Fair Friction, Not Abstention: Uncertainty-Aware Human–AI Workflows for Equitable Moderation

Sargeant et al. (2025) highlight a counterintuitive risk: uncertainty-based selective abstention can disproportionately burden and disadvantage underrepresented groups that most often fall into the “uncertain” bucket. Building on that critique, this project proposes uncertainty-aware “selective friction” for content moderation—when the AI is uncertain, it does not silently abstain; instead, it surfaces a lightweight, context-specific pause with salient, structured cues for the human, preserving transparency and agency. We fuse Lai et al.’s (2022) conditional delegation—where humans specify “trust regions” for the model—with adaptive friction that activates only outside those regions. Li and Chau (2023) studied information cues and time constraints; we extend this by designing friction that is dynamically tuned to time pressure and cognitive load, measuring impacts on both accuracy and fairness. Novelty comes from moving past the field’s default of abstention under uncertainty toward a scalable, equitable human–AI workflow that measures disparate impact and decision latency in tandem. The impact is a legally and ethically grounded collaboration pattern that reduces error and inequity without grinding moderation queues to a halt.

References:

1. 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.
2. Human-AI Collaboration in Content Moderation: The Effects of Information Cues and Time Constraints. Haoyan Li, Michael Chau (2023). European Conference on Information Systems.
3. Unequal Uncertainty: Rethinking Algorithmic Interventions for Mitigating Discrimination from AI. Holli Sargeant, Mackenzie Jorgensen, Arina Shah, Adrian Weller, Umang Bhatt (2025). arXiv.org.