With the rise of artificial intelligence, an overwhelming majority of university educators can't tell the difference between a student's work and an answer produced by ChatGPT, a recent British study shows.
Researchers at the UK's University of Reading have demonstrated that experienced exam markers find it extremely difficult to distinguish between their students' actual work and answers generated by artificial intelligence. Their findings are published in the scientific journal PLOS One.
To find this out, the researchers submitted the results of a test on various undergraduate psychology modules to lecturers, their sample mixing real answers with those written by ChatGPT. In 94% of cases, the AI-generated answers went undetected and, in most cases, even achieved higher grades than real student submissions. This situation raises crucial questions about the integrity of university assessments.
With the rise of generative AI, students now have tools capable of writing essays and a host of other academic assignments. This practice is all the more widespread as home-based exams and assessments, by definition unmonitored, have multiplied since the Covid-19 pandemic.
It goes without saying that if teachers are unable to distinguish authentic work from that produced by AI, the value of qualifications and confidence in educational institutions may one day be called into question.
In order to mitigate the effect of ChatGPT on students' grades, and in the face of teachers' dismay, the authors of the study recommend a return to in-person exams. Failing that, they recommend stepping up the monitoring of online exams, even if the tools for detecting AI in students' answers are not yet necessarily very reliable.
OpenAI provides students and teachers with ChatGPT Edu, a version of its intelligent conversational agent specially designed for universities. The aim is for generative AI to help students, faculty staff and researchers with their studies, not their exams. – AFP Relaxnews