NormGym: Evaluating Normative Competence through Social Interaction
Binze Li, Zihuan Zhang, Seokhyun Baek, Gillian K. Hadfield, and Tianmin Shu. arXiv preprint, 2026.
[arXiv] | [Project Page]
The Normativity Lab
Johns Hopkins University
I’m a junior at Johns Hopkins University pursuing a B.S. in computer science. I do research at the Normativity Lab under Gillian Hadfield, studying how AI agents learn social norms and reason about what people expect of them.
At the Alliance AI Lab, I apply this research to human–robot interaction. I’m also an external collaborator with the Social AI Research Group, where I’ve contributed to Sorrel, an open-source framework for building multi-agent environments.
I’m broadly interested in AI alignment, especially what it would take for models to reason more like people. My current interests include mechanistic interpretability, neuroscience-inspired machine learning architectures, emergent cognition, and multi-agent reinforcement learning (MARL).
· Reading the Room was accepted to the Machine Ethics and Reasoning in AI (MERe AI) Workshop.
· Our NormGym preprint is out. It introduces a benchmark for how AI agents learn social norms through interaction.
· Reading the Room was accepted as a poster at the Johns Hopkins DSAI Fall Symposium.
· Nominated for the CRA Outstanding Undergraduate Researcher Award.
· I received the Pistritto Research Fellowship and joined the Alliance AI Lab for summer research.
Binze Li, Zihuan Zhang, Seokhyun Baek, Gillian K. Hadfield, and Tianmin Shu. arXiv preprint, 2026.
[arXiv] | [Project Page]
Andrea Wynn, Harsh Satija, Seokhyun Baek, Anqi Liu, Eric Nalisnick, and Gillian K. Hadfield. arXiv preprint, 2026.
Accepted to the MERe AI 2026 Workshop.
[arXiv]