Open questions on using generative AI for content quality measurement with human involvement

Determine effective methodologies for deploying generative AI to scale measurement of content quality and policy-violation labeling, and ascertain whether and how human involvement should be maintained in individual content moderation decisions.

Background

Section 9.1 argues that generative AI could provide cheaper and faster quality measurement, potentially enabling more comprehensive monitoring that traditionally required human labeling. Despite this potential, the authors explicitly flag significant open questions about implementation details.

A central unresolved issue is the appropriate role of human reviewers when generative AI is incorporated into content moderation workflows, particularly at the level of individual moderation decisions.