From Descriptive to Prescriptive: Uncover the Social Value Alignment of LLM-based Agents

2026-05-13

AI Summary

Current large language model (LLM) agents exhibit insufficient alignment with human societal values in self-awareness, moral decision-making, and emotional expression. This work proposes a novel value-based alignment framework that integrates Maslow’s hierarchy of needs and Plutchik’s emotion wheel to formulate normative behavioral expectations. By leveraging GraphRAG, the approach translates abstract social principles into context-sensitive value instructions embedded within the LLM’s reasoning process. The retrieval mechanism facilitates a shift from descriptive knowledge to prescriptive guidance, enabling more spontaneous affective expression in AI agents. Evaluated on the DAILYDILEMMAS benchmark, the proposed framework significantly outperforms baseline methods—including Emotion Chain-of-Thought (ECoT), Plan-and-Solve, and metacognitive prompting—demonstrating markedly improved consistency between agent behaviors and societal values.

Abstract

Wide applications of LLM-based agents require strong alignment with human social values. However, current works still exhibit deficiencies in self-cognition and dilemma decision, as well as self-emotions. To remedy this, we propose a novel value-based framework that employs GraphRAG to convert principles into value-based instructions and steer the agent to behave as expected by retrieving the suitable instruction upon a specific conversation context. To evaluate the ratio of expected behaviors, we define the expected behaviors from two famous theories, Maslow's Hierarchy of Needs and Plutchik's Wheel of Emotion. By experimenting with our method on the benchmark of DAILYDILEMMAS, our method exhibits significant performance gains compared to prompt-based baselines, including ECoT, Plan-and-Solve, and Metacognitive prompting. Our method provides a basis for the emergence of self-emotion in AI systems.

Problem

• social value alignment
• LLM-based agents
• dilemma decision
• self-emotion
• human values

Innovation

• value alignment
• GraphRAG
• LLM-based agents
• social values
• self-emotion