According to this teacher, a very specific clue is enough to arouse suspicion.
In classrooms, the question of AI is no longer theoretical. It appears in the copies, in the presentations, sometimes even in the oral presentations. However, contrary to what one imagines, professors do not track down technical proofs. Their approach is more intuitive, almost artisanal, and is based on careful reading of the texts.
For Tiphaine Desprez, philosophy professor, it all starts with a form of habit, acquired over the years of reading student papers. “Whatever the means of detecting the use of AI is always based on experience”she explains. Each copy is part of a continuity: a style, turns of phrase, writing reflexes which evolve slowly but remain recognizable. When this thread breaks, doubt appears. “There is the experience of one’s own students which comes into play, because strangely the brilliant student does not cheat. So in reality we are used to reading what this or that student does, we recognize their writing style.”
This link between a student and his way of writing is central. When an assignment seems to come out of nowhere, unrelated to usual productions, it does not go unnoticed. And this discrepancy is all the more visible as teachers know how to situate each student, including within their limits. A highly mastered sentence can exist, but not in just anyone, nor in just any context. For example, some words may be revealing, not because they are impossible, but because they do not correspond to the student’s actual level. The professor explains that the use of specific terms like “shape”, for example, “puts the flea in your ear straight away” in an abstract argument. This can be enough to arouse suspicion when it appears in a student who never uses them.
The other element that attracts attention is the way in which the texts are constructed. Productions resulting from an AI often present a very clean organization, almost too regular. Ideas flow smoothly, arguments are balanced, and the tone remains consistent from start to finish. “The structure is very typical, the tone is very non-controversial, they always conclude by saying that it is multifactorial”observes Tiphaine Desprez. Some students try to correct this by asking the AI to introduce errors, thinking it will make their copy more credible: “This is why students ask the AI to add mistakes, but even if the AI only makes grammatical mistakes, that’s not enough.” Clumsiness, too, remains mechanical.
As the teacher summarizes, “in fact, even excellent, hyper-serious students still write like 17-year-olds.” This is precisely where the unmistakable detail lies. A 17-year-old student, even a brilliant one, writes with a thought in the process of being constructed, with sometimes hesitant choices of words, somewhat clumsy formulations, repetitions which betray the progression of ideas. “It’s deep, it’s thoughtful, but the choice of words is always a little clumsy, implicit, repetitive. We see that it’s thought in the process of being constructed.” Conversely, a copy generated by an AI gives the impression of an already finished, smoothed text, where each sentence seems definitive.
