Participation in AI: Notes from The Trenches

Abstract

As AI systems become more closely intertwined in public lives, there have been calls to build these systems with greater public participation through the AI development lifecycle. The goals for participation in AI are instrumental and normative. Participation can result in design improvements, greater transparency, and further belief in democratic ideals. Through the case study of one project's attempt to create a space for participation in AI, this paper surfaces the complexities of building participatory AI in practice such as heterogeneity within a community and the inevitability of opacity of technical processes. We highlight the need for translation between different communities involved in an AI process that generates a partial understanding of the skills and context of other stakeholders in the process. We propose that trust in participatory processes can exist despite the opacity and participation inequality. This however requires a commitment to processes for enabling participation, even if there is some distance in the ideal and reality of a participatory project.

Keywords: participatory AI, community led AI, Responsible AI