Students increasingly use Large Language Models (LLMs) to complete academic assignments, including essays and thesis projects. Educational institutions are attempting to counter this trend by implementing AI detection tools to identify machine-generated content.
The reliance on AI detectors creates a 'reward hacking' scenario where students focus on modifying text to bypass detection software rather than engaging with the course material. This process often results in lower-quality writing, as students intentionally introduce stylistic errors to mimic human patterns.
The author argues that AI detectors provide a false sense of academic integrity, allowing faculty to avoid the time-consuming process of verifying a student's actual understanding through oral examinations or individual questioning.
To address the issue, the article proposes a shift toward mandatory in-person assessments. Under this model, students would complete practical tasks—such as engineering calculations or data analysis—in real-time under the supervision of a commission, without internet access.
The author suggests that this approach to evaluation should be implemented starting from school and continuing through university. However, such a transition requires significant increases in teaching staff, time, and institutional effort compared to the current reliance on automated detection.
