AI Cheat Sheets: Graders Struggle to Spot Bots in Exams

Researchers at the University of Reading discovered that AI-generated exam answers were mostly undetected and often received higher grades than student submissions. The study highlights the need for educators to reassess how exams are conducted and evaluated in the age of artificial intelligence.

Exam graders face increasing difficulty in identifying answers generated by AI-based chatbots, according to a recent study by researchers from the University of Reading. The study revealed that AI-driven answers not only went largely unnoticed but were also graded higher than those authored by students.

The research team conducted the study using submissions from 33 'fake students,' generated by ChatGPT and submitted to the School of Psychology and Clinical Language Sciences at the same university. Alarmingly, they found that 94 per cent of AI-generated answers remained undetected and that 83 per cent of these obtained better grades compared to real student submissions.

'We found that within this system, 100 per cent AI-written submissions were virtually undetectable and very consistently gained grades better than real student submissions,' the authors noted in their study published in the journal PLoS ONE.

The researchers urged that these findings serve as a 'wake-up call' for educators globally, emphasizing the imperative for the education sector to evolve against the backdrop of advances in artificial intelligence.

'Many institutions have moved away from traditional exams to make assessment more inclusive. Our research indicates it is internationally important to understand how AI will affect the integrity of educational assessments,' said lead researcher Peter Scarfe, an associate professor at the University of Reading.