Artificial intelligence has “blurred the line” between what constitutes academic support and what should be seen as misconduct, necessitating a rethink on what is considered cheating, according to a new study.
A fifth (22 per cent) of students surveyed for the paper, “How vulnerable are UK universities to cheating with new GenAI tools?”, admitted using AI to cheat in their assessments in the past 12 months.
But how students said they had cheated with AI varied. While one simply said they had asked ChatGPT to “write my stuff lmao [laughing my ass off]”, others displayed more nuanced uses. One told the study that they had used it to “ask questions when I was stuck on a particular question, [as I] needed to be sure I was getting it right”.
Author Phil Newton, an academic integrity expert and neuroscientist at Swansea University’s medical school, writes in the paper that it is unclear whether this sort of behaviour “would constitute cheating under conventional interpretations of assessment security and integrity”.
ADVERTISEMENT
The paper in the journal Assessment and Evaluation in Higher Education further says that there is a “profound disconnect between simple ideas of ‘cheating’ and the more complex, nuanced uses for GenAI” which “blur the line between using GenAI for academic support versus misconduct”.
AI tools can offer “enormous benefits” to students, and “therefore, from a policy perspective, it is now unclear what it means to ‘cheat’, and what is an acceptable use of these tools to support learning”. Many of the behaviours identified by students “could reasonably be interpreted as legitimate use of GenAI for academic means”, it says.
ADVERTISEMENT
Newton questions, for example, whether an assessment draft outline generated by AI “represents the learning of the student” or whether it is cheating.
He told Times Higher Education that current university policies leave students in a “no-win” situation, as “either they use GenAI and risk being accused of cheating, or they don’t use it, knowing full well that many of their colleagues will, and so they will get a worse mark”.
Proofreading using AI could either be viewed as cheating “depending on the rules”, he said, adding: “If we are testing their ability to write, then using ChatGPT as a proof-reader would be cheating. If we are testing their ability to demonstrate learning, then proof-reading might be something we want them to do, especially if they have some challenges with writing or language.”
The paper also finds that 91 per cent of students report that they are assessed through written coursework, and that 55 per cent of students had been assessed using unsupervised online examinations, with such methods “vulnerable to cheating”.
ADVERTISEMENT
It concludes: “Most students are using GenAI, and so there are serious questions about the use of these assessment methods as valid ways to certify the learning of students. There is an urgent need for the sector to develop more appropriate summative assessments in the age of GenAI, and for appropriate policies to support the use of those assessments.”
juliette.rowsell
