Design and Governance of Generative AI for Child Users: Cultural-Ethical Translation and Family Relationship Impact Assessment

Abstract

Generative artificial intelligence (GenAI) can hand a child an answer that is factually sound and emotionally soothing, yet the child may walk away from it less willing to listen at home. Judging the answer on its own terms would never catch this. This paper introduces family relationship impact assessment (FRIA), an approach for tracing how conversational GenAI becomes involved in the ties between children and the people they live with. The framework was built through conceptual analysis, an integrative reading of the literature, and mapping onto current governance guidance. Five dimensions informed by Confucian relational ethics—relationality, contextual proportionality, judgment, truthfulness, and motivational orientation—form its core, with children's rights bounding how each is interpreted. For every dimension we derive two questions: one asks which human function an assistant must not displace, the other asks what assistance it may legitimately offer. Together they yield ten preliminary indicators dealing with role boundaries, developmental expectations, reasoned judgment, warranted correction, and responsibility repair. Five thought experiments test how the indicators work, and the Govern, Map, Measure, and Manage functions of the NIST AI Risk Management Framework anchor their use. Organizations gain concrete questions for setting product roles, reviewing interactions, and deciding who is responsible for correction. FRIA is still a conceptual proposal: the indicators remain unvalidated, and nothing here shows how often relational harm occurs or whether the suggested controls would actually work.