Recent research by Peter Scarfe and his team at the University of Reading, UK, has unveiled that a staggering 94% of university exam submissions created using ChatGPT went undetected as AI-generated content.
Even more surprising is that these AI-generated submissions often outperformed real students' work. This article delves into the study's findings, its implications for academic integrity, and potential solutions for the future. Start reading to understand the details!
The Study: Methodology And Findings
Scarfe's research team used ChatGPT to generate answers for 63 assessment questions across five psychology modules. The key elements of their methodology included:
- Real students taking exams at home.
- Students could reference their notes but were theoretically prohibited from using AI.
- AI-generated answers were mixed with real student submissions, constituting about 5% of the total scripts graded by academics.
The results were interpreted as both alarming and enlightening. Only 6% of the AI submissions were flagged as potentially not being a student's own work. On average, the AI-generated responses received higher grades than real student submissions, although performance varied by module. In some, none of the AI-generated work was flagged.
Scarfe observed that while current AI struggles with abstract reasoning and integrating information, there is an 83.4% chance that AI-generated work will outperform student submissions. It indicates a significant gap in current academic assessment methods.
Academic Integrity Under Threat?
Thomas Lancaster from Imperial College London highlighted the vulnerability of unsupervised assessments to AI-generated cheating. The increasing workload on academics further complicates the detection of AI-generated content.
According to Lancaster, generative AI can produce plausible responses to simple questions, and time-pressured markers are unlikely to raise AI misconduct cases impulsively.
The Need For Rethinking Assessments
Scarfe emphasized that tackling this issue at its source—preventing AI use in assessments—is nearly impossible. Instead, he suggests rethinking assessment methods to incorporate AI: "We're going to have to be building AI into the assessments we give to our students," Scarfe said. He also claims that the academic sector must collectively acknowledge and address this evolving challenge.
As the academic sector grapples with AI-generated submissions, embracing AI within assessments and rethinking evaluation methods seem like crucial steps. It shows that it might be time for academia to evolve in response to the digital age.