What participatory traditions shape AI design practice?

To make sense of the "enormous variation" in how AI researchers claim to do "participatory AI," the paper reviews participatory traditions across technology design, political theory, and the social sciences, pulling out eight distinct approaches with their own goals and power dynamics: user-centered design (UCD), which "involv[es] end users during needfinding and evaluation of designs"; service design, which considers "how... design decisions impact stakeholders" using service blueprints and value-stream mapping; participatory design (PD), which "enable[s] non-expert stakeholders to provide direct input on technology design" and is explicitly meant to "challenge power structures"; co-design, which shares PD's cooperative stance but "typically lacks its explicitly political component" and focuses more on specific artifacts than the larger sociotechnical system; value-sensitive design (VSD), which centers the values of direct and indirect stakeholders; participatory action research (PAR), which treats stakeholders as "co-inquirers" who help set research questions; social choice theory and mechanism design, which mathematically aggregate stakeholder preferences into a ranking; and participatory democracy and civic participation, which involve citizens in civic decision-making.

The paper draws these eight traditions together not to recommend one over the others but to supply a shared reference point: a way to locate any given "participatory AI" method along axes such as who gets a say, what counts as a legitimate claim, and how directly stakeholder input shapes an outcome. Naming PD's explicit politics against co-design's narrower artifact focus, for instance, makes visible that two methods both called "participatory" can grant very different amounts of agency.

Where How should humans and AI systems divide decision-making tasks? and How do AI novelty and scope shape employee relationships? build typologies of human-AI task and relationship structure after a system already exists, this framework reaches earlier, into the design traditions that decide who is even consulted before a system is built. It also gives How should AI systems evolve alongside human work practices? a more granular vocabulary for its design-stage level of analysis, naming specific participatory methods — PD, co-design, VSD, PAR — that could populate that level.

The excerpt presents this typology as the lens for the paper's own corpus analysis rather than as a tool validated independently of that analysis, and it does not resolve the tension Birhane and colleagues raise elsewhere in the literature, that there is "no clear consensus on what minimum set of standards or dimensions one should use to assess... a given potential participatory mechanism." The implication is that the typology is best read as a vocabulary for comparison across methods, not a settled standard for certifying any one method as adequately participatory.