Emergent Communication through Negotiation

Abstract

Multi-agent reinforcement learning offers a way to study how communication could emerge in communities of agents needing to solve specific problems. In this paper, we study the emergence of communication in the negotiation environment, a semi-cooperative model of agent interaction. We introduce two communication protocols -- one grounded in the semantics of the game, and one which is \textit{a priori} ungrounded and is a form of cheap talk. We show that self-interested agents can use the pre-grounded communication channel to negotiate fairly, but are unable to effectively use the ungrounded channel. However, prosocial agents do learn to use cheap talk to find an optimal negotiating strategy, suggesting that cooperation is necessary for language to emerge. We also study communication behaviour in a setting where one agent interacts with agents in a community with different levels of prosociality and show how agent identifiability can aid negotiation.

Published as a conference paper at ICLR 2018

References in corpus (3)

• Adam: A Method for Stochastic Optimization
• Cooperating with Machines
• Emergence of Language with Multi-agent Games: Learning to Communicate with Sequences of Symbols

Cited by in corpus (20)

• The Hanabi Challenge: A New Frontier for AI Research
• On the Critical Role of Conventions in Adaptive Human-AI Collaboration
• Emergence of Communication in an Interactive World with Consistent Speakers
• The Emergence of Compositional Languages for Numeric Concepts Through Iterated Learning in Neural Agents
• Open Problems in Cooperative AI
• End-to-End Trainable Non-Collaborative Dialog System
• Emergence of Pragmatics from Referential Game between Theory of Mind Agents
• Decoupling Strategy and Generation in Negotiation Dialogues
• Provable Fictitious Play for General Mean-Field Games
• Learning to Communicate in Multi-Agent Reinforcement Learning : A Review
• Re-conceptualising the Language Game Paradigm in the Framework of Multi-Agent Reinforcement Learning
• Emergence of Numeric Concepts in Multi-Agent Autonomous Communication
• To Populate is To Regulate
• A perspective on multi-agent communication for information fusion
• Focus on What's Informative and Ignore What's not: Communication Strategies in a Referential Game
• The emergence of visual semantics through communication games
• Visual Perspective Taking for Opponent Behavior Modeling
• TexRel: a Green Family of Datasets for Emergent Communications on Relations
• Seq2Seq Mimic Games: A Signaling Perspective
• Negotiating Team Formation Using Deep Reinforcement Learning