Children increasingly interact with AI systems that simulate epistemic authority without possessing it.This paper develops "trust transfer" as a normative concept for understanding and protecting children'sepistemic development in AI-mediated environments. Drawing on the philosophy of trust (Baier, Jones,Hawley, McLeod), attachment theory (Bowlby, Fonagy), and developmental epistemology, I argue thatchildren's capacity to evaluate AI outputs depends on a prior developmental achievement: the acquisitionof epistemic self-trust through interaction with responsive human informants. Trust transfer—thedevelopmental process through which interpersonally grounded epistemic confidence becomes thefoundation for autonomous evaluation of non-human information sources—has three constitutiveproperties: human precedence, embodiment, and developmental timing. I show that trust transfer, whilenecessary, is insufficient for AI evaluation: a second component—understanding of LLM operatingprinciples—is required, creating a "dual requirement" for verification competence that has distinctdevelopmental timelines for each component. This dual-requirement framework generates specific designimplications: AI systems for children should be designed to support trust transfer rather than substitutefor it, and the appropriate degree of AI epistemic restriction should track the child's progress along bothdevelopmental dimensions. The analysis connects children's rights philosophy (evolving capacities, theright to an open future) with concrete developmental mechanisms, providing a normatively groundedframework for age-differentiated AI design.
Franny Philos Sophia (Tue,) studied this question.