- This HEPI blog was kindly authored by Dr Fawad Khaleel, Associate Professor and Head of Online, Dr Patrick Harte, Senior Academic Integrity Officer and Senior Lecturer, and Dr Sarah Borthwick Saddler, Academic Integrity Officer and Lecturer, all at the Business School, Edinburgh Napier University.
Breaches in student academic integrity include, but are not exclusive to, acts such as cheating, plagiarism, and others which gain an unfair advantage over other students. Previously, the problem was persistent but manageable for the academic world. However, emerging technologies like Artificial Intelligence, Machine Learning, and sophisticated search algorithms have now revolutionised the way students gather information and complete assignments. These tools can significantly enhance learning experiences but they also make academic dishonesty more accessible and more attractive.
The threats from such emerging technologies stretch beyond solely the academic conduct issue. Most UK HEIs do not yet use nascent or ‘unproven’ plagiarism checker tools that may detect AI-generated text. However, academics are generally able to identify the use of AI-generated text due to their subject knowledge, the difference in the tone of the text and the distinctive feel of AI-generated discourse. The suspicion of misconduct without support and reinforcement from approved AI detection software often ultimately requires an oral examination or ‘viva’ to gather evidence on the originality of the work. This absorbs significant time from both academic and administrative staff. The recent exponential increase in academic integrity breaches due to the use of generative AI has resulted in rampant, unnoticed costs.
We are currently conducting a longitudinal study going back to 2019 on academic misconduct and the circumstances surrounding occurrences. Our study suggests there are direct and indirect costs associated with processing, preparing, investigating, recording, and reporting a single misconduct case that requires viva. The direct cost is based on our examination of 2075 cases of academic dishonesty since 2019 and results show it costs 56 minutes of academic time and 106 minutes of administrative time per student per case. These figures include an academic identifying the case, reporting it to the Academic Integrity Officer (AIO) (or equivalent), AIO then processing it and potentially inviting students for a viva, conducting the viva and completing post-viva reporting. Usually, a viva is 30 minutes long and requires the presence of an academic, an AIO as chair, an administrator for recording the proceedings, a Student Union representative and the student. Previously the cost of academic and administrative time, in this context, was neither calculated nor recorded within HEIs due to the infrequency of these occurrences.
However, since the public launch of generative AI, breaches have increased. For instance, Abertay University experienced a 411% increase, as breaches of academic integrity jumped from 36 cases in 2020/21 to 184 cases in 2022/23. Similarly, the breaches of academic integrity within TNE and online provision of Herriot Watt University increased from 323 cases in 2020/21 to 901 cases in 2021/22. In 2022/23 out of 952 breaches that were investigated, 825 necessitated viva examination, which required attendance of a minimum of 4 staff members (academic and admin). Similar is the situation in other UK HEIs as Glasgow Caledonian University had 422 cases in 20/21, which increased to 742 in 21/22 and 711 in 22/23, while at Edinburgh Napier University the Business School alone has had 1395 cases in the last two years. Over the same two years, the University of Stirling had 1827 cases and the University of Edinburgh has recorded 1552 cases of academic dishonesty. Most Universities do not record the specific data of academic dishonesty due to the use of generative AI nor on the number of viva examinations conducted to investigate the breaches.
This increase coincides with the growing availability of AI tools: OpenAI’s GPT-2 was launched in 2019, GPT-3 in 2020, and Jasper AI in 2021. Other available tools included QuillBot, Article Forge and Scribendi, which were specifically designed either as paraphrasing tools, essay-generating tools or both. The trends were accelerated with the launch of ChatGPT in November 2022 and Google Bard, now Google Gemini, in February 2023. Though we cannot be confident that AI is fully responsible for the increase, the sharp rise just as AI tools are made available suggest it is a significant factor.
The total number of breaches within a HEI is correlated to the size of the student population. For example, a university experiencing 1000 cases a year which must be investigated through a viva examination will cost the University 933 hrs in academic time and 1764hrs in administrative time (2697hrs in total). This translates to a total annual cost of £95,181.06, which includes £38,644.86 of academic time and £56,536.2 of administrative staff time. We calculated it using the gross hourly rate cost to HEI under Scottish PayScale (£41.42 per hour of academic time and £32.05 of admin time). If we generalise this cost over all the public universities in the UK it costs £12.4 million per year to the British economy and $196 million per year to the US economy. In the UK context it equates to 951 postgraduate scholarships per year, the cost of 33,715 ambulance trips to A&E, 294,608 GP appointments or the cost of creating 1658 more places at primary schools.
The majority of UK HEIs are investigating the use of generative AI. However, their focus is currently on the use of AI by students and academics to improve learning, teaching and assessment, or on the official policies on the ethical use of AI in pedagogy and/or research. The purpose of this report is to highlight to the policymakers that it is vital for HEIs to collect thorough data on the breaches of academic conduct due to AI and the cost related to it. It may also be useful to link assessment design with the breaches to explore the types of assessment questions, word count limits, and scope of assessments that result in high numbers of academic dishonesty cases.
I would argue that it’s not costing universities anything monetarily (as yet) unless there is evidence that extra academics or support staff are being hired or paid extra to handle these cases.
The human cost is real, though: having to work longer hours, increased workload, heightened stress, risk of burnout.
Thank you for the article!
Lots of contradictions in this piece most notably, for me, the fourth paragraph starts ‘However, since the public launch of generative AI, breaches have increased.’ and ends with ‘Most Universities do not record the specific data of academic dishonesty due to the use of generative AI nor on the number of viva examinations conducted to investigate the breaches.’ So how can the authors assume all academic misconduct is AI related? Yes, academic misconduct has increased but is it really down to AI?
And finally “Though we cannot be confident that AI is fully responsible for the increase, the sharp rise just as AI tools are made available suggest it is a significant factor.” How can the authors know it’s a ‘significant’ factor without the evidence? Increases in the availability of AI tools does not lead to causation. Perhaps increases would have happened without AI due to time poor students (due to needing to work because of cost-of-living crisis) choosing to paraphrase or ‘cut and paste’, etc. Perhaps this is the significant cause in the jump?
This research presents findings that are likely to resonate across a broader international context, suggesting that the issues addressed may be universal. Specifically, I observe the same increases in Germany, as mentioned in this blog. The aspect of associated costs that AI is causing adds an interesting dimension to the overall AI discussions around ethics and security. As the longitudinal study progresses, it will be interesting to observe whether the introduction of new technologies can mitigate these rising costs within the next one to two years, or if the quest for perfect tools remains elusive. My exposure to an early beta version of a new tool, tested by my colleagues and myself, revealed a remarkably precise outcome, further justifying the importance of this research.
Moreover, the process of data collection could benefit from a nuanced consideration of the financial implications for private educational institutions, such as private universities and business schools. In drafting this response, it occurred to me that, beyond the direct costs incurred by these institutions, there is an (not expected) potential for revenue generation through students required to revise their theses. Given the substantial market share of private universities in certain European countries, exceeding 20% in some instances, this financial perspective represents an intriguing avenue for further exploration.
So, we know the opportunity cost, how to convince the management to use some resources to save? They do not see these costs but will see spendings on any action, even planning for it.
Next issue: what then? Maybe changing assessment, which does not suit the current situation? The AI will be used in business so it must be used for the assessment as integral part of task. Other wise we will end with Wehrmacht pictures with full racial diversity created by AI:
Jo Zefka on X: “Gemini (le générateur d’images de Google) est si obsédé par la diversité que, quand on lui demande de générer des images de soldats allemands de 1943, il invente des nazis de toutes les couleurs! Nous avons créé une IA encore + bête que…
https://twitter.com/JZefka/status/1760781002085933522
This is a very interesting and relevant piece calling for a so much needed improvement in data collection regarding academic integrity and AI use. This is a clever call as it should catch the eye of university management who would be unbearable interested in the cost part of it, but at the same time the improvement on data collection and the analysis of this data would allow educators to better adapt assessment tools to assure real learning is achieved. If we are to stand any chance of getting on top of this trend it is imperative to have accurate data to show patterns and trends. I look forward to reading more about this and get the results of the full study.
In this blog post the authors shed light on an important yet not broadly discussed topic. Academic integrity and AI will change the way we create knowledge and, hence, the way we teach at HEIs. Therefore an awareness needs to be addressed globally and not limited to Scottish/UK HEIs. Also, are there other cause-effect variables, ie. revised academic integrity trainings in addition to a cost-based perspective?
Seeing this financial cost in black and white is a sobering reminder of the energy and resources expended on chasing dishonest use of generative AI. I suspect that collecting data on breaches of academic conduct using AI will remain and inexact science though, as crude uses will probably reduce and more sophisticated tools and skills will increase. Assessment design is probably key, as you recommend here, and integrating the value of learning throughout the curriculum, not just at the point of assessment would be a good start.
They need to change how they assess has been the case for years