The teaching staff understands this well: middle school, high school and even university students are increasingly using artificial intelligence to write their homework for them. But some teachers have found an unstoppable method to push cheaters to betray themselves.
All teachers know it, from middle school to high school, but also up to university: artificial intelligence has taken a significant, even downright invasive, place in classrooms and educational programs. The days when cheating was limited to copying from your friend or hiding a cheat sheet in your pencil case seem to be over. Today, students mainly use AI to help them with their revisions, to better understand and assimilate a notion of the program, to learn a lesson… Used wisely, the tool can prove to be a valuable aid for students, provided they do not give in to the easy way out. Because of course, some also use it to write their homework in full for them.
As software becomes more and more efficient and more and more “human” in the way it is written, it is sometimes difficult for a teacher to prove with certainty that a student has used it. Traditional AI detectors are often unreliable, producing false positives or, conversely, passing artificially generated text. But some teachers have found an unstoppable trick to punish cheaters: a formidable trap, because it is totally invisible to the naked eye, but which pushes the artificial intelligence to betray itself… without the student even realizing it.
This method is that of the “Trojan horse”: more and more professors are using it to unmask dissertations and other work written by generative AI. This is notably what Will Teague, professor of history at the university, did: as he tells HuffPostso he set a trap for his students to find out how many of them use AI to write their papers. “I inserted hidden text in the instructions for an assignment, text invisible to students, but visible to ChatGPT”he explains.
But no need to be a computer code pro: this method simply consists of slipping, in the middle of normal instructions, a white text on a white background, giving a completely different instruction to the AI so that it can be identified immediately. So, if the student copies and pastes the topic into ChatGPT without rereading it, the AI also receives the “invisible” text. For example, the teacher can choose to give a hidden instruction to necessarily mention the words “Banana” or “Frankenstein”: the AI will do it and the student, who may not have taken the trouble to reread everything, will denounce himself without even realizing it. As a result, Will Teague discovered that a large part of his students had cheated: out of a total of 122 assignments collected, 47 of them “were at least partially written by an AI”or nearly 39% of the work.
Unfortunately, this method may no longer remain a secret from students for long… at least not for fans of Simpson. Yes, a new episode of the series tells exactly how the “Trojan horse” trick works, after Bart used AI to get a good grade at school. It remains to be seen who, teachers or students, will win the next round of this technological battle.
