Why University AI Literacy Must Move Beyond Technical Prompting Skills
- University AI instruction currently focuses primarily on technical competencies like prompt engineering, hallucination detection, and bias recognition.
- Research from the University of Exeter advocates expanding curricula to cover socio-technical systems, ethical governance, student agency, and futures thinking.
- Critical literacy exercises encourage students to examine who benefits from specific AI architectures and whose knowledge is privileged.
- Verifiable academic practice requires transparent tool use alongside a clear understanding of the institutional choices embedded in automated systems.
University classrooms and academic departments increasingly find themselves managing generative software through mechanical checklists. Students craft prompt strings and check generated text for factual hallucinations. Before submitting coursework, they might also scan outputs for obvious algorithmic bias. These operational steps prevent immediate errors, but they frame technology purely as an engineering utility to be operated correctly rather than an institutional structure that needs critical evaluation.
A broader pedagogical proposal challenges this narrow operational focus. In Times Higher Education’s analysis of embedding critical AI literacy in university curricula, Katie Ledingham and Sarah Hartley argue that higher education must move beyond technical competency to teach students how artificial intelligence functions as a socio-technical system. Instead of treating software models as neutral tools, the authors outline four structural perspectives rooted in responsible innovation. These involve examining the physical infrastructure of data centres and supply chains, treating governance choices as political decisions, building student agency to challenge system designs, and using futures thinking to ask whether automated systems ought to be deployed in specific settings. Practical classroom exercises accompany these principles. Scholars might be asked to identify whose knowledge a specific algorithm privileges, or map which commercial entities control model development.
This perspective shifts how academic integrity and honest tool use are understood in scholarly work. Treating software solely as a prompt-and-output mechanism encourages reliance on superficial technical fixes. By contrast, examining underlying data architectures helps scholars understand why systems privilege certain sources over others. Verifiable academic practice requires transparent acknowledgement of how automated systems are used, alongside an understanding of their limitations. When institutions equip students to interrogate the material conditions, political choices and institutional priorities embedded within digital tools, scholarship moves away from passive adoption towards rigorous critical inquiry.
When universities limit AI education to technical instruction, do they risk training students to adapt to automated systems rather than preparing them to shape how technology serves society?