The Model Card Authoring Toolkit: Toward Community-centered, Deliberation-driven AI Design

Abstract

There have been increasing calls for centering impacted communities – both online and offline – in the design of the AI systems that will be deployed in their communities. However, the complicated nature of a community’s goals and needs, as well as the complexity of AI’s development procedures, outputs, and potential impacts, often prevents effective participation. In this paper, we present the Model Card Authoring Toolkit, a toolkit that supports community members to understand, navigate and negotiate a spectrum of machine learning models via deliberation and pick the ones that best align with their collective values. Through a series of workshops, we conduct an empirical investigation of the initial effectiveness of our approach in two online communities – English and Dutch Wikipedia, and document how our participants collectively set the threshold for a machine learning based quality prediction system used in their communities’ content moderation applications. Our results suggest that the use of the Model Card Authoring Toolkit helps improve the understanding of the trade-offs across multiple community goals on AI design, engage community members to discuss and negotiate the trade-offs, and facilitate collective and informed decision-making in their own community contexts. Finally, we discuss the challenges for a community-centered, deliberation-driven approach for AI design as well as potential design implications.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 44 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.