This thesis addresses the development of an AI-driven child avatar for realistic skills training in child interviewing. High-quality interviewing is crucial in cases of child abuse.
Two studies analyzing real-life interviews with young children, allegedly subjected to abuse, in child protection services and police investigations confirmed international findings: trained professionals rarely used recommended open-ended questions. Advances in AI have increased interest in avatar-based training tools. An evaluation of frontline professionals’ interaction with the mechanical LiveSim avatar for question training showed high user engagement and confirmed usability. A text-based AI child avatar was developed, and effects of training were examined. The study found improved use of open-ended questions, especially when paired with automated, individualized feedback. Feedback function aligned with expert assessments.
This dissertation supports further development of AI-driven conversation-generating avatars with integrated feedback for realistic and cost-effective interview training. Strengthening interview skills will enhance information quality from children in vulnerable situations.