Researchers at the University of Reading conducted a study where they secretly submitted exam answers generated by AI, tricking professors into giving higher grades than real students without their knowledge.
In this project, fake student identities were created to submit unedited responses generated by ChatGPT-4 in an online assessment for an undergraduate course.
University graders, unaware of the project, only flagged one out of 33 responses, with the AI-generated answers receiving scores higher than the students’ average.
The study revealed that AI technologies like ChatGPT are nearing the ability to pass the “Turing test”, a benchmark for human-like AI performance without detection.
Described as the “largest and most comprehensive blinded study of its kind,” the authors warn of potential implications for how universities evaluate students.
Dr. Peter Scarfe, an author and Associate Professor at the University of Reading, emphasized the importance of understanding AI’s impact on educational assessment integrity.
The study predicts that AI’s advancement could lead to increased challenges in maintaining academic integrity.
Experts foresee the end of take-home exams and unproctored classes as a result of this study.
Professor Karen Yun from the University of Birmingham highlighted how generative AI tools could facilitate undetectable cheating in exams.
The study suggests integrating AI-generated teaching materials into university assessments and fostering awareness of AI’s role in academic work.
Universities are exploring alternatives to take-home online exams to focus on real-life application of knowledge.
Concerns arise regarding potential “de-skilling” of students if AI is heavily relied upon in academic settings.
The authors ponder the ethics of using AI in their study and question if such utilization should be considered cheating.
A spokesman from the University of Reading affirmed that the research was conducted by humans.
Source: www.theguardian.com
