Abstract:

In the vivid discussions on generative AI (GenAI) there are people arguing for a bursting AI bubble. Regardless of a bursting bubble or an approaching AI winter, the challenges will remain for how to handle assessment in higher education. The use of GenAI involves a range of ethical issues where this paper investigates the challenge of keeping student submissions authentic and not bot completely generated. It can be argued that this challenge is not a new one since so called 'essay mills' and ghost-writing services have been around for years. However, the rapid development of GenAI bots has aggravated the issue and put stress on teachers. At the same time a new market has appeared for AI detection tools that at a first glance might seem to be the solution. The aim of this study is to evaluate and discuss the quality of bot generated essays and the results when they are evaluated with AI detection tools.

Educational action research was used as the overall strategy where the author had the multiple roles of conducting research on a course where the earlier roles have been subject matter expert, the content developer and one of the lecturers. Data were collected from solutions to an assignment submitted by course participants in four iterations of a 7.5 credit course on Artificial Intelligence in Education. With the Grounded theory methods of open coding and axial coding found themes have been related to the axial category of 'Detection tools'. Furthermore, the solutions of the assignment have been discussed in online course webinars.

Findings show that the tested AI detection tools are not reliable, and that some of the tool providers have also developed so called 'AI Humanizers'. Anti-detection tools that have been presented with an ability to help turn GenAI writing into a smooth and natural text while staying untraceable. Moreover, the involved GenAI bots fabricated deceptively real references. The conclusion is that with or without AI Humanizers, none of the tested tools could be recommended for evaluating student assignments, and that more cases of inappropriate use of GenAI are detected by fabricated citations. Finally, a surprising finding was that the output from one of the tested AI bots involved a warning about that the provided references only are suggestions and maybe has to be replaced with more accurate ones.

Keywords:

Artificial Intelligence in Education, AIED, GenAI, Detection tools, Fabricated citations.