My research asks how AI agents and their human teammates build models of each other: how an AI can infer and act on what its teammates want, how people make sense of an AI teammate's behavior, and what happens when an AI has to reason about how it is being read. Most of this work uses large-scale online experiments in custom multiplayer environments, combined with computational modeling.

Dissertation: Designing AI Agents for Human Group Coordination

• Prosocial human-AI interaction (with Honda Research Institute). In a spatial token-collection game, three studies build up a picture of when helping spreads. People reciprocate directly when a robot helps them. That reciprocity also extends upstream, to other humans and other robots they weren't directly helped by. But merely watching a robot help someone else usually isn't enough, because in a busy environment people often don't notice it happened at all. A companion POMDP framework treats a human's prosociality as a latent state a robot can infer and shape.
• Timing as a social signal in human-AI teams. In a multiplayer Rush Hour puzzle where teams vote on moves under time pressure, an AI teammate's timing (voting first vs. last) changes how humans vote, hesitate, and follow. AI teammates make teams faster but are rated worse than human teammates, and their presence makes humans rate each other worse too. Current work replaces fixed timing with an adaptive, belief-based agent that infers each teammate's preferences and decides when to act.

Other work

• Trust repair in shared-control driving (Toyota Research Institute). Online experiment (N = 600) testing 8 trust-repair strategies across 3 shared-control scenarios after an AI driving error. Neutral AI expression and increased user control produced the largest trust recovery. IEEE ITSC 2025, invited session; two patent applications.
• LLM calibration (with Mark Steyvers and Padhraic Smyth). What language models know versus what people think they know, and how confidence is communicated. Nature Machine Intelligence, 2025.
• RAG-grounded AI avatars for learning (UCI Beall Applied Innovation). User studies comparing a persona-based AI avatar against a generic chatbot and static text on engagement and intent to continue, plus a Wizard-of-Oz study on whether users notice out-of-character responses.
• Body ownership in VR (University of Copenhagen, with Kasper Hornbæk). Eye-tracking study (N = 40) with Unity avatars, combined with micro-phenomenological interviews, on how gaze patterns shape embodiment. HCII 2025.

Publications

• Hu, X., Zahedi, Z., Akash, K., & Steyvers, M. (2026). How real-time signaling improves ad hoc collaboration. Proceedings of the Annual Meeting of the Cognitive Science Society.
• Zahedi, Z., Hu, X., Mehrotra, S., Steyvers, M., & Akash, K. (2026). Strategic shaping of human prosociality: A latent-state POMDP framework. IEEE Robotics and Automation Letters, 11(4), 4729–4736.
• Steyvers, M., Tejeda, H., Kumar, A., Belem, C., Karny, S., Hu, X., Mayer, L. W., & Smyth, P. (2025). What large language models know and what people think they know. Nature Machine Intelligence, 7(2), 221–231.
• Hu, X., Mehrotra, S., Misu, T., Akash, K., & Steyvers, M. (2025). Toward promoting prosocial interactions between humans with autonomous agents. ACM Transactions on Human-Robot Interaction, 15(1).
• Hu, X., Mehrotra, S., Zahedi, Z., Misu, T., Akash, K., & Steyvers, M. (2025). Does observing helping robots promote prosociality? Challenges of learning from observation in spatial environments. Proceedings of the ACM Collective Intelligence Conference, 159–168.
• Hu, X., Dees, L., Gopinath, D., Silva, A., DeCastro, J., Rosman, G., & Sumner, E. (2025). Enhancing trust repair in driving: The role of AI emotional expression and perceived control. IEEE International Conference on Intelligent Transportation Systems (ITSC).
• Hu, X., Dalsgaard, T. S., & Hornbæk, K. (2025). "A false reality"? A micro-phenomenology of avatars in VR. International Conference on Human-Computer Interaction (HCII), 77–96.
• Hu, X., Akash, K., Mehrotra, S., Misu, T., & Steyvers, M. (2024). Prosocial acts towards AI shaped by reciprocation and awareness. Proceedings of the Annual Meeting of the Cognitive Science Society, 46, 2270–2277.

Patents

• Sumner, E. S., Rosman, G., Hu, X., DeCastro, J. A., Silva, A. M., Gopinath, D. E., et al. (2026). Building the trust of a vehicle occupant in an automated vehicle function. U.S. Patent Application 18/929,848.
• Sumner, E. S., DeCastro, J. A., Rosman, G., Gopinath, D. E., Silva, A. M., Balch, T. M., et al. (2026). Systems and methods for predicting dissonance during shared control of a vehicle. U.S. Patent Application 18/923,840.