Description
The question of whether AI possesses agency has been debated among researchers. With the growing integration of AI agents across the data visualization pipeline, these agents are gaining increasing levels of autonomy. Based on the data, these agents can already help users identify potential areas of interest, suggest suitable visualizations, and even present findings to users through narrative summaries. Although end users are generally assumed to remain at the center of these computational and decision-making processes, empirical research on how users actually interact with AI tools and use these agents in the data visualization process remains under-explored. Depending on users’ understanding of the system, their domain and visualization expertise, and their level of trust in AI, they may either overly rely on the system or restrict its role to that of a simple rule-based assistant. To address these dynamics, we plan to design an AI-assisted interface that monitors user interactions during data exploration and provides feedback when users either grant excessive agency to the AI and drift from their initial analytical goals or overly constrain the AI, effectively reducing it to a rule-based system. User evaluation is also expected to illustrate how AI agency is negotiated and exercised throughout the data visualization process.
Tasks
Develop a user interface within the chatbot that gives real-time feedback to the users whenever they deviate from their analysis goals and passively consume AI information. Relevant work: https://dl.acm.org/doi/10.1145/3449287
Requirements
- Knowledge of English language (source code comments and final report should be in English)
- Knowledge of LLMs is required
- Knowledge of Python and web-based technology is necessary
Environment
The project should be implemented as a standalone application, desktop or web-based (to be discussed).