Frictional Agent Alignment Framework: Slow Down and Don't Break Things

TL;DR

The paper introduces the Frictional Agent Alignment Framework (FAAF), a novel method for improving dynamic human-AI collaboration by generating context-aware friction to prompt reflection and realignment, outperforming existing methods in benchmarks.

Contribution

FAAF provides a new approach to belief alignment in dynamic settings by decoupling the alignment objective and enabling simple supervised training.

Findings

FAAF outperforms competitors in producing interpretable friction.

FAAF generalizes well out-of-distribution.

FAAF enhances the scalability of human-AI collaboration.

Abstract

AI support of collaborative interactions entails mediating potential misalignment between interlocutor beliefs. Common preference alignment methods like DPO excel in static settings, but struggle in dynamic collaborative tasks where the explicit signals of interlocutor beliefs are sparse and skewed. We propose the Frictional Agent Alignment Framework (FAAF), to generate precise, context-aware "friction" that prompts for deliberation and re-examination of existing evidence. FAAF's two-player objective decouples from data skew: a frictive-state policy identifies belief misalignments, while an intervention policy crafts collaborator-preferred responses. We derive an analytical solution to this objective, enabling training a single policy via a simple supervised loss. Experiments on three benchmarks show FAAF outperforms competitors in producing concise, interpretable friction and in OOD generalization. By aligning LLMs to act as adaptive "thought partners" -- not passive responders -- FAAF advances scalable, dynamic human-AI collaboration. Our code and data can be found at https://github.com/csu-signal/FAAF_ACL.