AI through an EDI lens: reflections from UWE’s AI in education conference

Steffie Denton is the EDI Projects Coordinator, Civic and Inclusivity, in Research and External Engagement Services (REE) at UWE. Steffie works on projects and evaluation for the Access and Participation Plan (APP), focusing on Intervention Strategy 2 – Financial Support and Intervention Strategy 4 – Deepening Practice Based Learning.

A third of new students have no AI experience

Hosting the AI and EDI panel discussion at our recent AI in Education conference back in September made one thing obvious: students use AI differently. There is no typical student user.

Some students use AI constantly. They feel confident experimenting and can successfully integrate AI tools into their everyday work and study. Some rely on AI for emotional support, to navigate assessments or to help with translation. Others, around a third of undergraduates according to HEPI’s most recent data, arrive at university with no AI experience at all. Some are cautious or unsure, choose not to use AI because of ethical concerns or personal values, or worry about “accidentally cheating” and avoid it altogether.

And exactly the same is true of Higher Education staff.

To me, this is an equity issue. It could ultimately affect KPIs like Graduate Outcomes because AI literacy and prompt engineering are already listed among the fastest-growing employability skills.

Another hidden curriculum

Unequal starting points also create a new kind of hidden curriculum. Students who already know how to write effective prompts, check outputs and use AI within assessment rules have an advantage. Those without prior knowledge may not even know what questions to ask. If we leave students to work this out for themselves, AI confidence risks becoming another axis of inequality.

For example, Black students are more likely than White students to use AI and may turn to it to fill gaps in support, including getting instant explanations, accessing help outside office hours and navigating unclear expectations. If AI policies are unclear, student groups who use AI more could be disproportionately impacted. And this could widen equity gaps further.

A neurodivergent student might use AI to explain a difficult concept, structure their ideas or organise their thinking before beginning a piece of work, helping them to overcome barriers around processing information. This is what makes AI particularly interesting from an EDI perspective. AI is not simply being used as a study tool. For some students, it is helping to remove barriers that already exist within Higher Education. AI therefore has real potential to widen inclusion and participation, but only if students have equitable opportunities to develop the skills and confidence to use it effectively.

Baked-in bias, stereotypes and dominant voices

In JISC’s recent findings, some students expressed concern about bias in AI outputs. They also raised concerns about hallucinations, where AI presents incorrect or assumed information confidently as fact, and about fair access to paid tools that may give wealthier students an advantage.

It is worth stating explicitly that AI systems can reproduce and amplify biases contained in their training data.

Tools like Copilot use prediction based on patterns in that data. This means their outputs often reflect dominant voices and mainstream perspectives. They can reproduce stereotypes, misinformation and misrepresentations of marginalised groups. Certain views or voices may only appear if we explicitly prompt AI to include them. Another equity issue.

So, what can we actually do?

In my opinion, we cannot remove every EDI issue from AI. But we can make sure students and staff are better equipped to recognise and challenge them.

Embed AI as a core activity: create low-pressure opportunities to practise in everyday teaching and work settings.
Make expectations explicit: be clear about where, when and how AI can be used, and exactly where it can’t.
Role-model critical reflection: actively ask people to question AI outputs. What sources, facts, perspectives or opinions have not been included?
Make prompts more inclusive: ask AI to flag uncertainty, bias and gaps in evidence, highlight missing or underrepresented voices and include global viewpoints.

If we want AI to support inclusion rather than widen gaps, EDI cannot be an afterthought. It needs to sit alongside how we talk about AI, how we teach it and how we use it ourselves.