students did not try to improve it. They rather took another path, explored other ways of argumentation. Their speeches were different, embodied. And the debates that followed … Passionate, and fascinating to listen.”

And above all, the message gets through: why would a business hire you to do what AI can already do alone?

Train centaurs: rethink, not detect

The real challenge is not to know if students cheat with AI, but to support them in intelligent, effective use and, dare to promote it, ethical. Companies are waiting for professionals capable of structuring a thought, being critical and choosing the right tools with discernment. What is needed are centaurs: half-humans, half-tool students, capable of combining reasoning, creativity and productivity. Not prompt performers. Not knowledge counterfeiters.

Generative AI is a revealer. It obliges us to reconsider the validity of our educational formats, our evaluation methods and our very design of skills. If a student can make duty without understanding anything about its content, it is because our system has a problem. And if our only answer is detection software, then we have already lost.

This observation is just as much for the company. Again, AI produces content, automation of tasks. But the real question is not: “Who uses AI?” She is: “Who knows how to make it an added value lever?” A good professional will know, according to the context, to delegate to the machine and to keep for him what requires judgment, analysis and nuance.

So what good is it to run after cheaters made in IA? Let’s form students to do better than AI. It is possible, provided they give them the means, to set them their requirement and to teach them to think with AI, without submitting to it.

The real question is not: “How to prevent students from cheating with AI?” But: “How to get them to do better than AI and with AI?” And to this one, no software will answer for us.