Two years after ChatGPT became freely available, many students have learned to use it, sometimes as a simple writing aid, sometimes to write entire assignments for them. A development which is already forcing teachers to adapt their correction or their exercises.
“I filmed the AI doing the homework for me and posted the video on Snapchat. Obviously, others were interested!” Tom* is in 1st grade when generative artificial intelligence (AI) bursts into the news with a bang, with the release of ChatGPT in November 2022. A program that can instantly write credible and coherent texts on any issue . And the potential of which the student easily grasps: “I was bored in physics class, I gave the exercise to the AI and it gave me an answer… I quickly said to myself that it could help me!”
Like him, many budding cheaters have gradually taken up the tool, and not only in higher education. In high school and even college, AI already acts as a crutch, a luxury assistant, or even a little more for certain students who are ready to do anything to get the best grades. And poses problems for teachers that are sometimes unanswerable.
“I had heard about it when it came out, but didn't think to try it, then I saw more and more videos on TikTok doing it”remembers Dan, just out of final year. How many students followed this example? “For text comments in philosophy, half the class used ChatGPT!”trumpets the recent high school graduate. It is impossible to precisely quantify the phenomenon, because there are no studies on this subject concerning middle or high schools in France. But also because AIs can be used in more subtle ways than asking them to do all the work. “I sometimes use it to review and practice, says Colombe, a second-year student in Paris. By generating standard questions before an exam, by asking them to summarize the themes to remember…” Nothing reprehensible, in this case.
The use of AI becomes more questionable when it is used to solve an exercise intended to be corrected and graded by a teacher. Rémy, that the school has “very quickly inflated”concedes having it “not badly used in French and in philosophy” in high school, particularly when faced with the nightmare of many schoolchildren: the essay. “I gave him as much detail as possible about our lesson and I asked him for a structuresays the student. He’s a kind of co-pilot.”
“But all of this remains my personal work”considers Rémy. He swears that he wrote the final text himself and checked each time the veracity of the information generated by the AI. According to him, the method bore fruit: “As I used it, I was able to pick up on certain ideas or plans during checks [en classe]and to better understand the exercise.”
There remains the most dishonest use, where the student's intention to cheat – sometimes guided by his parents – is beyond doubt: the copying and pasting of a text produced by this assistant. “For multiple choice questions, it's simple, just put the title in ChatGPT and take the answers”assures Tom. But the AI may be wrong. “We used it with my son in sixth grade for an exercise in which we had to associate a fruit with each planet in the solar system, while respecting the ratio of proportions”says Nicolas, parent of a student. “We spent the day without getting there…”
“ChatGPT gives an answer like any other. In this case, it was wrong. But we had no guarantee of doing better!”
Nicolas, parent of a 6th grade studentat franceinfo
This use is even more risky for writing exercises, where the copying and pasting of a long text, which is too crude, is easily spotted by a minimally informed teacher. “Signs can alert usassures franceinfo Maxime, a professor of digital sciences and technology in a high school in the Rhône. When the student gives ten sentences of response to a question that only called for one, without any spelling mistakes when he usually makes them. Or when he asserts facts or figures for which the source cannot be found anywhere.” The students interviewed by franceinfo mention numerous cases of classmates being unmasked with their responses written to the nearest comma by ChatGPT. This is particularly easy in college, because “AIs write with 'grown-up words'”observes Nicolas, the parent of a 6th grade student.
But some ChatGPT enthusiasts have also learned to refine their method. Double-check what the AI writes, give it as much context as possible in the initial instructions, and above all “do things in stages”, assure Tom : “For a dissertation, I first ask for issues, then an intro, then an adapted plan, then ideas, sub-part by sub-part…” The high school student then pushes the vice to the point of adding turns of phrase of his own, “and even spelling mistakes”.
Faced with these digital tricks, teachers have no choice but to remain on the lookout. Oral presentations in class are a good indicator for detecting excessive use of AI. “Students had read me a text of which they obviously did not understand everything, with a suitable plan, but extremely generic arguments and the vocabulary of a tired academic”remembers Jean-Marie, digital science and technology teacher in a high school in Ile-de-France.
“They didn't understand the subject. I asked them questions, they couldn't answer… And they didn't deny having used ChatGPT.”
Jean-Marie, teacher in a high school in the Ile-de-France regionat franceinfo
In fact, the teachers interviewed by franceinfo describe a climate of constant suspicion at the time of corrections. At the risk of making mistakes, because it is impossible to prove by A+B that a student has used a text generative AI. “The ‘AI detectors’ on the market are unreliable”recalls Philippe*, SVT teacher and digital referent in a college in Isère.
Thus, a high school student from Haut-Rhin who went before the disciplinary committee last summer, suspected of having used an AI during the philosophy baccalaureate, was finally exonerated. The correctors had used “basic software (…) which generates false positives”argues the student's mother to The Republican East. ChatGPT developer OpenAI has well-designed a system that accurately identifies its creations, according to the Wall Street Journalbut he refuses to make it accessible to the public.
Many teachers therefore feel helpless. “Those who are interested in it often have to train themselves, and the regular changes of minister do not help”deplores Jean-Marie. The Ministry of National Education has produced resources (PDF file) to help teachers understand generative AI and talk about it to students… but on what should be allowed or prohibited, the answers remain unclear. “Undeclared or unauthorized use of AI tools to produce work for educational or assessment purposes (…) could be considered a form of misconduct or even educational fraud”, explains the Paris Academy on its site. A conditional that doesn't help much. Contacted by franceinfo, the Ministry of National Education did not specify whether there was a list of uses of AI prohibited for students at the national level.
Professors are not all closed to the use of generative AI, as long as it is reasoned. “If the students know how to explain what they are submitting, that’s fine with mesummarizes Jean-Marie. My students had to create a site with HTML code. One of them gave me an impressive site that didn't seem to me to match his capabilities. I asked him to explain and I saw that he mastered the code he had submitted.”
But in reality, this control of the final result is rarely guaranteed. Which pushes some teachers to introspection. “If my exercise could be done stupidly by an AI, it was not very interesting”believes Philippe. “It calls into question the good old homework assignments given in spades, which not all teachers are ready for,” admits Jean-Marie.
Especially since the students are already not equal when it comes to these exercises. “Some have family to help them, and there was already the internet and friends to copy and paste”recalls Louann, French teacher in 6th and 5th grade in the Nantes region. “AI will cause the same problems as Wikipedia in its early days, but to the power of 1,000”believes Jean-Marie. Between students and teachers, the game of cat (GPT) and mouse is far from over.
* First names have been changed.
