mlsanityData to 14 Oct 2026

Training Language Models for Social Deduction with Multi-Agent Reinforcement Learning

arXiv 2502.06060 ↗First posted 9 Feb 2025cs.AI cs.CL cs.LGcited by 16

On the map: LLM agents and tool use → General LLM agent frameworks and benchmarks

Abstract (arXiv)

Communicating in natural language is a powerful tool in multi-agent settings, as it enables independent agents to share information in partially observable settings and allows zero-shot coordination with humans. However, most prior works are limited as they either rely on training with large amounts of human demonstrations or lack the ability to generate natural and useful communication strategies. In this work, we train language models to have productive discussions about their environment in natural language without any human demonstrations. We decompose the communication problem into listening and speaking. Our key idea is to leverage the agent's goal to predict useful information about the world as a dense reward signal that guides communication. Specifically, we improve a model's listening skills by training them to predict information about the environment based on discussions, and we simultaneously improve a model's speaking skills with multi-agent reinforcement learning by rewarding messages based on their influence on other agents. To investigate the role and necessity of communication in complex social settings, we study an embodied social deduction game based on Among Us, where the key question to answer is the identity of an adversarial imposter. We analyze emergent behaviors due to our technique, such as accusing suspects and providing evidence, and find that it enables strong discussions, doubling the win rates compared to standard RL. We release our code and models at https://socialdeductionllm.github.io/

Key facts. Cited by 16 papers in the map (data through 14 Oct 2026). Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use. Builds on 15 of its 37 references that are in the map. First posted on arXiv 9 Feb 2025.

Builds on

15 of 37 references.

References · 15

  1. 2017Proximal Policy Optimization Algorithmscited by 15,215
  2. 2019On the Utility of Learning about Humans for Human-AI Coordinationcited by 349
  3. 2023Text2Motion: From Natural Language Instructions to Feasible Planscited by 224
  4. 2020"Other-Play" for Zero-Shot Coordinationcited by 198
  5. 2023Large Language Models as General Pattern Machinescited by 197
  6. 2016Multi-Agent Cooperation and the Emergence of (Natural) Languagecited by 185
  7. 2017Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbolscited by 125
  8. 2023GPT-4 Doesn't Know It's Wrong: An Analysis of Iterative Prompting for Reasoning Problemscited by 89
  9. 2019Continual adaptation for efficient machine communicationcited by 28
  10. 2022Sample-Efficient Reinforcement Learning of Partially Observable Markov Gamescited by 27
  11. 2021Learning to communicate about shared procedural abstractionscited by 15
  12. 2024How Well Can a Long Sequence Model Model Long Sequences? Comparing Architechtural Inductive Biases on Long-Context Abilitiescited by 9
3 more
  1. 2022Hidden Agenda: a Social Deduction Game with Diverse Learned Equilibriacited by 7
  2. 2023Hidden-Role Games: Equilibrium Concepts and Computationcited by 6
  3. 2024Optimizing Robotic Manipulation with Decision-RWKV: A Recurrent Sequence Modeling Approach for Lifelong Learningcited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 9; 2026: 7.

16 papers.

Sort:

By area:

  1. 2025SPIRAL: Self-Play on Zero-Sum Games Incentivizes Reasoning via Multi-Agent Multi-Turn Reinforcement Learningcited by 100
  2. 2025SPICE: Self-Play In Corpus Environments Improves Reasoningcited by 83
  3. 2025LLM Collaboration With Multi-Agent Reinforcement Learningcited by 49
  4. 2025Chasing Moving Targets with Online Self-Play Reinforcement Learning for Safer Language Modelscited by 39
  5. 2025Among Us: A Sandbox for Measuring and Detecting Agentic Deceptioncited by 26
  6. 2025MetaMind: Modeling Human Social Thoughts with Metacognitive Multi-Agent Systemscited by 13
  7. 2026SPADE: Self-Play in Adaptive Synthetic Executable Environmentscited by 12
  8. 2025Bayesian Social Deduction with Graph-Informed Language Modelscited by 9
  9. 2026Learning Decentralized LLM Collaboration with Multi-Agent Actor Criticcited by 8
  10. 2025AI Agent Behavioral Sciencecited by 6
  11. 2026QUACK: Questioning, Understanding, and Auditing Communicated Knowledge in Multimodal Social Deduction Agentscited by 3
  12. 2025Shall We Play a Game? Language Models for Open-ended Wargamescited by 2
4 more
  1. 2026Evaluating Large Language Models in a Complex Hidden Role Gamecited by 1
  2. 2026From RLVR to RLSVR: Task Transformation Induces Self-Verifiable Rewards for Open-Ended LLM Self-Improvementcited by 1
  3. 2026Revac: A Social Deduction Reasoning Agentcited by 0
  4. 2026Synthetic Linguistic Agency: How an Embodied Mortal Agent Learns Linguistic Affordances through Consequential Social Experiencecited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Playing social deduction games with reinforcement fine-tuned large language models
  2. 2024Language Grounded Multi-agent Reinforcement Learning with Human-interpretable Communication
  3. 2025SpeakRL: Synergizing Reasoning, Speaking, and Acting in Language Models with Reinforcement Learning
  4. 2025The Stackelberg Speaker: Optimizing Persuasive Communication in Social Deduction Games
  5. 2024Teaching Embodied Reinforcement Learning Agents: Informativeness and Diversity of Language Use
  6. 2025Learning Robust Social Strategies with Large Language Models
  7. 2026The Five Ws of Multi-Agent Communication: Who Talks to Whom, When, What, and Why -- A Survey from MARL to Emergent Language and LLMs
  8. 2026The Rise of Verbal Reinforcement Learning