Researchers from the University of Reading secretly submitted exam answers generated by artificial intelligence, successfully fooling professors and achieving better results than real students. The project used ChatGPT-4 to generate unedited answers and submitted them under fictitious student identities to online assessments for undergraduate courses.
Image source note: The image was generated by AI, authorized by the service provider Midjourney
Surprisingly, the university examiners only detected one out of 33 answers as coming from AI, with the other AI-generated answers scoring even higher than the average student performance. The authors suggest that this finding indicates that AI processors like ChatGPT have now passed the "Turing Test," meaning they can pass without being detected by experienced markers.
The study concludes that AI's abstract reasoning capabilities are continually improving while its detectability is decreasing, posing a serious challenge to the integrity of educational assessments. Experts point out that this research heralds the end of take-home exams or unsupervised coursework. They call for deeper consideration within the education sector about the use of AI in teaching evaluations and seek solutions to avoid a crisis of trust.
Elizabeth McMullan, Deputy Vice-Chancellor for Education at the University of Reading, stated that the university is gradually phasing out take-home online exams and developing alternative solutions, including applying knowledge to real-life and work scenarios. She also noted that some assessments will support students' use of AI, teaching them to use it critically and ethically, and cultivating skills needed in the modern workplace. However, experts also worry that allowing students to use AI in exams may lead to a decline in their skills, potentially losing their ability to think, analyze, and write independently.
Key Points:
⭐️ University of Reading researchers successfully fooled professors, raising concerns in the education sector about the integrity of assessments
⭐️ AI's abstract reasoning capabilities are increasing, but its detectability is decreasing, posing a challenge to the integrity of educational assessments
⭐️ Universities are gradually phasing out take-home online exams; experts worry about a decline in students' skills and call for deeper consideration of AI use in education