Teachers have been using AI detection tools for several years, but the British higher education regulator gave it a not-so-good review: unreliable.
The Office for Students (OfS) recently released a report, concluding that existing detection tools cannot reliably determine whether students use AI to complete assignments, and there are many false positives. As the model becomes stronger and stronger, the difficulty of identification will only become higher.
Therefore, its suggestion is not to "buy a better detector", but to change the examination method: re-examine unsupervised assessments such as home papers and open-book examinations, and gradually shift to supervised assessment.
Many universities are already doing this: offline written examinations, oral examinations, and process assessments. For example, when writing a paper, you don't just look at the final product, but you also look at the literature review, research design, and mid-term report. If you can explain your reasoning clearly in person, it will be more difficult to fake than submitting a beautiful article.
The report doesn't shy away from side effects either. Closed-book exams are out of touch with the real work environment. In reality, no one will let you check information or use tools. In addition, for some students with disabilities, on-site examinations will cause new injustices.
There is also a more heart-wrenching observation: Many students already believe that using AI is an effective way to get high scores, and they will suffer if they don’t use it. The capability gap between paid AI and free AI may also become a new educational inequality.
The problem in schools is changing from "how to catch AI cheating" to "how to make the test meaningful." This is actually a question that should have been asked earlier.