AI literacy curricula usually follow the same script: a definition of AI, a demonstration of prompts, and an ethics module bolted on at the end. After two years of watching this in the field, I’m not seeing it move the needle on critical thinking. Ford and colleagues (2026) make a different proposal worth taking seriously. They argue AI literacy should teach AI’s failures first, with errors treated as the curriculum’s center, not its appendix.
The case study comes out of a four-month project with 14 librarians from four libraries in Greater Sydney, including the University of Technology Sydney, TAFE NSW, and two public library groups. The team used eight phases to test whether learning about AI errors could anchor critical AI literacy, and the methodology has lessons for any educator designing AI literacy work right now.
How the Misbehaving Machine Works
The central artifact of the project is what the team calls “The Making of Misbehaving Machines,” an exhibition built around a deliberately broken AI chatbot. The team fine-tuned a small open-source language model (TinyLlama, then Llama 3.2 for stability) on a co-created dataset of question-answer pairs and ran it on a Raspberry Pi for energy efficiency.
The interface displayed the system prompt, the question, the AI’s answer, and a playful “uncertainty rating” calculated from the absurdity, answerability, and misbehavior scores of each prompt.
The goal was to make the machinery visible to learners, not to build a better chatbot. The exhibit showed how prompt design shapes output, how absurd questions produce absurd answers, and how confident-sounding text often has no grounding in fact. It gave librarians and visitors a way to see what’s normally hidden inside commercial AI products.
The librarian interviews surfaced one of the most concrete findings in the paper. Students come to reference desks asking for help finding citations that don’t exist. One librarian called this “chasing reference ghosts.” That’s a visible consequence of AI hallucination, and it’s already changing the everyday work of library reference.
Three Pedagogical Principles for Critical AI Literacy
The paper’s most useful contribution beyond the case study is a set of three principles for teaching AI literacy critically. These travel beyond libraries and beyond Sydney.
The first is to situate AI literacy in everyday practice. Learning works better when grounded in real workflows than when delivered as abstract ethics modules separate from the work. I’ve made this argument myself in previous posts on scaffolded AI literacy in academic librarianship and on the AI literacy gap inside academic libraries. Generic AI courses treat ethics as a wrap-up section. Ford and colleagues invert this. The ethical questions emerge from practical use, which is where they belong.
The second principle is speculative design. A deliberately broken AI gave participants space to think critically about the design choices baked into commercial products. This is the move the field has been missing.
Most AI literacy work asks students to be smart consumers of existing tools. Speculative design asks them to imagine how the tools could be different, which is a more powerful pedagogical lever. Roe, Furze, and Perkins (2025) made a related point in their digital plastic metaphor: AI literacy is a way of seeing, not a workshop topic.
The third is participatory research methods. The librarians in this project weren’t subjects to be studied. They were co-creators of the curriculum and the exhibit. Ford et al. report that “the project improved librarians’ confidence in their ability to evaluate the results of genAI tools at the expense of those tools” (p. 58). One participant, quoted by Ford and colleagues, wrote that “now my lack of confidence is in the LLM, not in my ability to prompt it” (p. 58). That shift matters. It’s the exchange of agency McCrary (2026) was calling for in his ghost-versus-architect framework.
The authors put the pedagogical case clearly: “learning about AI through AI error helps to better calibrate our relations with genAI tools: from passive consumer to actively questioning results and capabilities, and imagining how tools could operate differently” (p. 59). I agree, and the calibration metaphor is right. Critical AI literacy gives students a calibrated relationship with AI tools, knowing where to trust them and where not to.
Limitations
The paper is candid about what it didn’t do. The pilot involved 14 librarians, with 10 completing the post-project survey. Confidence gains were self-reported, not measured against objective learning outcomes. The exhibition ran for four weeks at four libraries, but the paper doesn’t include data on public engagement at scale. These are reasonable limits for a pilot, worth flagging before treating the model as proven.
The bigger question is scaling. A project like this needs design researchers, programming time, and fine-tuning expertise. Most library teams don’t have any of those. The authors flag this and propose making the materials open and the exhibits mobile, which is the right move.
There’s also a temporal question. The exhibition ran in late 2024, and AI tools have improved fast since then. A 2024 demonstration of obvious GPT-4 hallucinations may not land the same way in 2026, when models are better at hiding errors. The principle holds, but the specific failures need to keep changing.
The authors close with this: “we have demonstrated how generative it could be for AI literacy initiatives to experiment with the concept of AI error and uncertainty at a local level if those initiatives are situated, speculative and participatory” (p. 60). The qualifier “if” is doing real work. Critical AI literacy doesn’t happen because errors appear in the curriculum. It happens because errors are taught well, in context, with learners as co-makers of meaning.
References
- Ford, H., Burrell, A., Monin, M., Narayan, B., & Jethani, S. (2026). Hacking AI chatbots for critical AI literacy in the library. Journal of the Australian Library and Information Association, 75(1), 42-65. https://doi.org/10.1080/24750158.2026.2614000
- McCrary, Q. D. (2026). Are we ghosts in the machine? AI, agency, and the future of libraries. The Journal of Academic Librarianship, 52, 103181. https://doi.org/10.1016/j.acalib.2025.103181
- Roe, J., Furze, L., & Perkins, M. (2025). Digital plastic: A metaphorical framework for Critical AI Literacy in the multiliteracies era. Pedagogies: An International Journal. Advance online publication. https://doi.org/10.1080/1554480X.2025.2557491