Welcome to the Social Reinforcement Learning lab website!
Social learning helps humans and animals learn complex behavior, rapidly adapt to new circumstances, and cooperate to achieve joint goals. Our lab is focused on whether Social Reinforcement Learning—algorithms that leverage social information in both human-AI and multi-agent interactions—can do the same thing for AI. Research areas include multi-agent reinforcement learning, learning from human feedback (RLHF), human-AI interaction and cooperation, social learning, modeling other agents’ goals and preferences, and emergent complexity.
Our lab was founded by Natasha Jaques in 2024 and is part of the Paul G. Allen School for Computer Science and Engineering at the University of Washington.
If you are a prospective Ph.D student or postdoc, please visit the contact page for more information on how to apply.
News
- Our paper on Modeling Others' Minds as Code received a Best Paper award at NeurIPS 2025 LAW Workshop!
- Our paper Chasing Moving Targets with Online Self-Play Reinforcement Learning for Safer Language Models won the Outstanding Paper Award at COLM 2025 AIA workshop!
- Our paper on Curiosity Reward for Personalized Multi-Turn Dialogue was accepted at NeurIPS and won the Madrona Prize at the Allen School’s Industry Affiliates Research Showcase.
- Released three exciting pre-prints on reinforcement learning fine-tuning of LLMs, with multi-agent RL (Self-RedTeam, SPIRAL) and multi-turn RL (CURIO).
- Our paper on Cross-Environment Cooperation was accepted as an oral paper (top 1%) at ICML 2025 and to CogSci 2025!
- Received a nomination for Best Paper at ICRA for our paper on the first AI-controlled robots to play table tennis in real time with human players.
- MARL for satellite assignment received an oral at AAAI (top 5%).
- Our paper on personalized reward learning for large language models, VPL, was accepted as a spotlight presentation at NeurIPS (top 2%) and was nominated for Best Paper at the Pluralistic Alignment workshop.
- Natasha taught a grad class on Social Reinforcement Learning.
- The SocialRL lab started in January 2024 at the University of Washington!