Building Conversational Information Sources as Human-Centered Digital Persona Twins for Probation Officer Training

Large language models (LLMs) offered opportunities for experimenting with the ideas involve building digital twin of humans as they are able to generate human like responses. This doctoral research leverages personas to develop digital persona twins as LLM-enabled conversational agents that simulate human actors in probation interview encounters for criminal justice training. By framing officers as information seekers and personas as dynamic information sources, my work enables scalable, ethical training tools for studying information interactions. A key innovation extends digital twin theory from continuous tracking to temporal snapshot instantiation, capturing a human actor’s information state at a specific moment to enable privacy-preserving simulation. This approach builds on personas to create diverse, individualized digital twins, opening new applications in social computing and information interaction. This dissertation aims to contribute empirical insights into professional information seeking behavior and validates new approaches for designing AI-mediated information sources that support skill development in high-stakes professional contexts.