Chain of Alignment: Integrating Public Will with Expert Intelligence for Language Model Alignment

Synopsis

Framework where laypeople specify normative objectives ("public will") and domain experts translate them into model behaviour rules; case study in mental health shows ~96% public agreement on objectives and r=0.841 correlation between expert rule scoring and human-expert assessments.

Keywords

·alignment ·public will ·rule-based reward ·pluralistic alignment ·expert rules

Related entries

• AI Organizations are More Effective but Less Aligned than Individual Agents
  April 11, 2026 · arXiv
• Collective Constitutional AI: Aligning a Language Model with Public Input
  June 12, 2024 · FAccT 2024
• Position: Towards Bidirectional Human-AI Alignment
  June 13, 2024 · ICML 2024
• Training Language Models to Follow Instructions with Human Feedback
  March 4, 2022 · NeurIPS 2022
• The Triadic Loop: A Framework for Negotiating Alignment in AI Co-hosted Livestreaming
  April 20, 2026 · arXiv
• Institutional AI: Governing LLM Collusion in Multi-Agent Cournot Markets via Public Governance Graphs
  January 16, 2026 · arXiv