Today, the word on everyone’s lips is ChatGPT, particularly in the world of higher education. Since its launch at the end of last year, the new AI tool, available as a free online service, has been tested by many, including myself. We can be confident that it will be used extensively by our students in the future, certainly as a possible tool to support their work but also most likely to replace their own work in exams – or, more clearly, to cheat.
So how should this be handled? The tools currently available to detect cheating, such as Ouriginal, are completely ineffective here because this is not plagiarised text but completely new writing – almost comparable to exam papers purchased from ghostwriters. It also means that our disciplinary committees will have as much difficulty dealing with these cases as previous ghostwritten exams and theses – how do you prove that the student did not actually write the text?
For my part, I think there are several possible approaches, and that a combination of measures is needed. For one, the team behind ChatGPT also developed a GPT Detector, (https://huggingface.co/openai-detector/), which gives a percentage value for the probability that the text in question was produced by ChatGPT. The problem here is that the results from this cannot be dealt with in a disciplinary committee without in-depth analysis. Even if the probability is “99.9% Fake”, no one can assess how the detector arrived at this result. However, each teacher also needs to use their own judgement. While ChatGPT can produce both logically coherent and linguistically acceptable answers, for those with a deeper understanding of the subject it may still be apparent that the answers produced by the chatbot are pure nonsense.
It will be easier to judge in cases where the questions are advanced. However, if they are elementary questions at the beginner level, the probability of detecting cheating is much lower. So, first and last, we need to review our examination forms, where take-home exams at the beginner level, if they need to exist at all, need to be supplemented with other types of examination that preclude the use of AI tools. The Senior Management Team has, of course, initially discussed the matter, but that discussion needs to be broadened and must also be addressed across the core activity going forward.
This article is written by Astrid Söderbergh Widding, President of Stockholm University. It appears in the section ”Words from the University’s senior management team”, where different members take turns to write about topical issues. The section appears in News for staff which is distributed to the entirety of the University staff.