Contestability Loops: LLM-Facilitated Micro-Deliberations for Culturally Contested Moderation

Shahid (2024) shows that current human–AI moderation ignores diverse mental models in the Majority World, leading to moral policing and missed local harms. Meanwhile, CounterQuill (Ding et al., 2024) demonstrates that AI can support learning and reflection, not just automated replies, in counterspeech. This project unites those insights: for culturally ambiguous flags, the system opens a “contestability loop,” a short, scaffolded deliberation among stakeholders. An LLM facilitates reflection, captures rationales, and produces a transparent summary for moderators—without deciding. Selective friction (per Sargeant et al., 2025) is explicitly invoked on these items, slowing the process and preserving contestability. The loop outputs two artifacts: (1) a final decision with traceable reasoning and (2) updated, machine-readable “micro-norms” that refine Lai et al.’s (2022) conditional delegation rules for the local context. This approach reframes appeals from a back-end process into a first-class, community-centered workflow. If successful, it improves legitimacy and cultural fit, and generates continuously improving, community-aligned moderation guidance.

References:

1. Human-AI Collaboration to Facilitate Culturally-Aware Content Moderation. Farhana Shahid (2024). CSCW Companion.
2. Human-AI Collaboration via Conditional Delegation: A Case Study of Content Moderation. Vivian Lai, Samuel Carton, Rajat Bhatnagar, Vera Liao, Yunfeng Zhang, Chenhao Tan, Q. Liao (2022). International Conference on Human Factors in Computing Systems.
3. CounterQuill: Investigating the Potential of Human-AI Collaboration in Online Counterspeech Writing. Xiaohan Ding, Kaike Ping, Uma Sushmitha Gunturi, Buse Çarik, Sophia Stil, Lance T. Wilhelm, T. Daryanto, James Hawdon, Sang Won Lee, E. Rho (2024). arXiv.org.
4. Unequal Uncertainty: Rethinking Algorithmic Interventions for Mitigating Discrimination from AI. Holli Sargeant, Mackenzie Jorgensen, Arina Shah, Adrian Weller, Umang Bhatt (2025). arXiv.org.