Explicit Cooperation Shapes Human-Like Multi-agent LLM Negotiation

A study presented at the First Workshop on Integrating NLP and Psychology to Study Social Interactions (NLPSI) @ICWSMS '25 found that GPT-based large language models (LLMs) achieve 50–90% lower success rates in negotiation role-play simulations without explicit cooperation instructions compared to instructed scenarios, and 40–80% lower success rates than human performance from past studies. Researchers Yanru Jiang and Gülşah Akçakır, who conducted the study using BATNA (Best Alternative to a Negotiated Agreement), reported that implicit personality-based cues like agreeableness had only marginal effects, suggesting explicit instructions remain essential for multi-agent LLMs to approximate human-like negotiation.

Abstract Humans develop cooperation heuristics in social decision-making, either intuitively or deliberatively. Large language models LLMs , which exhibit human-like heuristics across cognitive domains, may acquire prosocial tendencies through instruction tuning, latently encoded in their representations to foster cooperative behavior in social reasoning games. However, most studies of this kind either focus on cooperative language generation or explicitly instruct LLMs to cooperate, deviating from the inherent cooperation heuristics of humans. Our negotiation role-play simulations using BATNA Best Alternative to a Negotiated Agreement with a GPT-based LLM reveal that LLMs may struggle with cooperation in the absence of explicit instructions, showing a 50–90% lower success rate than in instructed scenarios and a 40–80% lower success rate than human performance reported in past studies. Implicitly inducing cooperation through personality traits had inconsistent effects, with agreeableness showing only a marginal influence and other traits exhibiting no systematic impact. These findings suggest that personality-based cooperation cues are subtle and that explicit instructions may still be essential for multi-agent LLMs to approximate human-like negotiation.- Anthology ID: - 2025.nlpsi-1.6 - Volume: Proceedings of the First Workshop on Integrating NLP and Psychology to Study Social Interactions NLPSI @ICWSM ’25 /volumes/2025.nlpsi-1/ - Month: - June - Year: - 2025 - Address: - Copenhagen, Denmark - Editors: Aswathy Velutharambath /people/aswathy-velutharambath/unverified/ , Sofie Labat /people/sofie-labat/ , Neele Falk /people/neele-falk/ , Flor Miriam Plaza-del-Arco /people/flor-miriam-plaza-del-arco/ , Roman Klinger /people/roman-klinger/ , Véronique Hoste /people/veronique-hoste/unverified/ - Venues: NLPSI /venues/nlpsi/ | WS /venues/ws/ - SIG: - Publisher: - Association for the Advancement of Artificial Intelligence www.aaai.org - Note: - Pages: - 57–71 - Language: - URL: https://aclanthology.org/2025.nlpsi-1.6/ https://aclanthology.org/2025.nlpsi-1.6/ - DOI: 10.36190/2025.34 https://doi.org/10.36190/2025.34 - Cite ACL : - Yanru Jiang and Gülşah Akçakır. 2025. Explicit Cooperation Shapes Human-Like Multi-agent LLM Negotiation https://aclanthology.org/2025.nlpsi-1.6/ . In Proceedings of the First Workshop on Integrating NLP and Psychology to Study Social Interactions NLPSI @ICWSM ’25 , pages 57–71, Copenhagen, Denmark. Association for the Advancement of Artificial Intelligence www.aaai.org . - Cite Informal : Explicit Cooperation Shapes Human-Like Multi-agent LLM Negotiation https://aclanthology.org/2025.nlpsi-1.6/ Jiang & Akçakır, NLPSI 2025 - PDF: https://aclanthology.org/2025.nlpsi-1.6.pdf https://aclanthology.org/2025.nlpsi-1.6.pdf