The second FemAIEdu session, led by Stockholm University, explores the concept of AI literacy and is anchored by a legal perspective on the EU AI Act. AI literacy is introduced as an emerging response to the profound transformation of societies by AI. A key distinction is drawn between education for AI, which prepares individuals to build and work with AI systems, and education with AI, which concerns the use of AI within teaching and learning practices. Within this framing, AI literacy is positioned as a foundational competence, necessary not only for technical participation but also for meaningful engagement with the societal implications of AI.

At the same time, the concept remains unsettled. Existing models of AI literacy vary widely, ranging from narrowly defined technical skills to broader frameworks that include ethical reasoning, collaboration, and critical evaluation. This lack of coherence is reinforced in the discussion, where participants emphasize that, across disciplines, literacies are defined in multiple and often conflicting ways, with no shared conceptual foundation.

The session then turns to the question of measurement, raising a fundamental challenge. If AI literacy is embedded in law, it must in some way be assessable. Two dominant approaches are outlined: (1) self-assessment and (2) performance-based testing. Both, however, are limited. Self-assessment is prone to bias, while performance-based methods struggle to remain relevant in a rapidly changing technological landscape. The discussion deepens this critique by questioning the underlying assumption that literacy can or should be measured in standardized ways. It highlights the risk that such approaches frame literacy as an individual attribute, thereby overlooking collective, institutional, and structural dimensions of knowledge and responsibility.

From this point, the session develops a detailed account of AI literacy within the EU AI Act. The law defines AI literacy as a combination of “skills, knowledge, and understanding” that enable informed engagement with AI systems, encompassing technical, practical, legal, ethical, and societal aspects. Providers and deployers of AI systems are required to ensure a “sufficient level” of literacy among those who interact with these systems. However, this requirement is deliberately open-ended. What counts as “sufficient” depends on context, including the type of system, its level of risk, the sector in which it is deployed, and the roles and responsibilities of the actors involved. AI literacy is therefore not a fixed standard but a relational and situational condition.

A central insight developed in the session is that AI literacy cannot be understood as an isolated legal requirement. It is deeply intertwined with other obligations within the AI Act, particularly those related to human oversight, transparency, and explainability. The requirement for meaningful human oversight in high-risk systems presupposes that individuals have the capacity to understand and critically assess system outputs. Similarly, transparency obligations only take effect if users have the literacy required to interpret the information provided. In this sense, AI literacy operates as a bridge between formal legal requirements and their practical implementation. Even where it is not directly enforceable, it remains essential for compliance.

The session also addresses the ongoing policy shift represented by the Omnibus proposal, which seeks to transform AI literacy from a binding obligation into a non-binding recommendation. This shift is situated within broader concerns about regulatory burden and economic competitiveness. While it has been criticized for potentially weakening governance structures and worker protections, it is also argued that its practical impact may be limited. Even without a binding requirement, organizations will still need to ensure a certain level of literacy to meet other enforceable obligations under the legal framework.

As the discussion unfolds, the very foundations of AI literacy are examined more critically. Questions are raised about whether it is meaningful to treat AI literacy as a stable concept, given the rapid evolution of technologies. The example of prompt engineering, once considered a key skill but now increasingly marginal, illustrates how quickly the object of literacy can shift. This leads to a broader reflection that both “AI” and “literacy” may function as overly abstract categories, potentially obscuring the dynamics they are meant to capture.

In response, the conversation expands toward a more systemic understanding of AI, emphasizing the importance of considering the full “AI stack”, from resource extraction and hardware production to data labor, cloud infrastructure, and application layers. From this perspective, AI is not merely a tool but a complex socio-technical system that reorganizes labor, resources, and social relations. AI literacy, therefore, must extend beyond technical knowledge to include an understanding of these broader structures. This expanded framing explicitly incorporates societal awareness, including issues of fairness, bias, and democratic implications, as integral components of literacy.

This broader perspective also enables a critique of skill-based approaches to addressing inequality. Parallels are drawn with earlier digital literacy initiatives, which often assumed that providing access and technical skills would be sufficient to reduce disparities. Such assumptions are challenged, as these approaches can overlook or even reproduce structural inequalities. AI literacy, if narrowly defined, risks repeating this pattern by focusing on individual competence while neglecting systemic conditions.

The pedagogical implications of these insights become a central concern. There is growing unease about the increasing influence of industry agendas on AI education, particularly through corporate training ecosystems that frame literacy in terms of productivity and tool use. In response, alternative approaches are considered, including low-tech or “unplugged” pedagogies that emphasize conceptual understanding and critical reflection. These approaches align with feminist pedagogical traditions that foreground embodiment, relationality, and situated knowledge, and aim to reveal rather than obscure the underlying dynamics of technological systems.

Feminist and critical perspectives play a crucial role in reframing AI literacy throughout the session. They shift the focus from individual skills to questions of power, responsibility, and justice. AI literacy is understood not only as the ability to use or understand systems, but as the capacity to critically engage with their implications and to challenge dominant narratives. Within this framing, AI literacy is closely tied to educational justice, emphasizing that without a critical approach, AI systems risk reinforcing existing inequalities and disproportionately disadvantaging marginalized groups.

The discussion of open source further complicates the relationship between transparency and literacy. While open source is often associated with greater accessibility and understanding, in contemporary AI systems, key elements such as training data and model construction often remain opaque. Nevertheless, engagement with open source can be seen as a valuable pathway toward a more active and critical form of literacy, enabling a shift from passive use to informed participation.

Across the session, several underlying tensions emerge. There is a tension between the need for stable concepts and the reality of rapid technological change, between individual and collective understandings of responsibility, and between legal operationalization and critical resistance to reduction. Rather than resolving these tensions, the session foregrounds them as central to the challenge of defining and operationalizing AI literacy.

In conclusion, the session moves beyond a narrow legal or technical understanding of AI literacy and develops a broader conception of it as a socio-technical, ethical, and political competence. AI literacy is not simply about understanding how systems function, but about understanding how they shape and are shaped by wider social, economic, and institutional contexts.