Whenever a new technology enters education, the first question is often predictable: where can we use it? With artificial intelligence, that question now appears in almost every conversation about teaching, learning and assessment.

I increasingly think it is the wrong starting point. The curriculum should not begin with AI. It should begin with the kind of graduate we are trying to develop, the competence they must demonstrate and the decisions for which they must remain accountable.

The tool-first trap

When we begin with the tool, curriculum design can become a search for places to insert it. A lecture gains an AI activity, an assignment gains a prompt-writing component, or a clinical case gains an AI-generated answer for students to critique.

Some of these activities may be useful. But unless they connect directly to an intended learning outcome, they can create activity without creating alignment. The better sequence is to define the competence, identify the thinking that must remain human, decide the permitted AI role and specify what evidence will make student reasoning visible.

AI literacy is not chatbot familiarity

Students may already use ChatGPT or similar tools, but exposure is not the same as literacy. AI literacy includes knowing when a tool is appropriate, checking what it produces, recognising bias and privacy risk, explaining how it contributed and accepting responsibility for the final decision.

This is especially important in physiotherapy. A plausible answer can still miss a red flag, ignore a contraindication, misrepresent evidence or fail to account for the person in front of us.

Three kinds of assessment

An AI-ready programme may need three deliberately different conditions: AI-free assessment to confirm independent foundational competence; AI-supported assessment to represent authentic practice while exposing the student's reasoning; and AI-critical assessment to test whether students can find omissions, challenge unsafe suggestions and verify evidence.

These conditions should be designed across the programme rather than improvised assignment by assignment. Otherwise, students may encounter duplicated activities, unaddressed gaps and contradictory expectations.

A curriculum pathway, not a single AI lecture

Early in a programme, students might learn about limitations, hallucinations, source checking, disclosure and privacy. Later, they could use AI in evidence appraisal, patient communication and data interpretation. In the clinical years, they could evaluate AI-supported reasoning and identify missing safety information. Near graduation, the focus could move to governance, equity and professional accountability.

This progression matters because responsible use depends on disciplinary knowledge. Students cannot reliably identify an AI error before they know enough physiotherapy to recognise it.

The educator becomes more important

AI can generate case variations, suggest rubric language and create examples for discussion. It cannot independently decide what is educationally valuable, clinically safe or appropriate for a particular cohort.

The educator's role is therefore not disappearing. It is becoming more demanding: curriculum architect, designer of meaningful practice, facilitator of judgement and verifier of quality.

The evidence still calls for restraint

There are encouraging signals, but enthusiasm is ahead of evidence. A 2026 systematic review of randomised studies in undergraduate health-professions education found possible benefits for satisfaction, confidence and theoretical knowledge, but results were inconsistent and evidence certainty was low. No included studies assessed workplace behaviour or health outcomes.

That does not mean waiting until every question is settled. It means treating each new AI activity as a curriculum intervention that requires a rationale, alignment, evaluation and willingness to change course.

A framework I want to explore

The sequence I am considering is: CILO → Human competence required → Permitted AI role → Learning activity → Evidence of student reasoning → Assessment and assurance.

This places AI in the curriculum without placing it at the centre. The centre remains the graduate: what they can understand, do, explain and take responsibility for. Perhaps the most useful question is no longer, ‘Where can we use AI?’ It is, ‘What must remain visibly human, and how can AI support that without quietly replacing it?’

Sources informing this article