In a recent study, university exams taken by fictitious students using artificial intelligence (AI) outperformed those by real students and often went undetected by examiners. Researchers at the University of Reading created 33 fake students and employed the AI tool ChatGPT to generate answers for undergraduate psychology degree module exams.

The AI-generated responses scored, on average, half a grade higher than those of actual students. Remarkably, 94% of the AI essays did not raise any suspicion among markers, with only a 6% detection rate, which the study suggests is likely an overestimate. These findings, published in the journal Plos One, highlight a significant concern: "AI submissions robustly gained higher grades than real student submissions," indicating that students could use AI to cheat undetected and achieve better grades than their honest peers.

Associate Professor Peter Scarfe and Professor Etienne Roesch, who led the study, emphasized the need for educators globally to take note of these findings. Dr. Scarfe noted, "Many institutions have moved away from traditional exams to make assessment more inclusive. Our research shows it is of international importance to understand how AI will affect the integrity of educational assessments. We won’t necessarily go back fully to handwritten exams - but the global education sector will need to evolve in the face of AI."

In the study, the AI-generated answers and essays were submitted for first-, second-, and third-year modules without the knowledge of the markers. The AI students outperformed real undergraduates in the first two years, but in the third-year exams, human students scored better. This result aligns with the idea that current AI struggles with more abstract reasoning. The study is noted as the largest and most robust blind study of its kind to date.

Academics have expressed concerns about the impact of AI on education. For instance, Glasgow University recently reinstated in-person exams for one course. Additionally, a study reported by the Guardian earlier this year found that most undergraduates used AI programs to assist with their essays, but only 5% admitted to submitting unedited AI-generated text in their assessments.