Loading Events

« All Events

Digital Agents as Social Actors: From Theory to Agent-Supported Lifestyle Changes – PhD Defense

December 10 @ 13:00 - 17:00

Kaan Kilic defends his doctoral thesis “Digital agents as social actors: from theory to agent-supported lifestyle changes”, at Umeå University on December 10.

Abstract:

Most digital interventions for behavior change to improve health employ methods to support individuals such as providing research-based rationales, setting goals, sending reminders about goals, and tracking progress. However people differ, both between individuals and within the same individual over time, in how much support or challenge they want or need, and existing frameworks offer limited ways to model this variation or to act on it. Person-tailored support with artificial intelligence (AI) also brings questions on privacy, security, motivation, and how the AI-based support is perceived.

This thesis investigates how a digital companion can both support and challenge a person in lifestyle change while staying aligned with that person’s motives, readiness, and shifting preferences. Activity Theory and Argumentation Theory form the theoretical framework to develop the companion’s underlying models: agent roles, dialogue protocol, and two novel argumentation schemes for support and challenge. Participatory design is the overarching design methodology engaging an interdisciplinary team of health and domain experts, end users, and the research team iteratively design, prototype and refine subsequent versions of agent-based coaching systems. Across nine empirical user studies, data were collected through questionnaires, participant ratings, semi-structured interviews, use of prototypes, and observations during multi-week onboarding periods. Questionnaire data were analyzed quantitatively, and the interview and observation data qualitatively, through thematic analysis.

This thesis makes four contributions. First, an activity-theoretical reframing distinguishing exogenous (healthcare driven) from endogenous (person driven) behavior change, moving beyond persuasion and adherence to an intervention. Second, a formal model of supportive and challenging dialogue built on two novel argumentation schemes Argument from Position to Support and Argument from Position to Create Tension, each specified with critical questions and a dialogue protocol, which structures the grounds upon which an agent can act in a supportive or challenging manner. Third, a person-tailored, multi-domain coaching system built through participatory design, together with embedded clustered behavior change techniques. The approach is extended to a swarm of collaborating agents and single agents, whose role the person can choose or write, that supports recovery activities to manage everyday stress.
Finally, empirically grounded design implications for onboarding and role personalization.
Together, these provide a theoretical foundation, empirical insights, and a functional system of AI agents that tailor support and challenge to the individual; further identifying multi-agent coordination as a viable path toward personalized coaching for improved health through lifestyle change.

To summarize, the thesis contributes a deepened understanding of how an AI-based agent and a team of agents can be adapted to people’s individual and varying needs.

See full thesis

Supervisors

Helena Lindgren, Professor
Victor Kaptelinin, Professor

Opponent:

Prof. Judith Masthoff

Details

Venue

  • SAM A:230