A study conducted by researchers at the University of Reading has shown that artificial intelligence (AI) can beat real university students on exams. The researchers found that exams ended up being not only superior to real student-taken personalized tests but also largely unnoticed by markers as fakes.
Overview of Study To do so, the investigators designed 33 fake learners and registered them in an undergraduate psychology degree module using ChatGPT – an AI software – to write answers on test paperwork inside a UK College Cumming School report. On average, the AI-generated answers scored half a grade boundary higher than authentic student submissions. Astonishingly, 94 percent of the markers could not identify any concerns with the AI essays, making them almost imperceptible. In other words, the 6 percent detection rate is probably on the high side of what they found, according to their findings published in this week’s issue of the journal PLOS ONE.
Implications for Practice & Conclusion The results of the study come as a shock since AI submissions have outperformed real student submissions across the board, at an alarming rate. In other words, students could potentially use AI to cheat without getting caught and earn a higher grade over their honest classmates. Associate Prof. Peter Scarfe and Prof. Etienne Roesch, who co-led the study, stressed what their findings mean to educators around the globe. Dr. Scarfe said: “Many institutions have followed our lead and dropped traditional exams to make their assessments more inclusive, but this study suggests we need to understand the implications of AI for educational assessment security.”
Challenges and Considerations The answer to this is certainly not going back to handwritten exams. But as AI continues to progress, the global education sector will require a strong solution sooner rather than later, Dr. Scarfe said. During the study, AI-produced exam answers and essays were submitted to anonymous academics in 1st, 2nd & final year modules. The AI managed to outperform real students in the two first years, but not in its third year. This disparity is reflective of the lack of more abstract reasoning present at the moment with AI, which they point out as key for success on a third-year exam.
Academic Reactions The use of AI in the educational field has left many academics worried. One of their courses updated to include in-person exams, as seen at Glasgow University. A study reported by The Guardian earlier this year also showed that although most undergraduates use AI to assist with their essays, only 5 percent admitted to pasting unedited text produced by AI into their assessments.
Call to Action The University of Reading has dubbed their findings a “wake-up call” for educators in the age of AI: “These results represent a wake-up call to modern assessment and exam boards around the world as they look towards implementing novel forms of assessment.” As AI continues to advance, the education sector must adapt to ensure the integrity and fairness of academic assessments. This re-adaptation could include devising new ways to detect AI-generated writing, restructuring assessment structures, and even curriculum components to ensure that advanced education shall see the light of day in a world with copious amounts of artificial intelligence.