cs.HC提交于 2026-10-26已译

PAIR-SAFE：面向 AI 辅助心理健康支持的运行时审计与改进配对智能体方法

PAIR-SAFE: A Paired-Agent Approach for Runtime Auditing and Refining AI-Mediated Mental Health Support

Jiwon Kim · Violeta J. Rodriguez · Dong Whi Yoo · Eshwar Chandrasekharan · Koustuv Saha

中文摘要

大语言模型(LLM)正日益用于心理健康支持，但其回复可能过于指令化、不一致或偏离临床规范。现有风险缓解方法主要依靠训练或提示进行隐式对齐，运行时问责能力有限。PAIR-SAFE 将 Responder 智能体与 Judge 智能体配对，后者依据临床验证的动机式访谈治疗完整性(MITI-4)进行审计，并对每条回复给出允许(allow)或修改(revise)决定，以指导实时改进。研究使用源自人工标注数据的支持寻求者模拟器评估互动。Judge 监督显著提升了 Partnership、Seek Collaboration 及整体关系质量等关键 MITI 维度。专家评估进一步佐证了量化结果，并揭示运行时监督的细微作用，表明配对智能体审计可为 AI 辅助心理健康支持提供具有临床依据的透明监督。

关键要点

01现有 LLM 心理健康支持回复可能指令化、不一致或偏离临床规范。
02PAIR-SAFE 以 Responder 和 Judge 配对，依据 MITI-4 实时审计回复。
03Judge 对回复作出 allow 或 revise 决定，支持运行时改进与问责。
04在人工标注数据衍生的支持者模拟器中，关系质量等 MITI 指标显著提升。
05专家评估佐证量化结果，显示临床透明监督的价值与细微差异。

解读

尚无解读。

原始英文摘要

arXiv:2601.12754v2 Announce Type: replace-cross Abstract: LLMs are increasingly used for mental health support, yet they can produce responses that are overly directive, inconsistent, or clinically misaligned. Most risk mitigation approaches rely on implicit alignment through training or prompting, offering limited runtime accountability. We introduce PAIR-SAFE, a paired-agent framework that integrates a Responder agent with a Judge agent grounded in the clinically validated Motivational Interviewing Treatment Integrity (MITI-4). The Judge audits each response and issues \allow{} or \revise{} decisions to guide real-time refinement. We evaluate interactions using a support-seeker simulator derived from human-annotated data. The Judge-supervised interactions show significant improvements in key MITI dimensions, including Partnership, Seek Collaboration, and overall Relational quality. Our quantitative findings are corroborated by expert evaluation, highlighting nuances of runtime supervision. Our study reveals that paired-agent auditing can provide clinically grounded, transparent oversight for AI-assisted mental health support.