More about HKUST
Human Agency in the Age of AI: Towards Human-AI Co-Creation with High Control and High Automation
The Hong Kong University of Science and Technology
Department of Computer Science and Engineering
PhD Thesis Defence
Title: "Human Agency in the Age of AI: Towards Human-AI Co-Creation with High
Control and High Automation"
By
Mr. Leixian SHEN
Abstract:
Before the advent of Generative AI (GenAI), digital content creation largely
depended on manual Authoring tools (e.g., Adobe Illustrator), where humans
acted as operators who exercised precise control through labor-intensive
operations. Today, multimodal foundation models can generate images, videos,
animations, and other media from natural language prompts, dramatically
lowering the barrier to creation. Yet this shift introduces a central
tension: as AI becomes more powerful and autonomous, human agency—the
meaningful ownership and control over the creative process—tends to diminish.
Users risk being downgraded from creative authors to passive spectators.
To address this issue, this thesis asks: How can we scale AI automation
without sacrificing human agency? I argue that high automation and high
control are not inherently contradictory goals. Rather, they can be jointly
achieved when interaction systems provide shared, manipulable representations
through which human intent can be expressed, inspected, revised, and carried
forward. This thesis develops this argument through a progression from
concrete system explorations to theoretical abstraction, expanding the scope
and temporality of agency across the human-AI co-creation lifecycle.
The research unfolds across four progressive works. First, Data Player
explores the Automation stage by automatically generating data videos with
narration-animation interplay from static visualizations and descriptive
text. In this stage, the human becomes a consumer of AI-generated media: the
system achieves high efficiency, but the user's agency is compressed into
providing initial inputs and accepting or rejecting final outputs. This
exposes the limitation of full automation without fine-grained human input.
Second, Data Playwright advances to the Alignment stage by introducing
annotated narration, a paradigm that allows users to embed natural language
authoring commands directly within narrative text. Here, the human role
shifts from passive consumer to active director: users can specify what
should happen, when it should happen, and which visual elements should be
affected, while the AI handles interpretation, synchronization, and
rendering. Third, Live Artifacts extends agency into the Adaptation stage by
introducing persistent generative media whose specifications remain embedded
in visual layers and can be re-evaluated across time, context, and
modalities. In this stage, the human becomes an architect of living media:
rather than authoring a single static output, the creator defines what
remains stable, what changes, and how generative behaviors propagate across
the artifact's lifecycle. Finally, the Interaction-Augmented Instruction
model provides a theoretical Abstraction of these systems by formalizing how
natural language prompts and graphical interactions jointly construct
executable instructions for GenAI. At this meta-design level, the human
becomes an interaction designer: a designer of the representational
structures, interaction flows, and instruction paradigms through which future
users can communicate intent to AI systems with both flexibility and
precision.
Together, these contributions show that human agency in the age of AI is not
simply preserved by keeping humans in the loop. It must be deliberately
designed into the representations, interaction structures, and media
infrastructures through which humans and AI create together. Ultimately, this
thesis demonstrates that human agency is not what we save from machines; it
is what we systematically evolve and expand alongside them.
Date: Thursday, 25 June 2026
Time: 3:00pm - 5:00pm
Venue: Room 3494
Lifts 25/26
Chairman:
Committee Members: Prof. Huamin QU (Supervisor)
Prof. Pedro SANDER
Dr. Arpit NARECHANIA
Prof. Celine Yunya SONG (EMIA)
Prof. Yuanchun SHI (Tsinghua University)