文章背景与核心概要

《成交还是拒绝：情绪在多智能体谈判中的作用》（Deal Me Maybe）是一项探讨情绪状态对大语言模型（LLM）智能体在价格谈判中影响的前沿研究。传统基准测试通常将AI智能体视为情感中立的实体，而该研究则突破性地证明了情绪是驱动讨价还价行为的关键因素。通过在36种场景中为买卖双方智能体赋予六种不同的情绪状态，研究人员发现情绪会显著影响交易成功率、价格轨迹以及终止行为。

该研究的核心意义在于揭示了情绪条件作用不仅改变了智能体的语言风格，还从根本上改变了谈判的战略结果。这一发现为未来在真实商业环境中部署AI智能体提供了重要的参考依据，凸显了在设计和监管AI时考虑情绪细微差别（如愤怒、快乐等）的必要性，这往往决定了一笔交易的成败。

Deal Me Maybe: The Role of Emotions in Multi-Agent Negotiation

Authors: Massimiliano Luca, Apoorva Singh, Bruno Lepri
Date: August 7, 2026
Subject: Artificial Intelligence (cs.AI)
Identifier: arXiv:2608.06922

Summary

Deal Me Maybe investigates the impact of emotional states on Large Language Model (LLM) agents during price negotiations. While traditional benchmarks often treat AI agents as emotionally neutral, this study demonstrates that emotions are a critical driver of bargaining behavior. By assigning six distinct emotional states to buyer and seller agents across 36 scenarios, the researchers found that emotions significantly influence deal success rates, price trajectories, and termination behaviors. The findings highlight the necessity of considering emotional conditioning when deploying AI agents in commercial environments.

概要

《成交还是拒绝》探讨了情绪状态对大语言模型（LLM）智能体在价格谈判过程中的影响。虽然传统的基准测试通常将AI智能体视为情感中立的实体，但本研究表明，情绪是驱动讨价还价行为的关键因素。通过在36个场景中为买方和卖方智能体分配六种不同的情绪状态，研究人员发现情绪会显着影响交易成功率、价格轨迹和终止行为。这些发现凸显了在商业环境中部署AI智能体时考虑情绪条件作用的必要性。

Key Findings

1. The Impact of Emotion on Deal Success

The study reveals a stark contrast in outcomes based on the assigned emotional state: * Anger: Acts as a major deterrent, with angry buyers reaching an agreement in only 0.39% of cases. * Happiness: Promotes the highest agreement rate at 28.91%, though happy buyers often secure less favorable prices compared to those exhibiting fear.

1. 情绪对交易成功率的影响

该研究揭示了基于所分配的情绪状态所产生的结果的鲜明对比： * 愤怒（Anger）： 充当主要的威慑力量，愤怒的买家达成协议的比例仅为 0.39%。 * 快乐（Happiness）： 促进了最高的达成协议率，达到 28.91%，不过与表现出恐惧的买家相比，快乐的买家通常获得的优惠价格较少。

2. Role-Dependent Dynamics

The influence of emotion is not uniform; it varies significantly based on the agent's role: * Buyer Emotion: Primarily dictates the acceptance or rejection of a deal. * Seller Emotion: Primarily shapes the concession dynamics and the path toward a final price.

2. 角色依赖的动态机制

情绪的影响并不是统一的；它会根据智能体的角色而产生显著变化： * 买方情绪： 主要决定交易的接受或拒绝。 * 卖方情绪： 主要塑造让步动态以及通往最终价格的路径。

3. Broader Implications

The research concludes that emotional conditioning affects more than just the linguistic style of the agents; it fundamentally alters the strategic outcome of the negotiation. This raises important considerations for the development and regulation of AI agents designed for real-world commerce, where emotional nuance can be the difference between a closed deal and a failed interaction.

3. 更广泛的启示

该研究得出结论：情绪条件作用不仅影响智能体的语言风格，还从根本上改变了谈判的战略结果。这为面向现实世界商业环境而设计的AI智能体的开发和监管提出了重要的考量，在这些场景中，情绪的细微差别往往决定了一次交易的达成与否。

Research Metadata

研究元数据

Access & Citations

• BibTeX: Available via arXiv
• External Tools: Google Scholar | Semantic Scholar | NASA ADS

访问与引用

• BibTeX： 可通过 arXiv 获取
• 外部工具： Google Scholar | Semantic Scholar | NASA ADS