The (im)possibility of AI literacy

Synthesis: Pangrazio (2026) offers a critical editorial questioning whether AI literacy is a meaningful or even achievable goal. Tracing the history of literacy from its elite origins through mass institutionalization, she argues that AI literacy has been positioned as a "cure-all" — a solutionist, normative response to the complex and evolving phenomenon of AI. The analysis asks what "text" AI literacy actually refers to when AI is embedded, opaque, self-evolving, and deeply entangled in platforms and power structures, and whether current AI literacy discourse can move beyond instrumental skill-based framings to address the epistemological and political dimensions of AI — proposing that literacy, rights, and sovereignty responses must be developed in tandem.

Key Findings

• Literacy is never neutral: tracing literacy from sixteenth-century elite domains through the Protestant Reformation, mass institutionalization, and Freirean critical literacy, Pangrazio argues literacy has always been tied to power, access, cultural authority, and social order — yet this contested history is forgotten in the rush to make AI literacy a "cure-all."
• What is the "text" of AI? Conventional literacy applies to a text (a book, movie, article) that can be decoded and comprehended. For AI, the "text" is unclear — the algorithm? the Large Language Models (LLMs)? the data? This makes a critical-literacy deconstruction (identifying assumptions, contradictions, intertextual references) very difficult.
• AI is embedded, opaque, and evolving: AI is often not an isolated tool but part of an ecosystem (e.g., Copilot inside Microsoft 365), can be self-evolving/autonomous, is deliberately opaque (its construction is the domain of computer scientists), and most literacy approaches equate AI with generative AI, overlooking content-moderation and other instantiations.
• Programmatic practices: AI reshapes practices, values, and behaviors in ways that make critique harder — it curates and shapes public discourse while also being an object to discuss and regulate (Ananny's public object/public agent distinction). Criticality requires independent thought not shaped by the very technology being scrutinized.
• Digital literacies are declining and commercially captured: despite ~30 years of digital-literacies research, standardized tests (ACER 2023: only 46% of year-10 students proficient; ICILS 2023 declining) show digital literacy is falling. Meanwhile commercial tech companies (Microsoft, Google) have moved in to define literacies around their products, and AI literacy has been "cleaved off" from the broader field of digital literacies.
• Literacy may not be the right response: following Eynon, literacy "responsibilises the end user" rather than exploring AI Regulation in Education. Pangrazio proposes that literacy, rights (safe/ethical, unbiased, privacy-protecting AI), and sovereignty (countries/institutions developing their own AI and controlling data) responses should be developed in tandem, alongside speculative and interpretive responses.

Study Design & Method

This is a critical editorial (not an empirical study) in Learning, Media and Technology. It draws on a short history of literacy (Graff; Freire; Green; Janks; Luke), scholarship on critical digital literacies and algorithmic accountability (Ananny & Crawford; Seaver; Suchman; Nichols et al.; Pangrazio & Sefton-Green), and international assessments of digital literacy (ACER 2023; ICILS 2023). The argument proceeds through conceptual analysis of three challenges to AI literacy — the embedded/opaque/evolving nature of AI, the "programmatic practices" of AI, and the history of digital literacies — before proposing alternative (rights and sovereignty) responses.

What this means for practice

• Instructors. Locate AI literacy inside the broader field of critical digital literacies rather than treating it as a separate, rarefied program, and be candid that programs supplied by commercial vendors such as Google and Microsoft can produce only partial forms of the knowledge and skills required.
• Instructors. Make the missing "text" of AI an explicit teaching problem: ask students to name what exactly they are being asked to deconstruct — the algorithm, the model, the training data, or the interface that embeds it inside a product such as Microsoft 365 — since conventional critical decoding assumes a bounded text that AI does not clearly provide.
• Instructors. Defend curriculum time for critical engagement rather than skill drills, supporting students to write with, redesign, or even resist AI, and teach AI as a public object as well as a tool — it curates and shapes public discourse while also being the thing to be discussed and regulated — because criticality requires independent thought that is not shaped by the very technology under scrutiny.
• Instructors. Do not assume a baseline of digital fluency: digital literacies are declining on international measures (ACER 2023 found only 46% of year-10 students reached the proficient ICT standard, and ICILS 2023 recorded a further decline in 14-year-olds' computer and information literacy), so critical AI literacy has to be built on skills students may not have.
• Researchers. Stop treating literacy as the sole or sufficient response: develop literacy, rights (AI that is safe, unbiased, and does not jeopardize privacy), and sovereignty (institutions controlling their own AI and data) in tandem, because literacy on its own shifts responsibility onto the end user.

Limitations

As an editorial, the article presents an argumentative and conceptual position rather than empirical evidence; it does not offer a practical AI literacy framework or curriculum. Its historical and theoretical framing draws on Western (particularly Anglophone and Freirean) literacy traditions, and the proposal for rights- and sovereignty-based responses is programmatic rather than operationalized. The critique of commercial capture is illustrative rather than systematically evidenced.

Citation

Pangrazio, L. (2026). The (im)possibility of AI literacy. Learning, Media and Technology, 51(1), 1–7.