Please note: This PhD seminar will take place in MC 1056.
Xinyu Shi, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Jian Zhao
Current AI models have demonstrated remarkable progress on verifiable domains such as math and coding. However, open-ended creative problems such as visual content authoring — which are ill-defined, deeply contextual, and lack a single optimum — remain a significant challenge: humans struggle to express their intent fully, and AI models often fail to understand and align with that intent. As a result, there is substantial friction in making AI truly useful for creative professionals.
In this talk, I will present my research on defining new interaction paradigms among humans, AI, and creative artifacts to better support creative thinking and human-AI alignment. The conceptual key to my approach is designing interactive visual abstractions that visualize the key factors in creative artifacts and reveal the implicit structure behind the creative process. Through three projects in video color authoring, image design, and animation keyframing, I will demonstrate how such abstractions help users (1) better perceive the problem space, (2) intuitively express visual intent, and (3) structurally interpret AI's reasoning to ensure alignment.