AI essays beat human performance in university exams, reveals study

Using ChatGPT, an AI tool, the study crafted 33 fictitious student profiles and found that AI-written essays scored half a grade higher than human submissions

Artificial intelligence (AI) has proved its prowess in academia, surpassing real students in university exams, according to a recent study by the University of Reading. 

Researchers found that AI-generated answers in undergraduate psychology exams consistently outperformed those from actual students, often slipping past markers without detection.

Creating 33 fictitious student profiles, the study used ChatGPT, an AI tool, to craft responses for module exams. Results revealed that AI-written essays scored, on average, half a grade higher than human submissions. Alarmingly, 94 per cent of these AI essays raised no suspicions among markers, highlighting their near-invisible footprint in assessment processes. The study, published in PLOS ONE, suggested that the 6 per cent detection rate might overestimate AI detection capabilities.

Associate Professor Peter Scarfe and Professor Etienne Roesch, who led the study, stressed the implications for educational integrity worldwide. Dr Scarfe noted the shift away from traditional exams to foster inclusivity, but cautioned against AI’s potential to compromise assessment fairness.

While reverting to handwritten exams isn’t practical, Dr Scarfe urged swift global action to address AI's evolving role in education. The research found AI excelled in early undergraduate modules but lagged in more complex, third-year exams requiring abstract reasoning.

The study also emphasised on the broader concerns about AI’s impact on education. At Glasgow University, in-person exams were reintroduced for select courses in response to such challenges. A separate study reported by The Guardian revealed that while many undergraduates use AI to aid essay writing, only a minority admit to directly inserting unedited AI content into assessments.

The University of Reading’s findings serve as a wake-up call for educators, calling for a re-evaluation of assessment strategies amidst advancing AI technologies. As AI continues to evolve, ensuring the integrity and fairness of academic evaluations remains a critical priority for the education sector.