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

TL;DR

This paper presents a method called Chain of Alignment that combines public input and expert rules to evaluate and improve language model alignment with societal values, demonstrated in mental health domains.

Contribution

It introduces a novel approach to align language models with public will using normative objectives and expert-crafted rules, validated across mental health prompts.

Findings

Public normative objectives achieved with 96% public support

Expert-developed rules effectively evaluate model responses

High correlation (r=0.841) with human expert judgments

Abstract

We introduce a method to measure the alignment between public will and language model (LM) behavior that can be applied to fine-tuning, online oversight, and pre-release safety checks. Our `chain of alignment' (CoA) approach produces a rule based reward (RBR) by creating model behavior $\textit{rules}$ aligned to normative $\textit{objectives}$ aligned to $\textit{public will}$. This factoring enables a nonexpert public to directly specify their will through the normative objectives, while expert intelligence is used to figure out rules entailing model behavior that best achieves those objectives. We validate our approach by applying it across three different domains of LM prompts related to mental health. We demonstrate a public input process built on collective dialogues and bridging-based ranking that reliably produces normative objectives supported by at least $96\% \pm 2\%$ of the US public. We then show that rules developed by mental health experts to achieve those objectives enable a RBR that evaluates an LM response's alignment with the objectives similarly to human experts (Pearson's $r=0.841$, $AUC=0.964$). By measuring alignment with objectives that have near unanimous public support, these CoA RBRs provide an approximate measure of alignment between LM behavior and public will.