"Taking Stock at FAccT": Using Participatory Design to Co-Create a Vision for the Fairness, Accountability and Transparency Community
| Authors | Shiran Dudy et al. |
| Year | 2026 |
| Field | AI / ML |
| arXiv | 2604.16224 |
| Download | |
| Categories | cs.HC, cs.AI, cs.CY |
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.
Engineering Breakdown
Plain English
This paper describes a large-scale participatory design process used to govern the ACM FAccT (Fairness, Accountability, and Transparency) conference, demonstrating how to involve diverse stakeholders—academics, civil society, and government—in shaping conference governance and agenda. The authors combined three methodologies: an in-person CRAFT (Collaborative Reflection And Forward Thinking) session, an asynchronous Polis poll for distributed voting, and synthesis into a governance report for FAccT leadership. Participants directly shaped the conference by authoring seed statements, adding new statements, and voting to reveal patterns of agreement, disagreement, and uncertainty across the community. This represents one of the first large-scale attempts to apply reflexive participatory governance to a major AI/ML conference, making stakeholder positions visible and actionable.
Core Technical Contribution
The core contribution is a scalable, mixed-method participatory governance framework that combines synchronous in-person deliberation with asynchronous distributed decision-making via Polis (a digital platform for finding common ground). Rather than top-down conference planning, the authors designed a process where the community itself votes on substantive priorities, with voting patterns made transparent to reveal genuine consensus, disagreement, and nuance. The novelty lies not in individual components but in the orchestration: seed statements provide scaffolding, Polis aggregation reveals statistical patterns in participant beliefs, and the synthesis report translates community input into actionable governance decisions. This approach treats governance itself as a design problem amenable to participatory methods typically applied to product and system design.
How It Works
The process unfolds in three sequential phases. First, the in-person CRAFT session brings together a curated group of participants to author initial seed statements and engage in small-group dialogue about conference priorities and governance challenges. Second, these seed statements (plus new ones submitted by the broader community) are loaded into Polis, an asynchronous polling platform where participants vote on agreement/disagreement/uncertain for each statement; Polis uses clustering algorithms to identify groups of participants with similar voting patterns and surfaces majority and minority positions. Third, the team synthesizes voting results, disagreement patterns, and open-text responses into a structured governance report highlighting areas of consensus, key fault lines, and actionable recommendations. The output is a community-validated governance mandate that explicitly acknowledges where stakeholders diverge rather than imposing top-down priorities.
Production Impact
For organizations building AI/ML systems or governance structures, this framework offers a practical template for inclusive stakeholder engagement at scale without requiring everyone to attend in-person meetings. The Polis-based approach reduces facilitator bias and exhaustion that occur in purely deliberative processes; voting and clustering can handle hundreds to thousands of participants asynchronously, making it feasible for distributed communities. Engineers implementing similar governance systems should plan for: (1) careful curation of seed statements to avoid anchoring bias, (2) integration of Polis or similar platform APIs into governance workflows, (3) trained synthesis teams to translate voting data into actionable decisions (this is the bottleneck), and (4) feedback loops where the governance report is validated back to participants. The trade-off is increased process complexity and slower decision timelines compared to top-down governance, but the output gains legitimacy and captures minority positions that inform safer, more inclusive systems.
Limitations and When Not to Use This
This paper addresses conference governance rather than technical AI/ML challenges, so it has limited applicability to engineers building models or inference systems; the insights are primarily organizational and political rather than algorithmic. The Polis platform assumes participants have sufficient context and literacy to meaningfully evaluate statements—this may not hold for deeply technical governance questions or participants from non-English-speaking backgrounds, creating potential exclusion. The paper does not quantify how faithfully the final governance report reflects the voting data, nor does it address how dissenting minorities experience the process or whether minority positions are overwritten in synthesis. The approach also assumes good-faith participation; it is unclear how it scales when stakeholders have adversarial rather than collaborative intent, or how it handles participants who dominate statement creation despite efforts to democratize the process.
Research Context
This work builds on participatory design (PD) and deliberative democracy literature, adapting methods originally developed for product design (CRAFT sessions, consensus-building workshops) to conference governance. It is situated in the FAccT community's broader mission to democratize AI governance discourse and move beyond academic silos; prior FAccT work has focused on algorithmic auditing and stakeholder mapping, while this paper advances methodology for stakeholder participation itself. The paper also contributes to the emerging field of digital deliberation and participatory governance, extending platforms like Polis (developed for city budgeting and policy feedback) into the science governance space. This opens research directions in: (1) scaling participatory governance to global communities, (2) designing better synthesis methods to translate voting data into policy, (3) testing the long-term impact of participatory governance on community legitimacy and institutional outcomes, and (4) studying how Polis clustering algorithms might amplify or suppress minority voices.
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::