Every student needs a minimum AI requirement
This blog was kindly authored by Roney Lima do Nascimento, an IB Diploma Programme mathematics teacher.
Under my last HEPI blog, a reader asked the right question: where can universities find the time to teach students to evaluate artificial intelligence as well as use it? If AI literacy becomes another optional workshop then the students already best served will attend, and the students most at risk of being left behind will miss it. The sector should stop treating AI capability as enrichment and start treating it as an equality-of-opportunity issue.
Universal use can still produce unequal benefit
HEPI and Kortext’s Student Generative AI Survey 2026 found that 95 per cent of undergraduates use AI in at least one way and 94 per cent use it to help with assessed work. Yet only 48 per cent feel their teaching staff are helping them develop the AI skills needed for their future careers. The survey also found statistically significant social-grade differences in the use of text-generation, coding and data-analysis tools.
Near-universal use has not abolished the digital divide. It has moved it. The old question was whether a student could get online. The new question is whether they can expose a false citation, recognise when a model is merely agreeing with them, distinguish evidence from fluent invention and know when not to use the tool at all. A free account does not supply those capabilities, and a university licence does not guarantee them.
A minimum AI learning entitlement
Every student should receive a minimum AI learning entitlement by the end of their first term. It should not be a generic certificate in prompting. It should guarantee four things:
- equal access to an institutionally supported tool, or a genuinely equivalent no-cost route where the tool is required or advantageous;
- discipline-specific teaching on verification, disclosure, data protection and the limits of the systems used in that field;
- low-stakes practice before students are assessed either with AI or under AI-free conditions; and
- clear, assessment-specific rules, alongside human review when a student is suspected of misuse.
The test of success should not be whether a student can produce a polished output. It should be whether they can explain why they accepted, rejected or modified what the system produced. That is the point at which access becomes capability.
Put AI capability where equality commitments already live
In England, access and participation plans already require providers to identify risks to equality of opportunity, set out interventions and evaluate their impact. The Office for Students’ Equality of Opportunity Risk Register identifies unequal knowledge and skills and insufficient academic support as sector-wide risks. Uneven AI capability now sits across both.
Providers should therefore examine whether students’ ability to use and question AI varies by social background, entry route, disability, age, subject or mode of study. Where the evidence shows a risk, the intervention should appear in the access and participation plan: supported access, discipline-level teaching, staff development and a way to measure whether students can verify rather than merely generate.
The Office for Students should also name AI capability explicitly in the guidance attached to Risk 1 on knowledge and skills and Risk 6 on insufficient academic support. This would not create a new regulatory bureaucracy. It would make visible a new mechanism through which existing inequalities can be reproduced.
The divide universities can still close
The next AI divide will not be between students who use a chatbot and students who do not. It will be between students who can ask a system for an answer and students who can decide whether that answer deserves to be trusted.
Universities can close that divide. But they will do so only when AI literacy stops being an optional innovation project and becomes part of the promise that students from different backgrounds should have an equal opportunity to succeed and progress.
Comments
Andy Buckley says:
My bias is that the best way to develop that second-sense that an AI answer is dubious and needs a cross-check is to learn the material first without AI. But maybe this just shows that academics are not the people with the best experience in the newest tech, as well as having the same low level of institutional access to models as the students do!
Mid-hype-cycle seems a bit early to make official pledges, though: my understanding from reporting on commercial uptake of AI is that except in specific domains (notably software engineering) the value has not been demonstrated, and the costs are going to have to go up: who know what this will look like in 5-10 years. If successful, we won’t be giving it a special name anymore, it’ll be background tech like word processors and the Web.
Reply
Replies
Sascha says:
This is definitely an issue, but in the context of math so long that you have a basic understanding of prerequisite material you can see when a step seems off. As well, you can ask AI very specific questions and get very specific answers, making it easier to learn material. This gives students with AI proficiency a leg up over those who dont. This is a problem.
Reply
Replies
Paul Vincent Smith says:
Two things:
“so long that you have…” This is I think is Andy’s point here. If you don’t, you won’t gain an understanding of how AI is doing it. And if AI is involved in your learning, you don’t have a sense of that thing outside of AI involvement.
Second: I’ve yet to see an account of “AI proficiency” in any field, as compared to less proficient use of AI. This may come with time as more studies are generated, but right now I don’t see much to disabuse me of the position that sheer practice with AI prompting (etc.) and universities’ general lack of interest in ensuring assessment validity, contribute to a general AI proficiency in students other than those who do not use it as a matter of principle.
Reply
Jonathan Alltimes says:
The background to the argument is the continued bureaucratisation by the OfS regulator of the academic guilds and the industrial training organizations through the proxies of the legal persons of the providers. The argument proposes extending the rules about equality of opportunity to the capability of students to use a generic technology called AI. Is the proposed programme of work for student capability in AI likely to cause equality of opportunity for students? Equality of opportunity could be defined as equal provision of entitlement, but the provision of an entitlement can not guarantee student capability, so we can imagine the proposed programme could cause student capability and yet we do not know from experience if such a programme is likely to cause student capability in AI for the entire class of students. The programme of work needs to be matched to the students and the specific characteristics of the technology. The argument is definitely unproven and unlikely given the mismatches. What is provided for students should be decided by the provider dependent on their capabilities and the characteristics of specific students. Defining rules, their application, and their standard of conformity is another problem.
Reply
John Elijah says:
The low-stakes practice in the third part of the entitlement could also give universities a way to test whether the teaching helps. Give students a plausible but flawed answer in their own subject, ask them to locate the supporting source and explain what they would keep or reject, then revisit the skill with a different example later. Record justified uncertainty as well as mistakes; confident acceptance is not the same as capability. That would give the entitlement an observable learning outcome beyond tool access. My essay The Truth Will Set You Free considers this distinction, for readers who want to explore it further: https://johnelijah1.substack.com/p/the-truth-will-set-you-free
Reply
Add comment