Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Collaborating with Humans without Human Data
DJ Strouse, Kevin R. McKee, Matthew Botvinick, Edward Hughes, Richard Everett
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 |
|---|---|---|---|
| add past partner checkpoints representing different training skill levels (ablation) | FCP · H_proxy | deliveries: 4.7±0.4 -> 10.6±0.5 | Improved |
| add past partner checkpoints representing different training skill levels (ablation) | FCP · Diverse SP | deliveries: 6.9±0.1 -> 11.2±0.1 | Improved |
| add past partner checkpoints representing different training skill levels (ablation) | FCP · Random | deliveries: 1.0±0.1 -> 8.6±0.2 | Improved |
| add architectural variation to the partner population (ablation) | FCP with past checkpoints · H_proxy | deliveries: 10.6±0.5 -> 9.9±0.6 | No clear effect |
| add architectural variation to the partner population (ablation) | FCP with past checkpoints · Diverse SP | deliveries: 11.2±0.1 -> 11.1±0.4 | No clear effect |
| add architectural variation to the partner population (ablation) | FCP with past checkpoints · Random | deliveries: 8.6±0.2 -> 8.4±0.4 | No clear effect |
| add architectural variation when past checkpoints are unavailable (ablation) | FCP−T · H_proxy | deliveries: 4.7±0.4 -> 7.0±0.8 | Improved |
| add architectural variation when past checkpoints are unavailable (ablation) | FCP−T · Diverse SP | deliveries: 6.9±0.1 -> 8.6±0.4 | Improved |
| add architectural variation when past checkpoints are unavailable (ablation) | FCP−T · Random | deliveries: 1.0±0.1 -> 3.2±0.5 | Improved |
| increase partner population size from N=4 to N=32 (ablation) | FCP · H_proxy | deliveries: 8.4 (0.3) -> 10.6 (0.5) | Improved |
| increase partner population size from N=32 to N=64 (ablation) | FCP · H_proxy | deliveries: 10.6 (0.5) -> 10.4 (0.3) | No clear effect |
| train with diverse self-play partners and their past checkpoints instead of self-play, population-play, or behavioral-cloning play | FCP · held-out agent partners and human partners | deliveries | Improved |
Closest papers by meaning
- Learning to Cooperate with Humans using Generative Agents (NeurIPS 2024 · similarity 0.91)
- Cross-environment Cooperation Enables Zero-shot Multi-agent Coordination (ICML 2025 · similarity 0.89)
- Diversity Is Not All You Need: Training A Robust Cooperative Agent Needs Specialist Partners (NeurIPS 2024 · similarity 0.88)
- Beyond Single Stationary Policies: Meta-Task Players as Naturally Superior Collaborators (NeurIPS 2024 · similarity 0.88)
- Adaptively Coordinating with Novel Partners via Learned Latent Strategies (NeurIPS 2025 · similarity 0.88)
- Partner Modelling Emerges in Recurrent Agents (But Only When It Matters) (NeurIPS 2025 · similarity 0.88)
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.
