Hao Zhu
he/him/他
朱昊
Postdoctoral
Stanford University
zhuhao@stanford.edu
AI, Cooperation, Language, CogSci, Robotics.
I take an interactionist approach. This means the priority of my research is to bring up agents in the realistic open-ended multi-agent embodied environments.
For examples of environments, check out Sotopia and WebArena. For examples of methods for training agents in these environments, check out Sotopia-π and NNetNav-live. Lately, EgoNormia benchmarked the preparedness of frontier models for embodied interactions with humans. AutoLibra allows agents to develop self-regulation metrics from human open-ended feedback.
Self-identified as a hedgehog , my research is scattered along three major dimensions: the behavior of agents, e.g. like a computer, a robot, or a simulated human; the cognition of agents, e.g. reasoning about humans, other agents, or time and space; and the learning of agents, e.g. through exploration, environment feedback, or human feedback.
My ultimate goal is to unravel the mysteries of human intelligence through building AI. My research not only takes inspiration from but also contributes to a better understanding of humans themselves. You will have the sense of this research taste if you read the first sentence of each of the above papers.
If you approach me, consider the following topics that I am mostly interested in:
I got asked this question a lot, so I decided to put it here. Most of my work aims to build algorithms that are general enough to be applied to a wide range of agents, including computer agents, embodied agents, and social agents. The reason why I work on a broad range of them is that I believe that the central challenges of all of them are common, and the only long-lasting solution to solve individual challenges is the one that can solve the common challenges.
I am always looking for talented students and collaborators from all backgrounds for both research and impactful open source projects. That being said, I invest at least 10 hours per week in each project I manage, so I am currently at full capacity. Feel free to shoot me an email if you are interested in my research, but please understand if I am not able to respond to every email.
Yes! I am always happy to talk to people. Please check out my calendar and book a slot . If you are at Stanford, I host an office hour every Friday at 9am in Gates 256. Feel free to drop by! Sorry, no online office hours at the moment.
Happy to give a talk! I have several keynotes covering my latest research which are ready to present. Please email me for more details. Current talk topics: “Training agents in the open social world: exploration and reinforcement”, “Building the common ground between humans and AI: norms and risks”.