TLDR:

• Researchers at the University of Reading used AI-generated exam answers to fool their own professors.
• AI-generated answers received higher grades than real students, raising concerns about academic integrity.

Researchers fool university markers with AI-generated exam papers

Researchers at the University of Reading conducted a project where they submitted AI-generated exam answers that went undetected by their professors. The project involved creating fake student identities to submit unedited answers generated by ChatGPT-4 in take-home online assessments for undergraduate courses. Surprisingly, only one out of 33 entries was flagged by the markers, with the AI answers receiving better grades than real students.

The study concluded that AI processors like ChatGPT have passed the “Turing test,” indicating their ability to go undetected by experienced judges. The implications of this study on how universities assess students are significant, as it raises concerns about the integrity of educational assessments in the face of AI advancements.

Experts believe that allowing the use of AI in exams could lead to “deskilling” students and dependency on machines for critical thinking and analysis. The study authors also suggest that universities should consider incorporating AI material generated by students in assessments and prepare students to use AI critically and ethically.

In response to the findings, the University of Reading is moving away from using take-home online exams and exploring alternative assessment methods that include real-life application of knowledge. The study highlights the need for the education sector to adapt to the growing influence of AI in academia and society at large to maintain academic integrity.