Participatory Design of AI Systems for Public Services

Abstract

This paper examines how participatory design methods can be adapted for the development of AI systems in public services. Through a year-long collaboration with Seattle-area social service agencies and their clients, we developed a framework for inclusive AI design that centers the needs and concerns of system users—particularly those from marginalized communities who are often most affected by algorithmic decision-making.

Key Findings

• •Traditional software design methods exclude voices most impacted by AI systems
• •Co-design sessions with service recipients identified critical blind spots in proposed AI tools
• •Iterative prototyping with community feedback reduced bias-related errors by 35%
• •Human oversight mechanisms emerged as a core requirement from participant input

Methodology

We employed a mixed-methods approach combining participatory design workshops, ethnographic observation, and quantitative analysis of AI system outputs. Our research team partnered with three Seattle-area agencies to co-design AI tools for resource allocation, intake screening, and service referrals. Participants included agency staff, current service recipients, and community advocates with lived experience of the systems being designed.

Implications

This research demonstrates that participatory design is not only feasible for AI systems but essential for developing tools that are fair, effective, and trusted by the communities they serve. The additional time and resources required for inclusive design processes are offset by improved system performance, reduced errors, and greater community acceptance. We recommend that public agencies mandate participatory design requirements for any AI procurement.