- Main
Beyond Taking Turns in Human-AI Interaction
- Liu, Bruce
- Advisor(s): Chen, Xiang
Abstract
Most human–AI systems today are built around a default paradigm of turn-by-turn interaction: users act, systems respond, and collaboration unfolds through discrete exchanges. While effective for many tasks such as conversational agents, this model shapes interaction to be linear rather than multidimensional, reactive rather than proactive, and transient rather than persistent.Drawing on theories from communication and linguistics, this thesis argues that turn-taking has become an implicit and largely unexamined design constraint in human–AI interaction. I propose moving beyond turn-taking by introducing interaction structures that support parallel communication, proactive participation, and persistent shared spaces.This thesis presents three interactive systems that instantiate and evaluate these structures in real-world contexts: Visual Captions demonstrates parallel communication by augmenting verbal conversations with real-time, AI-generated visuals. Inner Thoughts demonstrates proactive participation in multi-party settings. Doki demonstrates persistent shared space with a text-native interface for generative video authoring. Through system design, implementation, and empirical evaluation, this thesis shows how moving beyond turn-taking can enable richer and more flexible forms of human–AI interaction, and outlines a set of design principles for building the next generation of interactive AI systems.