Voices in the Loop: Mapping Participatory AI

Participatory AI remains geographically concentrated: The documented initiatives in the atlas are concentrated in a small number of countries, suggesting that participatory approaches to AI development are not yet globally distributed and may reflect existing inequalities in AI research and deployment.

Participation happens at specific stages, not throughout: When participation does occur, it is most often coded at problem formulation, evaluation, and governance stages rather than during model development or training—meaning communities typically have input on what problems to solve and how to assess solutions, but less influence on how the AI systems are actually built.

Documentation of participatory AI is fragmented: The paper addresses a significant gap by creating a unified repository, indicating that evidence about how participation is organized across different AI projects has been scattered across different sources, making it difficult for researchers and policymakers to learn from existing efforts.

The atlas enables ongoing scrutiny and improvement: By creating a "living inventory" with versioned releases, issue channels, and mechanisms for communities to contest or request redactions, the framework allows participatory AI initiatives to be continuously updated and challenged rather than treated as static case studies.

A reproducible protocol for tracking participation matters: The paper establishes standardized methods for discovering, vetting, and documenting participatory AI initiatives, which creates a foundation for consistent comparison and learning across different projects and contexts.