Designing Human–AI Communication Systems for Distributed Collaboration

ABSTRACT

Human collaboration increasingly happens in distributed digital environments, where teams communicate through video calls, shared artifacts, and emerging GenAI tools. However, many nuances of real-world communication, such as spatial reference, shared attention, and unspoken vision, are difficult to express through existing interfaces. At the same time, AI is transforming how people create and communicate ideas, enabling new forms of collaboration between humans and AI systems. In this talk, I will present my research on designing AI systems that restore these missing layers to distributed collaboration. I will discuss approaches for enabling collaborators to share and manipulate physical objects in video-mediated interaction, supporting teams in externalizing and aligning shared unspoken intent, and providing practitioners with principled frameworks for authoring multi-party human–AI conversation structures with generative agents as designable artifacts. These components become building blocks of a new design paradigm for AI communication systems, one that takes seriously the physical, intentional, and structural dimensions of human communication, and that strives to bridge the theoretical foundations of human-computer interaction with the expressive capabilities of modern AI.

BIOGRAPHY

Erzhen Hu is a fifth-year PhD candidate in the Department of Computer Science at the University of Virginia. Erzhen’s research lies at the intersection of Human-Computer Interaction, Artificial Intelligence, and Collaborative and Social Computing, focusing on designing and evaluating human–AI communication systems that support collaborative and creative work. Her work has been published in top HCI/CS venues including CHI, CSCW, and UIST. She was one of the recipients of the 2024 Google PhD Fellowship, and has been selected to participate in the MIT EECS Rising Stars workshop in 2025. She has interned at Microsoft Research, Google, and Autodesk Research.