Three wooden dice on a pale background. One shows a gavel, one a set of scales, and the other says 'AI'.

I spoke to Jason Billenness-Clarke, Senior Lecturer in Criminology, about how he has used GenAI to make an assessment based around critical debate workable in a large cohort, and how it has developed their thoughtful and critical engagement with GenAI.

Foundations

Much of the discourse around Generative AI (GenAI) and assessment seems to have been about whether it’s possible to prevent and detect its use. This has been complicated by the speed of GenAI’s evolution, and the fact that studies have shown GenAI detectors to be inconsistent, with a disproportionate negative impact on already minoritised groups (e.g. Gotoman et al., 2025).

Some people have chosen, instead, to make the use of GenAI central to students’ experience of assessment.

“I started by asking myself what I actually wanted students to get from the assessment,” Jason told me. “My teaching philosophy isn’t about memorising and recalling content. I want students to connect deeply with the subject. My focus is prisons and probation, which can feel like a hidden world unless you’ve experienced it directly.”

With a very large module – over 300 students from Criminology, Psychology and Law – and only four tutors, Jason’s interest in using assessed debate to promote deep thought and reflective engagement was a challenging prospect. And then he had a ‘lightbulb moment’ – what if the students could debate with AI?

Jason researched whether there was a model of this working, but while he found similar ideas there was nothing quite like it that he could find, either in the literature or at Nottingham Trent University (NTU). “That was exciting, but also a little unnerving,” he said.

He worked carefully with colleagues on the prompt design for the assessment, using our institutional tool Microsoft Copilot. One of the key aspects he wanted was for the GenAI tool to act as a ‘devil’s advocate’, providing reasonable arguments that students could respond to and defend against.  A transcript of the debate would then be submitted for marking by the tutors on the course, ensuring that human judgement was central to the assessment process.

Careful design

I asked about the biases that GenAI could bring to the debate – given the ‘hidden world’ of prisons and probations, and the limited information from within it in the AI’s dataset, might there not be controversial external opinions that would need to be mitigated?

Jason had thought of this. “Actually, that’s part of what students are assessed on.

We discuss these controversies and biases throughout lectures, workshops, and in-person debates. If AI produces a biased argument, that’s an opportunity. Students can recognise it, challenge it, and argue against it using evidence.”

“They’re graded on critical thinking, not on “beating” the AI.”

Jason designed a prompt for students to use that would cause Copilot to challenge the student’s position regardless of what they argued, with the assessment based on the student’s responses; how they evaluated evidence, identified weaknesses in the GenAI tool’s arguments, and constructed stronger arguments of their own.

“Those are exactly the tools we want them to develop.”

Building to a strong assessment

How did Jason prepare his students to participate in a critical debate?

Crucially, by making the debate environment available from early in the module. Students were able to repeat the process multiple times, building their skills and rehearsing their arguments while never encountering the same exchange twice.

“The idea was to create a process of continuing learning and development.”

The debate preparation was also integrated into the curriculum, with formative activities and practical sessions offering opportunities for students to discuss their experiences and feedback from the AI tool with tutors, making sense of what they could do to improve. 

And how did students react to the assessment?

“The response,” Jason told me, “was quite divided.”

Many students enjoyed engaging with the assessment task, and reported that they felt more connected with the subject and more engaged with existing debates within the field.

“Interestingly, applications for our prison officer placement year increased significantly afterwards.”

Some students were less positive about the assessment, and Jason feels that this may have been impacted by the fact that it was an innovative assessment format that students didn’t have a familiar template to navigate.

However, even the students who felt less positive towards the assessment format did acknowledge that they learned a lot from it.

“Students learned to question AI outputs, verify information, and think critically. That’s an incredibly important skill. Many told me they no longer copy and paste AI outputs blindly. They now check everything first,” Jason said.

“For me, that’s one of the most valuable outcomes of the entire project.”

References

Gotoman, J., E.J., Luna, H.L.T., Sangria, J.C.S., Santiago Jr, C.S. and Barbuco, D.D., 2025. Accuracy and reliability of AI-generated text detection tools: A literature review. American Journal of IR 4.0 and beyond, 4(1), pp.1-9.

Author: Bethany Witham

Learning Technologist

SOC Digital Learning Team
School of Social Sciences
Nottingham Trent University