Description: With the rise of multi-agent systems that are optimized to support humans with different tasks (solving logic, summarization, programming, ideation etc.), we face several challenges, such as loss of agency and transparency (i.e., who contributed to the solution), (over-)reliance or lack of trust in the AI assistants, because workers lack an explanation which agent contributed to the solution.

Goal of Thesis

The goal of this Bachelor/Master thesis (level will determine the scope) is to design virtual AI-agents that can communicate their function/capabilities through affordances, requiring little to no explanation. This thesis follows a Design Science Research approach and includes: 

1. Problem Awareness: Review of the related literature on (conversational) agent/ avatar / social cue design and identify typical tasks at work that involve AI agents.
2. Solution Design: Conduct a design workshop / create AI-based designs to produce affordance-based agent designs that are suitable for tasks at work.
3. Validation: Evaluate avatar designs in a larger quantitative (online) study.
4. Implement: Develop a prototype that supports a subset of the identified tasks + integrate the most promising avatar designs.
5. Evaluate: Conduct a user study to investigate whether the developed prototype improves users' trust, etc., in the system.

Requirements

• Strong interest in embodied AI and avatar design
• Familiarize yourself with the affordance-actualization theory, representation theory and role theory
• You have read the following papers: [1] [2]
• Experience with frontend development