Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Emergent Communication through Negotiation
Kris Cao, Angeliki Lazaridou, Marc Lanctot, Joel Z Leibo, Karl Tuyls, Stephen Clark
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use binding, game-grounded proposals instead of linguistic cheap talk | Self-interested negotiating agents · Negotiation environment, random termination | fraction of joint reward: 0.75 ± 0.22 -> 0.87 ± 0.11 | Improved |
| Use linguistic cheap talk instead of the proposal channel | Prosocial negotiating agents · Negotiation environment, random termination | fraction of joint reward: 0.93 ± 0.10 -> 0.99 ± 0.02 | Improved |
| Add the proposal channel alongside the linguistic channel | Prosocial negotiating agents · Negotiation environment, random termination | fraction of joint reward: 0.99 ± 0.02 -> 0.92 ± 0.11 | Hurt |
| Use the proposal channel instead of no communication | Prosocial negotiating agents · Negotiation environment, random termination | fraction of joint reward: 0.95 ± 0.11 -> 0.93 ± 0.10 | Hurt |
| Allow the full 10 turns rather than random termination | Prosocial agents using the proposal channel · Negotiation environment | fraction of joint reward: 0.93 ± 0.10 -> 0.96 ± 0.07 | Improved |
| Reveal opponent IDs to the fixed selfish agent | Fixed selfish agent A, 1 prosocial opponent, proposal channel · Society negotiation | personal reward: 0.81 ± 0.17 -> 0.96 ± 0.06 | Improved |
| Reveal opponent IDs to the fixed selfish agent | Fixed selfish agent A, 1 prosocial opponent, linguistic channel · Society negotiation | personal reward: 0.97 ± 0.05 -> 0.97 ± 0.10 | No clear effect |
| Reveal opponent IDs to a fixed prosocial agent | Fixed prosocial agent A, 5 prosocial opponents · Society negotiation, linguistic channel | joint reward: 0.98* ± 0.06 -> 0.97 ± 0.06 | Hurt |
| Use linguistic cheap talk instead of no communication | Prosocial agents · Negotiation environment, random termination | fraction of joint reward: 0.95 ± 0.11 -> 0.99 ± 0.02 | Improved |
| Use linguistic cheap talk instead of no communication | Self-interested agents · Negotiation environment, random termination | fraction of joint reward: 0.77 ± 0.22 -> 0.75 ± 0.22 | No clear effect |
| Use linguistic messages to convey hidden utilities | Prosocial-prosocial agents · Linguistic message probing | agent B value accuracy (%): 20 -> 81 | Improved |
Closest papers by meaning
- Explaining Decisions of Agents in Mixed-Motive Games (AAAI 2025 · similarity 0.88)
- Semantics and Spatiality of Emergent Communication (NeurIPS 2024 · similarity 0.87)
- Cost-Effective Communication: An Auction-based Method for Language Agent Interaction (AAAI 2026 · similarity 0.87)
- Learning Multi-Agent Communication with Contrastive Learning (ICLR 2024 · similarity 0.87)
- Language Grounded Multi-agent Reinforcement Learning with Human-interpretable Communication (NeurIPS 2024 · similarity 0.87)
- Multi-Agent Reinforcement Learning with Communication-Constrained Priors (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
