| KEYWORD |
WebAgents: Designing Controllable UI Agents
keywords ARTIFICIAL INTELLIGENCE, HUMAN-COMPUTER INTERACTION, INTERACTION TECHNIQUES, TRUSTABILITY, USER INTERFACE AGENTS, VISUALIZATION
Reference persons LUIGI DE RUSSIS
External reference persons Tommaso Calò
Research Groups DAUIN - GR-10 - Intelligent and Interactive Systems - e-LITE
Thesis type EXPERIMENTAL, RESEARCH
Description The potential of deep learning to revolutionize user interface (UI) agents is immense, promising to automate and personalize user interactions by learning from vast datasets of goal-oriented behaviors. However, creating UI agents that users can control, trust, and understand remains a significant challenge. This thesis sets out to address this by developing and assessing interaction techniques and interfaces for pre-trained WebUI agents, emphasizing their roles as "teachers," "companions," and "students." Through these lenses, we will explore ways to enhance user control, trust, and comprehension of UI agents, underpinned by rigorous user studies. This research aims to establish a set of principles for designing UI agents that align with user-centric values, fostering their reliable and beneficial incorporation into everyday digital experiences.
The thesis will aim to:
1. Review and analyze current WebUI agents, focusing on design, functionality, and interaction methods.
2. Develop new interaction techniques that enhance the roles of UI agents as:
- Teachers: Helping users perform new tasks on existing interfaces.
- Companions: Collaborating with users on interface tasks.
- Students: Learning to autonomously perform routines or tasks on an interface as taught by users.
3. Evaluate these interaction techniques through user studies to gauge their impact on user control, trust, and visualization.
4. Formulate guidelines for developing controllable and trustworthy UI agents, contributing valuable insights to the HCI and AI fields.
If satisfactory, the result of the thesis will be released as an open-source project.
Required skills Good programming skills are required.
Python and JavaScript knowledge is required.
Deadline 11/03/2025
PROPONI LA TUA CANDIDATURA