Learning AI primarily through use
This framework enables students to develop AI literacy while learning disciplinary content. Rather than teaching AI as a separate topic, embed AI engagement into existing learning activities so students develop capability through authentic practice. Some foundational understanding of what AI systems are and how they work will make that practice more effective — but this orientation is a prerequisite to learning through use, not a substitute for it.
Purpose
This framework guides educators in embedding AI literacy development into existing modules and courses. The primary mode is learning through use: students develop capability by using AI to engage with disciplinary content, not by studying AI in isolation. The module content remains primary; AI literacy develops as a by-product of structured engagement.
This does not mean AI can be learned without any knowledge about it. Understanding what AI systems are, how they generate responses, and where they fail will make practice more effective. That orientation belongs at the start, before embedded activities begin — as a prerequisite rather than as a substitute for learning through use.
The six dimensions of AI literacy in this framework are not derived from AI literacy literature alone. They appear consistently across information literacy, media literacy, digital literacy, and data literacy traditions — suggesting they reflect the structure of literacy itself, not any single domain. The framework applies this common architecture of literacy to the specific context of AI engagement.
The six dimensions of AI literacy
AI literacy is multidimensional capability that cannot be reduced to any single skill. When embedding AI into learning activities, consider which dimensions each activity develops:
1. Access and recognition
Understanding what AI systems are and recognising when they are relevant. Students learn to distinguish AI from search engines or databases, understand AI as language-based cognitive extension, and identify when AI might help versus hinder their learning.
In practice: Activities that ask students to consider whether to use AI for a task, not just how.
2. Critical evaluation
The capacity to assess AI outputs for quality, accuracy, and reliability. Students learn that AI can hallucinate, recognise the difference between fluent and accurate responses, and maintain appropriate scepticism while remaining open to genuine assistance.
In practice: Activities requiring verification of AI outputs against authoritative sources or disciplinary standards.
3. Functional application
The practical ability to use AI systems effectively for specific purposes. Students develop competence with prompting, learn to structure requests for better outputs, and deploy AI capabilities appropriately.
In practice: Structured prompting exercises with clear expectations for output quality.
4. Creation and communication
The skill to generate meaningful outputs through collaboration with AI. Students move beyond consumption to production—using AI to develop arguments, decompose problems, and communicate ideas while maintaining their distinctive voice.
In practice: Activities where AI supports the creative process but students remain responsible for the intellectual contribution.
5. Ethical awareness and responsibility
Understanding the social, ethical, and professional implications of AI engagement. Students learn about transparency, attribution, academic integrity, and when AI use might undermine rather than support their learning goals.
In practice: Explicit discussion of appropriate AI use for each activity, with clear expectations.
6. Contextual judgement and metacognition
Developing taste and professional judgement about meaningful AI engagement. Students learn to distinguish genuine competence from mere familiarity, recognise when AI helps versus hinders, and develop domain-specific judgement that evolves with experience.
In practice: Reflection activities that ask students to evaluate the quality of their AI engagement, not just their outputs.
Developmental progression
AI literacy develops through three stages. Design activities that meet students where they are and scaffold progression. The conceptual basis for this trajectory — why it unfolds in this order and what each stage requires — is explored in developing AI literacy.
Before the stages begin — orientation prerequisite: Students need a foundation before embedded activities can work. This means understanding that AI generates responses based on patterns rather than retrieving information, that conversational engagement produces better outputs than transactional queries, and that AI and humans make different kinds of mistakes — which is why critical evaluation remains the student’s responsibility, not something AI can do for them. This orientation can be delivered as a short session at the start of a module or embedded in an existing introductory lesson; it does not need to be a standalone course.
Stage 1: Substitution
Focus: Using AI to complete existing tasks more efficiently.
Key learning:
- Structured prompting (Role, Goal, Instruct, Discuss) improves outputs
- Not every task benefits from AI—learn to select appropriately
- Track time honestly, including all revision
- Efficiency serves learning goals, not productivity for its own sake
Activity characteristics:
- Bounded tasks with clear structure and expectations
- Time tracking to evaluate actual efficiency gains
- Honest evaluation: did AI actually help with this task?
Stage 2: Adaptation
Focus: Reshaping learning approaches around AI capabilities.
Key learning:
- Build genuine competence, not just familiarity
- Test understanding immediately rather than assuming comprehension
- Use staged complexity: basic idea → contextual understanding → procedural detail
- Develop arguments through structured dialogue
Activity characteristics:
- Extended engagement building understanding through conversation
- Multiple rounds of testing and refinement
- Activities that require application, not just comprehension
Stage 3: Transformation
Focus: Integrating AI into sustained practice with professional judgement.
Key learning:
- Context management enables more sophisticated engagement
- Taste and judgement develop through reflective practice
- Structural integration sustains practice better than willpower
- Ongoing maintenance and adaptation required
Activity characteristics:
- Ongoing projects with accumulated context
- Metacognitive reflection on AI engagement patterns
- Student-directed decisions about when and how to use AI
Designing embedded activities
Principles for embedding AI literacy
1. Start with the learning outcome, not the technology Design activities that achieve disciplinary learning goals. Then consider how AI engagement might support that learning while also developing AI literacy.
2. Make AI use authentic, not artificial Students develop literacy through genuine use. Avoid contrived “use AI for this” exercises. Instead, create situations where AI engagement is a natural part of the learning process.
3. Require critical evaluation, not acceptance Every AI-assisted activity should include evaluation of AI outputs against disciplinary standards. Students learn that AI assistance requires verification.
4. Build in reflection Include structured reflection on the AI engagement process, not just the output. What worked? What didn’t? When did AI help versus hinder?
5. Progress through stages deliberately Early activities provide more structure and guidance. Later activities give students more autonomy to make judgements about AI use.
Activity design template
For each embedded activity, consider:
| Element | Questions to address |
|---|---|
| Learning outcome | What disciplinary knowledge or skill does this develop? |
| AI literacy dimensions | Which dimensions does this activity develop? |
| Developmental stage | Substitution, Adaptation, or Transformation? |
| Structure provided | How much prompting guidance do students receive? |
| Evaluation requirements | How must students verify or evaluate AI outputs? |
| Reflection component | What metacognitive questions will students address? |
| Success criteria | What distinguishes effective from ineffective AI engagement? |
Example activity types by stage
Substitution activities (general):
- Use AI to generate first-draft summaries of course readings, then evaluate and refine
- Create study materials (flashcards, practice questions) with AI assistance
- Draft routine professional communications with AI support
Substitution activities (health professions education):
- Use AI to draft patient education materials, then verify clinical accuracy against current guidelines
- Generate OSCE station scenarios or assessment questions, then evaluate for clinical validity and appropriate level of challenge
- Draft clinical placement feedback templates, then refine for professional tone and specificity
Adaptation activities (general):
- Build understanding of complex concepts through staged AI dialogue
- Develop arguments by testing reasoning with AI as interlocutor
- Decompose complex problems with AI support before solving
Adaptation activities (health professions education):
- Build understanding of clinical reasoning frameworks (e.g., hypothetico-deductive reasoning, pattern recognition) through staged dialogue that tests comprehension at each step
- Test clinical arguments against AI-generated counterarguments, then evaluate which challenges reveal genuine reasoning gaps versus introduce unnecessary complexity
- Decompose complex cases or clinical problems into components before attempting synthesis
Transformation activities (general):
- Maintain ongoing project context across multiple AI sessions
- Design personal AI engagement strategies for different learning tasks
- Evaluate when AI engagement serves learning goals versus undermines them
Transformation activities (health professions education):
- Integrate AI into curriculum design workflows with documented context that builds across sessions (e.g., module learning outcomes, student cohort characteristics, assessment constraints)
- Develop domain-specific judgement about when AI engagement would be appropriate in clinical or supervision contexts versus when it would undermine professional development
- Evaluate AI engagement against professional competency frameworks (e.g., CanMEDS, HCPC standards) rather than generic quality criteria
Assessing AI literacy development
What to assess
Assessment should focus on the quality of AI engagement, not just the quality of outputs. The distinction matters: a student can produce a polished output through poor AI engagement (uncritical acceptance, no verification, generic prompting) and a less polished output through excellent engagement (structured prompting, rigorous evaluation, honest reflection on limitations). Both process and product deserve attention.
Evidence of development by dimension
| Dimension | Evidence of development |
|---|---|
| Access and recognition | Can explain why AI was or wasn’t used for a specific task; selects appropriate tools and timing without prompting; identifies when AI would be unhelpful or inappropriate |
| Critical evaluation | Identifies specific errors or limitations in AI outputs; documents what was changed and why; verifies claims against authoritative sources or professional standards |
| Functional application | Demonstrates iterative prompt refinement rather than single-shot queries; outputs meet disciplinary or professional quality standards; uses contextual prompting that provides relevant background |
| Creation and communication | Can articulate their specific intellectual contribution distinct from AI’s; maintains consistent professional voice; AI role is transparent and bounded |
| Ethical awareness | Identifies relevant ethical considerations before beginning a task; applies appropriate transparency conventions; can explain when AI use would be professionally inappropriate |
| Contextual judgement | Makes appropriate decisions about AI use without being directed; reflects on the quality of AI engagement not just outputs; judgement demonstrates improvement over time |
In health professions education contexts, also consider: does the student recognise when AI limitations could have clinical consequences? Do they apply different standards to clinical content than administrative tasks? Are they developing professional habits they would be comfortable defending to a regulatory body?
Assessment approaches
Process logs: Students submit the actual AI interaction alongside their output — prompts, responses, and the revisions they made. This provides direct evidence for dimensions 1–4 and reveals whether engagement was iterative or single-shot.
Structured reflection: After each AI-assisted task, students respond to a short set of prompts: What did you use AI for and why? What did you change and why? When did AI help or hinder? What would you do differently? Open-ended reflection portfolios produce variable evidence; structured prompts produce comparable evidence across students.
Comparative tasks: Students complete a closely related task with and without AI assistance, then reflect on what differed. This surfaces genuine judgement about when AI adds value and when it doesn’t.
Calibrated self-assessment: Students rate their own engagement quality against the dimension criteria above, with instructor feedback on calibration. The goal is accurate self-assessment, not high scores — students who consistently overrate their engagement quality are demonstrating poor metacognition regardless of output quality.
Implementation guidance
Module-level integration
- Audit existing activities: Identify where AI engagement could support disciplinary learning while developing AI literacy
- Sequence deliberately: Ensure activities progress through developmental stages appropriately
- Be explicit about expectations: Clarify AI use policies for each activity rather than blanket policies
- Build in reflection points: Regular opportunities for students to evaluate their AI engagement patterns
In health professions education
Several features of HPE make AI literacy both more important and more complex than in most other disciplines.
Patient safety raises the stakes for critical evaluation. An uncritical approach to AI outputs that would be a minor learning issue in other contexts can generate clinical misinformation that harms patients. This makes Dimension 2 (critical evaluation) a professional responsibility, not just a good habit. Design activities that reinforce verification as non-negotiable — particularly for any AI-generated clinical content.
The efficiency framing of Substitution requires care. In HPE, some friction in skill development is intentional. Students learning clinical reasoning, procedural skills, or communication competencies sometimes need the effortful process, not the shortcut. The Substitution stage should prompt students to ask not just “does AI make this faster?” but “is this a task where doing it myself develops competence I need?” This is itself an AI literacy skill — recognising when substitution undermines development rather than supporting it.
Professional frameworks provide ready-made evaluative criteria. Competency frameworks familiar to this audience — CanMEDS, NHS frameworks, HCPC standards for allied health professionals — can serve as disciplinary standards against which students evaluate AI outputs. Rather than generic quality criteria, students can ask: does this AI-generated case adequately address the communicator role? Does this feedback template reflect professional responsibility standards? Using existing frameworks grounds AI literacy development in recognisable professional expectations.
Supervision and placement contexts introduce additional considerations. AI use in practice settings raises questions that don’t arise in classroom contexts — confidentiality of patient information, appropriate disclosure to supervisors, and whether AI engagement aligns with the practice culture students are entering. The Transformation stage is where students develop the professional judgement to navigate these contextual differences.
Avoiding common pitfalls
Don’t: Create artificial “use AI for this” exercises disconnected from learning goals Do: Embed AI engagement into authentic learning tasks
Don’t: Allow unstructured AI use without guidance or reflection Do: Provide scaffolding appropriate to developmental stage
Don’t: Focus only on outputs without attention to process Do: Assess quality of AI engagement alongside quality of outputs
Don’t: Assume students will develop literacy through exposure alone Do: Design deliberate progression with explicit attention to all six dimensions
Don’t: Apply blanket AI policies across all activities Do: Set context-appropriate expectations for each learning task
Quick reference: The framework at a glance
┌─────────────────────────────────────────────────────────────────┐
│ SIX DIMENSIONS OF AI LITERACY │
│ (common across all literacy traditions) │
├─────────────────────────────────────────────────────────────────┤
│ 1. Access & Recognition │ 4. Creation & Communication │
│ 2. Critical Evaluation │ 5. Ethical Awareness │
│ 3. Functional Application │ 6. Contextual Judgement │
└─────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────┐
│ PREREQUISITE ORIENTATION │
├─────────────────────────────────────────────────────────────────┤
│ What AI is · How it generates responses · Where it fails │
│ Basic prompting · Complementary errors · Critical evaluation │
└─────────────────────────────────────────────────────────────────┘
↓
┌─────────────────────────────────────────────────────────────────┐
│ DEVELOPMENTAL PROGRESSION │
├─────────────────────────────────────────────────────────────────┤
│ Substitution → Adaptation → Transformation │
│ (Efficiency) (Reshaping) (Integration) │
└─────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────┐
│ DESIGN PRINCIPLES │
├─────────────────────────────────────────────────────────────────┤
│ • Start with learning outcomes, not technology │
│ • Make AI use authentic, not artificial │
│ • Require critical evaluation, not acceptance │
│ • Build in reflection on process │
│ • Progress through stages deliberately │
└─────────────────────────────────────────────────────────────────┘
Sources
The six dimensions are grounded in the common architecture of literacy — a cross-framework analysis showing the same six dimensions appearing across information, media, digital, and data literacy traditions. They are not derived from AI literacy literature alone. Additional sources supporting this framework:
- Long, D., & Magerko, B. (2020). What is AI Literacy? Competencies and Design Considerations. Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems.
- UNESCO. (2024). Guidance for generative AI in education and research.
- Allen, L. K., & Kendeou, P. (2023). ED-AI Lit: An Interdisciplinary Framework for AI Literacy in Education. Policy Insights from the Behavioral and Brain Sciences.
- Mollick, E., & Mollick, L. (2023). Assigning AI: Seven approaches for students, with prompts. SSRN Electronic Journal.
- Association of College & Research Libraries. (2016). Framework for information literacy for higher education. ACRL.