"Taking Stock at FAccT": Using Participatory Design to Co-Create a Vision for the Fairness, Accountability and Transparency Community

Abstract: As a relatively new forum, ACM FAccT has become a key space for activists and scholars to critically examine emerging AI and ML technologies. It brings together academics, civil society members, and government representatives from diverse fields to explore the broader societal impacts of both deployed and proposed technologies. We report a large-scale participatory design (PD) process for reflexive conference governance, which combined an in-person CRAFT session, an asynchronous Polis poll and the synthesis of a governance-facing report for the FAccT leadership. Participants shaped the substantive agenda by authoring seed statements, adding new statements and making patterns of agreement, disagreement and uncertainty made visible through voting.Our endeavors represent one of the the first instances of applying PD to a venue that critically interrogates the societal impacts of AI, fostering a niche in which critical scholars are free to voice their concerns. Finally, this work advances large-scale PD theory by providing an effective case study of a co-design paradigm that can readily scale temporally and epistemologically.

• The paper demonstrates a scalable, infrastructure-mediated participatory design approach using the Polis platform to enable collective agenda-setting for conference governance.
• It utilizes a robust three-stage protocol—including seed statement generation, asynchronous community engagement, and report synthesis—to gather actionable insights from over 4,500 votes.
• The study reveals consensus on key issues like inclusivity and tiered pricing while highlighting persistent divisions on matters such as industry funding and governance structures.

Using Participatory Design to Shape Reflexive Conference Governance in FAccT

Introduction

"Taking Stock at FAccT": Using Participatory Design to Co-Create a Vision for the Fairness, Accountability and Transparency Community (2604.16224) presents a methodologically rigorous case study of large-scale participatory design (PD) for the governance of an AI ethics and socio-technical conference. The work advances PD itself as both theoretical and practical apparatus for collective agenda-setting and disagreement mapping in a distributed, interdisciplinary research community, with an emphasis on the FAccT (Fairness, Accountability, and Transparency) conference. The primary contribution is the demonstration of a scalable, infrastructure-mediated PD process using the Polis platform, culminating in actionable outputs for conference leadership and the broader community.

Methodology: Design and Execution of Large-Scale PD

The study operationalizes a three-stage PD protocol:

1. Seed Statement Generation: Initial agendas are co-constructed in-person using structured prompts, facilitating the articulation of both descriptive and normative perspectives.
2. Asynchronous Polis Engagement: A community-wide deployment via Slack and mailing lists enables open submission and voting on statements, with all contributions subject to machine-assisted and manual thematic classification.
3. Report Synthesis: Voting and statement patterns guide the construction of a governance-facing report, explicitly balancing operational and aspirational dimensions while revealing consensus, dissent, and uncertainty.

The empirical backbone comprises 128 participants generating 59 analyzed statements and 4,500+ votes, with sustained community engagement through both reminders and feedback incentives.

Figure 1: Temporal profile of participant recruitment, statement submission, and voting activity indicating coordinated engagement spikes aligned with process milestones.

Notably, the process preserves contributor anonymity in Polis, thereby reducing social pressure and potentially surfacing suppressed opinions, but at the cost of demographic granularity.

Results: Participation, Themes, and Consensus Mapping

Participation Dynamics

The analysis confirms robust engagement across participants, with minimal evidence of domination by hyperactive voters. Median and mean votes per participant are balanced ($\tilde{n}_{votes} = 40$ and $\bar{n}_{votes} = 35.5$), and statement authors also tend to participate at or above the average engagement rate.

Figure 2: Distribution of votes per participant demonstrating evenness of engagement without high-leverage outlier behaviors.

Statement exposure analysis reveals a recency and ordering effect, as early statements capture more votes, emphasizing infrastructural bias inherent in digital PD tools.

Figure 3: Voting counts per statement by order of submission, illustrating exposure bias and the imperative for mechanisms to remediate temporal advantage.

Thematic Landscape

Thematic analysis yields 11 salient categories, with the most prominent relating to conference format, inclusivity, geographic and disciplinary diversity, industry engagement, and community values.

Voting outcomes display strong consensus around openness, support for tiered pricing by economic context, and the provision of expanded collaboration/networking modalities (e.g., poster sessions, speed dating, non-paper-based tracks). At the same time, division persists on questions of industry funding, the boundaries of practical impact versus academic discourse, and optimal governance structures.

Figure 4: Distribution of statements along normative (future-oriented), descriptive (current state), and other classes; the majority foreground aspirational agenda setting.

Importantly, for areas such as the role of corporate sponsorship or the centrality of US/Western narratives, the process effectively reveals both the existence and contours of controversy without immediate forced consensus.

Deliberative Patterns and Active Negotiation

The platform enables direct and indirect participant reflection: some statements function explicitly as counters or elaborations to previously submitted claims, though the lack of explicit reply-linking is a technical constraint. The report presents both high-consensus, actionable recommendations and low-consensus items requiring ongoing negotiation.

Discussion

PD as Agenda-Setting and Legitimation Mechanism

This work demonstrates that scalable, platform-mediated PD can assemble broad-based, temporally flexible input into agenda setting for academic and activist collectives. The approach provides legitimacy for subsequent policy evolution through transparency and plurality, even when high consensus remains elusive.

However, the process does not extend to the implementation or sustained co-iteration of the governance artifact. This is both a limitation and a reflection of existing conference infrastructure and authority relationships—a constraint that future work must address through integration of formalized leadership response loops and proceduralized follow-on deliberation.

Scaling, Infrastructuring, and Visibility Regimes

Expanding participation socio-spatially (globally, asynchronously) introduces both reach and new opacity (e.g., exposure bias, loss of demographic tracking). The platform's design admits broad input but provides limited conversational depth; consensus mapping is prioritized over deep negotiation. Infrastructure choices—such as statement surfacing algorithms and voting queue properties—are not neutral and shape what becomes collectively legible.

Polis' scalability in facilitating many-to-many opinion clustering and mapping is clear, but it does not natively support iterative deliberation on mechanism or implementation, nor does it afford traceability for response threading.

Institutional and Theoretical Implications

The case demonstrates a concrete pathway for institutionalizing reflexive governance processes in research communities. It provides empirical backing for models wherein collective capabilities, as articulated by participant-authored statements and voting, can be surfaced and translated into governance artifacts. It also refines theoretical distinctions between participation, co-design, and infrastructuring—especially in transnational, epistemically pluralistic collectives.

Limitations

The voluntary, anonymous participation model introduces unknown sample bias and the possibility of multiple voting. There is substantial exposure bias owing to temporal ordering of statements. Polis' limited discourse expressivity and lack of feedback mechanisms constrain both analysis and process evolvability.

Conclusion

This work establishes a methodologically sound and theoretically reflective approach for large-scale participatory governance in the AI ethics community, operationalized in FAccT as a distributed sociotechnical collective. It provides both actionable recommendations and critical mapping of persistent disagreements, demonstrates the utility and boundaries of infrastructure-mediated PD, and offers a template for analogous interventions in other domains. Future iterations should seek deeper integration with organizational authority, mechanisms for iterative deliberation, and further mitigation of infrastructural bias.