Why AI literacy must be discipline specific

A student’s attitude to artificial intelligence (AI) depends heavily on what subject they study, recent research has shown.

An engineering student described AI as “augmentation, not replacement”: using it for repetitive tasks while retaining responsibility for core logic and final validation. A graphic design student drew a very different line: AI could serve as a tutor or supporter, but never as author. A nursing student’s view was shaped less by personal preference than by the weight of professional accountability that the discipline carries by design. These distinctions are grounded in the structure of each discipline, and in what AI therefore threatens or supports. 

How AI literacy differs across disciplines

In STEM subjects, AI literacy is bound up with the logic of the scientific method.  Reproducibility, audit trails, the conditions under which a result can be trusted: these are not abstract values but working practices. A biology or chemistry student learning to use AI tools needs, above all, to understand where a model’s output comes from and under what conditions it might fail. In research contexts, hallucination is a methodological problem that corrupts the research process, not a minor inconvenience to be worked around. AI literacy in STEM means learning to interrogate a tool the way a scientist interrogates data.

Humanities disciplines carry a different burden. The authorial voice, the interpretive act, the reading of sources for argument rather than information retrieval: these are what the disciplines are for, not merely assessment conventions. When an AI tool summarises a primary source, it produces a statistically probable description rather than a reading. A history student who reaches for AI as a substitute for that act of interpretation is bypassing the very act the subject exists to teach. AI literacy in the humanities is most usefully framed as provocation rather than prohibition: what does this tool produce, what has it left out, and why does that matter to your argument?

Professional fields present a third and arguably more pressing challenge. Medicine, law, social work, nursing and teacher education all involve regulated practice. The professional standards that govern these fields pre-exist AI and are not optional. A law student who submits a brief citing cases that do not exist has produced work that could harm a client. A nursing student who relies on an AI output without applying clinical judgement may one day harm a patient. The stakes are not hypothetical. AI literacy for professional programmes must be built around accountability: who is responsible for this output, under what regulatory framework, and how do you demonstrate that you exercised appropriate judgement?

What field-specific AI training looks like in practice

The variation between these fields is wide enough that a single institutional framework cannot meaningfully address all three. What works is guidance built at module level, specific to the discipline, the assessment type and the stage of the work.

Some of the most effective approaches currently emerging share a common feature: they ask students to account for their AI use rather than simply declare it. In engineering programmes, this can mean students walking through the logic of a piece of code in a short oral, explaining which design decisions were theirs and where AI contributed. 

In philosophy seminars, students compare an AI-generated argument with their own position and articulate precisely where and why they diverge. In clinical education, oral defence formats require students to justify a clinical decision regardless of what tool assisted their initial workup. 

A postgraduate design course currently requires students to submit a research folder of 100 to 200 pages with reflective analysis on every page, a format that makes AI-assisted shortcuts visible precisely because the assessment traces the student’s thinking throughout the project rather than only at the point of submission.

These approaches share a second feature: the presence of a visible accountability moment changes how students use AI throughout the module. Students who know they will need to explain their work in person use AI to test and deepen their understanding. Students who face no such moment have no equivalent incentive.

• Spotlight guide: Bringing GenAI into the university classroom
• Spotlight guide: AI and assessment in higher education
• AI and assessment redesign: a four-step process

What institutions could usefully do

Drawing on both the disciplinary analysis and the emerging evidence on student experience, five practical steps are worth prioritising.

Design AI guidance at module level, not institution level: the variation across creative arts, STEM, healthcare and humanities is too wide for any uniform approach to be genuinely operable. Module teams are best placed to define what AI use looks like in the context of their specific learning outcomes, assessment types and disciplinary norms.

Assess explanation of AI use, not declaration: a brief oral component, a process reflection, or a structured conversation about submitted work creates a far stronger incentive for genuine engagement than any checkbox.

Align AI guidance with disciplinary purpose: the question for each programme team is not “how much AI is acceptable?” but “what does AI do to the specific kinds of knowledge and judgement this programme develops?” Starting from that question produces guidance students can actually use.

Coordinate expectations across teaching teams: a 10-minute conversation at the start of the academic year, producing a shared position on AI use across a programme, would resolve the single most commonly reported source of student confusion.

Address unequal access to AI tools: where AI use is expected or encouraged, the quality differential between paid and free tools is a socio-economic issue. Students who cannot afford paid tools are working with worse outputs and spending more time correcting errors. Institutional licensing is a practical and direct way to close that gap.

There is a sixth move worth making alongside these five. Students can describe with precision when AI supports genuine learning and when it replaces it. That clarity is currently invisible and largely unused. Bringing it into curriculum design, through structured seminar discussions, peer case analysis or disciplinary ethics workshops, produces better guidance than any top-down framework and treats students as partners in working out what learning means in their field rather than subjects to be governed.

The discipline problem with AI literacy does not require a complicated solution. It means treating each field on its own terms and building guidance that starts from what the subject is actually for.

Rose Luckin is a professor emeritus of Learner Centred Design at UCL. 

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.