AI From the Margins (AIM): Rethinking Participatory AI Design Through the Lived Experience of Minoritized Communities

Researchers propose AI From the Margins (AIM), a methodological framework that centers the lived experiences of minoritized communities in participatory AI design before problem definitions are established. The approach was tested in a Dutch healthcare context through narrative elicitation, co-constructed rule-making, and policy dialogue, demonstrating that grounding AI design in community experience fundamentally reshapes project goals and outcomes.

This research addresses a critical gap in participatory AI design: the timing and depth of community involvement. Traditional participatory approaches engage stakeholders after foundational decisions are made, effectively limiting influence to implementation details rather than fundamental purpose. AIM inverts this sequence by establishing lived experience as the starting point, allowing marginalized communities to define what problems AI should address and whether AI involvement is appropriate at all.

The Dutch healthcare study demonstrates practical implementation through four interconnected techniques. Narrative elicitation using BNIM captures complex, contextual experiences that quantitative methods often miss. Co-constructed rule-making gives participants agency in establishing boundaries and principles. Critically, participants determined whether and how AI should be involved—a power typically retained by technologists and administrators. The translation phase bridges community insights with policy language, ensuring experiences inform institutional decision-making rather than disappearing into project documentation.

For AI development practitioners, this framework addresses growing accountability pressures and documented harms in deployed systems affecting marginalized populations. Healthcare AI systems have consistently demonstrated disparate impacts on communities of color, making this methodological contribution practically valuable beyond academic interest. The participant feedback emphasizing substantive engagement and requesting continuation suggests the approach generates genuine buy-in, potentially improving both ethics and adoption outcomes.

Looking forward, the field should examine whether AIM principles scale beyond small-group sessions to influence large-scale AI governance. The framework's flexibility—described as preconditions rather than fixed protocol—may enable adaptation across sectors and cultural contexts. Critical questions remain about how organizations integrate preparatory, experience-centered work into timelines and budgets dominated by technical development cycles.

• →AIM establishes lived experience of marginalized communities as the foundation for AI design rather than as input after core decisions are made.
• →The Dutch healthcare trial combined narrative elicitation, co-rule-making, and participant-driven AI involvement determination with policy translation.
• →Participants described the engagement as substantive and requested continuation, indicating genuine rather than performative participation.
• →The framework addresses documented disparities in AI outcomes for communities of color, particularly in healthcare applications.
• →Implementation requires organizational commitment to timeline flexibility and genuine power-sharing with marginalized communities.