Title: Designing Adaptive Responsible AI through Participatory Value Alignment 

Date: Monday, November 10th, 2025 

Time: 1:00 - 3:00 PM (EST) 

Location (in-person): TSRB 509        

Location (Zoom): https://gatech.zoom.us/j/96916027886 

Committee: Dr. Brian Magerko (chair), Dr. Noura Howell, Dr. Richmond Wong, Dr. Mark Riedl, Dr. Alireza Karduni 

Summary:  

This dissertation begins by asking: How do different stakeholders define 'responsible behavior' in professional practice, and what does it mean for an AI agent to embody that (or not)? I approach this question from a stakeholder-centric perspective that views responsible behavior as situated and shaped by diverse professional values that may exist in tension. 

Using academic advising as a research site, I explore how value articulation processes and artifacts support stakeholders in surfacing and reasoning about diverse values around responsible AI behavior. Drawing from literature in HCI, values in design, and STS, and through empirical work, I investigate: (1) what dimensions characterize the affordances and constraints that value articulation processes and artifacts create for stakeholder engagement with values; (2) how stakeholders engage in collaborative meaning-making and reasoning around responsible behavior through collaborative value elicitation activities; and (3) how explicit configuration of conversational agents supports stakeholders in exploring and reflecting on the operationalization of responsible behavior and what resists it in their professional practice. 

This research contributes: (1) the Values Articulation framework, providing nine dimensions for characterizing how processes and artifacts shape value articulation; (2) Values Mapping artifacts and process template for collaborative value articulation around responsible AI behavior; (3) a configurable conversational agent as a technology probe for exploring value operationalization; and (4) empirical documentation of relationships between activity characteristics and value articulation. These contributions have implications for researchers designing value elicitation activities and organizations developing configurable AI systems.