Most AI literacy frameworks I’ve come across share the same problem. They target one learner group, usually university students, and assume everyone else will figure it out. But a ten-year-old learning about AI for the first time needs something fundamentally different from a mid-career nurse working alongside AI diagnostic tools. And both need something different from a graduate student training machine learning models. The conversation about AI literacy has been growing for years now. The specificity hasn’t kept pace.
Chee, Ahn, and Lee (2025) try to close that gap. Their paper, published in the British Journal of Educational Technology, develops an AI literacy competency framework from a systematic review of 29 studies. They propose 8 core competencies with 18 sub-competencies, organized along a developmental pathway from K-12 through higher education and into the workforce. It’s a structured attempt to answer a question the field has been circling for a while: what should learners actually know about AI, and when should they know it?
The model fills a real gap. It also has blind spots worth talking about.
Beyond Tool Skills
Chee et al. identify eight competency areas: AI device and software operations, data and algorithm literacy, problem solving with AI, communication and collaboration, AI ethics, career-related competencies, AI content creation, and affective competencies.
Two elements jumped out at me. First, the affective domain. Confidence, self-efficacy, and a reflective disposition toward AI rarely show up in literacy models. Most frameworks zero in on knowledge and skills, what learners should know and be able to do. Chee et al. argue that how learners feel about engaging with AI matters just as much. I think the research supports them.
Bilbao-Eraña and Arroyo-Sagasta (2025) found that an 8-hour AI literacy intervention improved pre-service teachers’ awareness and attitudes, but trust didn’t move. You can teach someone what AI is, they can feel positive about its potential, and they’ll still hesitate to actually use it if they don’t trust it. Affective readiness isn’t a bonus feature. It’s a precondition for everything else in the framework.
Second, career-related competencies. Chee et al. define these as the ability to “recognize the potential for incorrect outcomes, verify AI-generated results for accuracy, and make rational decisions based on validated data” (p. 2155). That’s professional judgment. A nurse who catches an AI recommendation that doesn’t match clinical evidence. A journalist who spots factual errors buried inside confident-sounding AI text. Most frameworks stop at “understand how AI works” and “use it responsibly.” Chee et al. ask a harder question: can you use this tool well enough to make reliable decisions in your specific field?
From Childhood to Career
Chee et al. organize their framework along a developmental arc that makes intuitive sense. K-12 education focuses on foundational AI knowledge, basic tool operations, and early exposure to ethics and social implications. Higher education shifts toward data literacy, algorithm literacy, AI-integrated problem solving, and career-aligned skills. The workforce stage becomes field-specific: interpreting outputs in professional contexts, verifying results under real constraints, making informed decisions when the stakes are high.
I covered UNESCO’s AI Competency Framework for Students on this blog, and the two models work well together. UNESCO builds a rich, values-driven framework grounded in human agency, ethics, and civic responsibility. Chee et al. extend the timeline past graduation and into the working world, where AI literacy takes on a different shape entirely. For anyone designing workforce training or professional development programs, that extension fills a space UNESCO wasn’t trying to address.
Chee et al. also frame AI literacy within a historical sequence. They argue it’s the next step in an evolution that started with computer literacy and passed through digital literacy. Computer literacy was about operating machines. Digital literacy added networked information and online participation. AI literacy now requires understanding how algorithms shape decisions, where automated systems fall short, and where human judgment has to remain in the loop.
In their conclusion, the authors write: “AI education should progress from teaching how to use AI to teaching how to integrate it critically, strategically, responsibly, and ethically into our lives and professions” (p. 2164).
That trajectory sounds right to me. It also exposes an uncomfortable reality: many institutions are still parked at step one, teaching tool use without building the critical, ethical, and professional layers that give AI literacy real substance.
Where the Model Runs Thin
My main concern with the framework is its foundation. The entire model is built from published literature, 29 peer-reviewed studies analyzed through PRISMA methodology. Methodologically rigorous, yes. But literature-based models inherit the blind spots of the field they draw from. And AI literacy research concentrates heavily on higher education students in well-resourced countries. The framework will naturally reflect those populations and their priorities, leaving other groups underserved.
Chee et al. acknowledge several limitations. They note that the “exclusion of non-English studies likely resulted in the omission of diverse perspectives and approaches from researchers worldwide, narrowing our understanding of global trends in AI-based education” (p. 2163).
They also deliberately excluded studies on AI developers and programmers, focusing on AI literacy as a fundamental skill akin to reading and writing. That’s a defensible methodological choice, but it narrows the scope. And the sample of 29 studies is relatively small for a field this broad.
I’d press the point further than the authors do. What does AI literacy look like for a teacher in a rural school with unreliable internet? For a healthcare worker in a country deploying AI diagnostic tools without strong regulatory frameworks? For a creative professional whose entire industry is being reshaped by generative AI in real time? The contexts where this framework gets tested hardest are exactly the ones least represented in the research behind it.
Algorithm literacy raises a practical tension too. Chee et al. write that:
Understanding AI models and their outputs is vital for model interpretability, enabling users to trust AI recommendations, explain decisions, and identify biases. For example, generative AI, optimized for pattern recognition and natural language processing, has limitations in traditional quantitative analysis. Choosing the right AI model for a specific task is vital for effectively using AI in problem solving, indicating the need for comprehensive educational strategies. (p. 2160)
I agree with the principle. But there’s a huge distance between knowing that algorithms carry biases and being able to spot those biases in a specific output you’re looking at on a Tuesday afternoon at work. “Understand AI models” means something very different for a fifth grader than it does for a data analyst. The framework could be more precise about what level of algorithmic understanding is realistic, practical, and useful at each stage of the pathway.
Keeping Humans in Control
One argument in the paper deserves particular attention. Chee et al. (citing Bond et al. 2019, and Shneiderman, 2020) state that “Humans must be central throughout the AI lifecycle—design, implementation, evaluation, operation, maintenance, decommissioning, and governance—to ensure high levels of oversight and control” (p. 2163). They go further, calling for a “human consciousness” approach to AI, one that aligns the technology with human values, protects human agency, and embeds social and ethical responsibility into every stage of AI use.
And the reason this needs saying is that the opposite is already happening. Shaw and Nave (2026) documented cognitive surrender, the tendency for people to accept AI outputs without critically evaluating them, especially when those outputs sound fluent and confident. If AI literacy teaches people to operate AI tools but never trains them to question what those tools produce, the whole project misses its most important target.
Chee et al. have built a well-organized, ambitious model. The developmental pathway from K-12 through higher education to the workforce addresses a gap other frameworks leave open. The inclusion of affective and career-related competencies broadens what AI literacy means in ways the field needs. The real test comes next: turning these competency categories into classroom activities, assessment tools, and professional training programs that work across very different contexts, resource levels, and learner populations.
We have the architecture. Now the profession needs to build what goes inside it.
References
- Bilbao-Eraña, A., & Arroyo-Sagasta, A. (2025). Fostering AI literacy in pre-service teachers: Impact of a training intervention on awareness, attitude and trust in AI. Frontiers in Education, 10, 1668078. https://doi.org/10.3389/feduc.2025.1668078
- Bond, R. R., Mulvenna, M. D., Wan, H., Finlay, D. D., Wong, A., Koene, A., Brisk, R., Boger, J., & Adel, T. (2019, October). Human centered artificial intelligence: Weaving UX into algorithmic decision making. In RoCHI (pp. 2–9)
- Chee, H., Ahn, S., & Lee, J. (2025). A competency framework for AI literacy: Variations by different learner groups and an implied learning pathway. British Journal of Educational Technology, 56, 2146-2182. https://doi.org/10.1111/bjet.13556
- Shaw, S. D., & Nave, G. (2026). Thinking fast, slow, and artificial: How AI is reshaping human reasoning and the rise of cognitive surrender. Working paper, The Wharton School, University of Pennsylvania. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646
- Shneiderman, B. (2020). Human-centered artificial intelligence: Reliable, safe & trustworthy. International Journal of Human–Computer Interaction, 36(6), 495–504.