The rise of artificial intelligence in education has sparked a growing concern about plagiarism and academic integrity but can we really tell when student work is AI-generated? Gary Henderson shares some insights
Plagiarism is defined as “the practice of taking someone else’s work or ideas and passing them off as one’s own” and has long been an issue within education especially since the age of the internet and Google. For some students, and I would suggest these are a minority, there is a wish to minimise the effort they need to expend in getting to the academic outcome they seek, be this a GCSE, a B-tec, an A-Level or a degree qualification. And I get this to some extent. We live in a world of convenience, with next day delivery, on-demand TV, on-demand taxis (e.g. Uber), takeaways and home delivery shopping from our supermarkets, to name but a few of the conveniences which now pervade our lives. Why would a student choose to undergo the stress and difficulty of creating a piece of coursework when AI can do it in a fraction of the time, with no real effort, allowing the student to focus on other things?
Looking to the JCQ guidance we see mention of the need for the work to be the student’s “own” work. It needs to be original, but how possible is it to be original where the amount of data produced yearly is estimated to be 9000% higher than it was in 2010, with this only set to grow through generative AI. There is greater and greater potential for similarities to exist coincidentally never mind where it happens purposefully. And what does “own” work mean? If a student used a grammar checker or spellchecker, is this ok, but what if they used a more complex AI powered tool such as Grammarly or the Editor built into Microsoft Word where the student now gets recommendations as to phraseology, tone and more; Is it still their own work? So maybe we need to start by considering what originality and a student’s “own” work actually means in this world of technology and tools designed to assist and help.
Can we tell when work is AI generated?
Let’s briefly assume AI use isn’t to be permitted, and I am not suggesting this is the stance we should take. Our next challenge would be to detect where students are using it and presenting work that is not representative of their knowledge and understanding. The problem here is that the whole purpose of AI is to create human like outputs. An AI solution will be judged by its ability to replicate what humans can do. So how can we honestly expect to detect a tool where its principal aim is to behave like we do. Large Language Models have a temperature variable built into them which allows for the predictability of the responses to controlled with one study identifying the trade-off between low temperature values and “boring responses” and a higher temperature leading to “more creative, human-like responses”. This implies the importance, even in fact seeking business applications never mind educational applications, of more human-like responses.
So, we can’t accurately detect AI, no matter what the flashy sales info from vendors says regarding their AI detection tools. In fact, most of these tools present their findings as a probability that a student has used AI rather than as fact. How would we treat a 50% likelihood that AI was used? What if AI was used simply to help a student with dyslexia or where English is an additional language; Would this be ok and how would we identify how it was used? And what about when the AI detection tool gets it wrong and a student is wrongly accused, or maybe even proceeds to sue the school as has happened in the US? Or what if a student’s future suffers through a lost scholarship or a university place as the result of an 81% chance that they may have used AI, but where this could be ChatGPT, or it could be Grammarly or even the Editor built into Word? For me AI detectors ask more questions than they answer so I tend to shy away from them.
What now?
I suspect at this point I have posed more questions than I have answered however I think it is important that we think about these challenges. We need to think deeper than “how do we stop students cheating using AI” or “how do we detect AI”. We need to consider what we are trying to achieve through coursework and maybe consider if there is another way as AI is here to stay.
Be the first to comment