DiPS: Dialogue Policy Selection for High-Stakes Persuasion Agents

2026-07-01

AI Summary

This study addresses the challenge that large language models struggle to dynamically adapt persuasive strategies to individual differences in high-stakes scenarios, limiting their effectiveness. Focusing on fire evacuation persuasion, the authors propose DiPS, a dialogue policy selection framework that introduces reinforcement learning to high-risk human–AI persuasion for the first time. DiPS leverages a Q-learning mechanism combined with dialogue history and real-time semantic understanding to dynamically select the optimal persuasion strategy from a predefined policy set at each interaction turn. Experimental results demonstrate that DiPS significantly improves evacuation compliance rates in both simulated and real human interactions, outperforming zero-shot large language models and general retrieval-augmented generation (RAG) approaches.

Abstract

Large Language Models (LLMs) often struggle with persuasion in high-stakes scenarios. People's individual personalities and concerns require tailored strategies rather than a one-size-fits-all approach. To address this challenge, we focus on a fire-rescue scenario in which an operator must persuade a resident to evacuate as a high-stakes persuasion domain and propose Dialogue Policy Selection (DiPS), a Q-learning framework to dynamically select persuasion strategies adapted to the evolving conversational context. Specifically, we train a critic, trained to maximize the chance of evacuation success, to select a persuasion policy at each turn based on the resident's recent utterances.We then evaluate DiPS against multiple baselines in both simulated and real human interactions. We find that DiPS achieves higher evacuation success than a zero-shot LLM and generic RAG-augmented approach.

Problem

• persuasion
• high-stakes scenarios
• dialogue policy
• personalization
• evacuation

Innovation

• Dialogue Policy Selection
• Q-learning
• High-stakes Persuasion
• Personalized Strategy
• Conversational AI