To deceive AI detectors, students add mistakes to their homework
The goal? That homework is “Natural” With faults added to the texts generated by AI. The technology is so trivialized in the academic environment that students redouble their imagination to avoid detection, even if it means doing it manually.
Voluntary faults to avoid AI detectors
Other users ask AI to simplify their style. In a Tiktok video, a student explains that she asks her chatbot to write “Like a somewhat stupid first -year student”.
Second year student in Stanford, Eric explains to New York Magazine That his comrades are “really good at manipulating the systems”. “We submit a chatgpt setpoint, then transfer the result to another AI system, then to another one. At this stage, if it is passed through an AI detection system, the detected AI percentage decreases at each stage.”
The situation is all the more ironic since students make efforts to modify their duties generated by AI instead of using this energy to write themselves. Teachers are also concerned about this motivation to cheat with these tools.
Sam Williams, teaching assistant at the Iowa University, explains: “They use AI because it is a simple solution and an easy way for them not to spend time writing dissertations. And I understand, because I hated writing dissertations when I was a student.”
Sam Williams noted a striking change between the first homework of his students and their following work on the history of jazz in New Orleans. The texts contained rough mistakes such as the mention of Elvis Presley which has never made jazz or even belonged to the city’s music scene.
“I literally said to my class: ‘Hey, don’t use AI, but if you are going to cheat, do it intelligently. You can’t just copy exactly what it produces'”explains Sam Williams. For him, the observation is bitter: “Whenever they encounter a little difficulty, instead of overcoming it and learning from it, they fall back to something that makes their task a lot.”
