The first FemAIEdu meeting, led by KTH, began with an introduction to the Co-Liberative Computing research group's research perspective. The presentation outlined the group’s work at the intersection of feminism, decolonial thinking, and AI, with a focus on data practices, visualization, and large language models. The central goal of this work is to understand how power operates through data and AI systems and how these systems can be critically examined and challenged.

A key part of the presentation addressed how the group understands feminism in its research. While some feminist approaches focus on increasing women’s representation in positions of power, for example, through the idea of “breaking the glass ceiling”, this perspective was contrasted with more structural approaches to feminism. The research group primarily draws on intersectional feminist traditions, particularly ideas associated with “feminism for the 99%”. From this perspective, inequalities arise from the interaction of multiple social dimensions such as gender, race, class, culture, and sexuality. Addressing inequality, therefore, requires challenging the systems and structures that produce these forms of privilege and oppression.

Within this framework, the concept of data feminism provides an important foundation. Data feminism emphasizes that data is not neutral but functions as a form of power, unevenly distributed across society. Rather than rejecting data, the goal is to use data critically in order to challenge existing power structures. This perspective also stresses collaboration and co-liberation, meaning that social change should benefit all groups rather than only a privileged minority.

The presentation also emphasized the need to reconsider how success is measured in technological systems. In computer science and machine learning, systems are typically evaluated through technical metrics such as accuracy or performance. While these metrics remain important, they are not sufficient when the broader goal is social justice. Additional considerations, such as trust, fairness, resource sharing, and collective learning, should also be part of how technological systems are evaluated.

To analyze power within socio-technical systems, the presentation introduced Patricia Hill Collins’s concept of the “matrix of domination”. This framework describes how power operates across four interconnected domains: structural, disciplinary, hegemonic, and interpersonal. Structural power refers to laws and policies; disciplinary power concerns the technical systems that implement them; hegemonic power relates to cultural narratives and stereotypes; and interpersonal power appears in everyday interactions. According to this framework, meaningful change requires engagement across all four domains, rather than addressing only one dimension of the problem.

The discussion also highlighted three critical questions that help examine power in data science: who builds technological systems, for whom they are designed, and whose interests they ultimately serve. These questions reveal how technological systems can embed social inequalities. For example, when the developers of technology come from a narrow demographic background, their assumptions can shape system design in ways that marginalize other groups. In addition, many technologies are designed around the needs of dominant groups, leading to biases or exclusions in practice. Finally, the discussion noted that problems may arise both from missing data, when certain issues are ignored, and from excessive data collection, which can produce forms of surveillance that disproportionately affect marginalized communities.

The conversation then turned to the question of how these power structures can be challenged. While many efforts in the technology sector focus on data ethics, emphasizing fairness, transparency, and accountability, the discussion suggested that these approaches often treat inequality as a technical problem rather than a structural one. From a data-feminist perspective, addressing inequality requires moving beyond data ethics toward data justice, which emphasizes collective responsibility and structural change.

An example discussed during the presentation was Joy Buolamwini’s research on bias in facial recognition systems. Her work demonstrated that technical systems can reproduce social inequalities when training data are not diverse. While improving datasets can help address these biases, Buolamwini’s work also showed the importance of broader interventions, including policy engagement, public awareness, and education. This approach illustrates how challenges to technological systems must occur across multiple domains simultaneously.

Following the presentation, participants reflected on several conceptual questions relevant to the project. One issue raised was the distinction between data justice and AI justice. Some suggested that data itself may not be inherently unjust, but that contemporary AI systems, particularly large-scale systems such as generative models, are embedded in socio-technical infrastructures that can concentrate power and reproduce inequality. This raises broader concerns about environmental costs, geopolitical dynamics, and the hidden labor involved in producing AI systems.

Another topic of discussion concerned how feminist principles can inform design practices, as feminist approaches may shape not only the technologies produced but also the processes through which they are developed. Even when technological outputs appear similar, differences in design processes, such as collaboration, reflexivity, and inclusion, may lead to significantly different outcomes.

The discussion also emphasized the importance of AI literacy in higher education. Understanding AI systems is closely connected to questions of power. AI literacy may therefore be relevant for multiple groups, including engineers who design systems, educators who deploy them, students who interact with them, and institutional actors responsible for governance and procurement.

Finally, the team reflected on the need to identify a clear, distinctive contribution in the rapidly growing field of AI literacy research. Rather than attempting to develop a completely new literacy framework, the project may focus on integrating feminist theory, design practices, and legal or policy perspectives to produce a more critical, interdisciplinary approach to AI education.

The meeting will be followed by two more meetings led by SU and UoM. After these sessions, the team will work to narrow the project’s scope and define concrete outputs, such as joint publications, workshops, and future research proposals.