Metacognitive AI literacy: going beyond the AI skills gap agenda

Synthesis: Shapiro, Souto-Otero, and Watermeyer (2026) argue that conventional AI literacy frameworks anchored in functional skills acquisition fail to address the fundamental epistemological challenges posed by probabilistic, opaque algorithmic systems. They reconceptualize AI literacy as a metacognitive social practice that transcends individual competencies to encompass collective capacity for critical engagement with AI design, deployment, and governance. Drawing on case studies from higher education institutions and the Nordic financial sector, they illustrate how participatory co-design processes and experimental pedagogical spaces can cultivate metacognitive awareness and democratic agency — arguing that universities must evolve beyond skills transmission to become sites of collective intelligence that anchor AI literacy as a public good.

Key Findings

• AI literacy is a paradigm shift, not an extension of digital literacy: the probabilistic nature and epistemic opacity of contemporary AI systems (particularly large language models) render traditional rule-based digital literacy paradigms obsolete; learners must navigate probabilistic reasoning, continuously question outcomes, and consider AI's broader impacts.
• Reconceptualizes AI literacy as a metacognitive social practice — transcending individual skill acquisition toward collective critical engagement with AI design, deployment, and governance, rather than just functional tool use.
• Case studies show participatory co-design and experimental pedagogical spaces cultivate metacognitive awareness: examples include Aalborg (interdisciplinary project-based engineering), Aalto, Strathmore (GAEIA, engaging ~200 postgraduate students across Global North and South), Northeastern's "robot-proof" university / Humanics model, Stanford's sandbox environments, and a Nordic bank's Agentic AI implementation that used co-design as a vehicle for metacognition.
• Efficacious AI literacy requires institutional transformation: universities must evolve beyond skills transmission to become sites of collective intelligence where learners interrogate algorithmic power structures and mobilize alternative AI futures; AI literacy is framed as a governance problem, not merely a technical or educational challenge.
• The framework transforms universities across three dimensions: epistemically (beyond functional skills to interrogating how algorithmic systems reshape knowledge production), civically (cultivating ethical agency to contest power asymmetries), and institutionally (reclaiming legitimacy as custodians of collective intelligence rather than vendors of digital credentials).

Study Design & Method

This is a conceptual/argumentative article (not an empirical study) drawing on case studies and literature analysis. It synthesizes scholarship on digital/media literacies, critical pedagogy, algorithmic governance, and higher education transformation (including Giroux, Marginson, Williamson, Pangrazio, Crawford, Eubanks, and Watermeyer). The authors present illustrative case studies of AI literacy in practice across higher education institutions (Aalborg, Aalto, Strathmore, Northeastern, Stanford) and the adult continuing training sector (a Nordic bank's agentic-AI implementation), examining how metacognitive AI literacy is operationalized across pedagogical design, organizational governance, and socio-technical experimentation.

What this means for practice

• Instructors. Teach AI literacy as interrogation rather than tool operation: set tasks in which students work out how an algorithmic system reshaped a knowledge claim, so the probabilistic and opaque character of the model becomes the object of study instead of a background condition an AI course would leave untouched.
• Instructors. Assess the habit of monitoring and adjusting strategy under uncertainty — require students to say where a model's answer cannot be trusted — rather than rewarding fluent tool use, since scientific skepticism, not narrow functional proficiency, is the outcome the framework targets.
• Instructors. Build literacy through participatory co-design and experimental, low-stakes pedagogical spaces rather than a standalone AI module, following the cases the article documents: Aalborg's interdisciplinary project-based engineering, Aalto, Strathmore's GAEIA work with almost 200 postgraduate students across the Global North and South, Northeastern's "robot-proof" Humanics model, and Stanford's sandbox environments.
• Instructors. Run challenge-based collaboration with external partners so students meet real power and governance questions: in the Nordic bank case an Agentic AI rollout was treated as a facilitated learning process, with a cross-disciplinary team mapping customer journeys to decide what to automate while preserving human judgment.
• Administrators. Fund AI literacy as infrastructure for civic resilience rather than as a marketable credential, because the argument is that universities must become sites of collective intelligence whose learners interrogate algorithmic power structures — which means reconfiguring mission, AI Governance, and pedagogy, not adding courses. Institutions also owe this to learners in the three dimensions the article names: epistemic, civic, and institutional.

Limitations

As a conceptual article, the argument is not empirically tested, and the case studies are illustrative rather than systematically evaluated. The authors acknowledge the tension that universities operate under neoliberal funding regimes and "efficiency logics" that marginalize critical pedagogy, which limits the practical feasibility of the proposed transformation. The Nordic financial-sector case is Adult Learners-focused rather than higher-education-specific, and the institutional examples span very different national contexts without comparative assessment. The proposal for institutional transformation is programmatic rather than operationalized.

Citation

Shapiro, H., Souto-Otero, M., & Watermeyer, R. (2026). Metacognitive AI literacy: Going beyond the AI skills gap agenda. Learning, Media and Technology.