Answers are grounded in this paper and cite their sources.
As large language models (LLMs) increasingly shape children’s learning, fostering informed and critical engagement has become an urgent educational task. This study examines children’s AI literacy and trust in LLMs during a learning interaction. Thirty children aged 8-10 years participated in semi-structured interviews combined with an interaction task using a customized LLM. Children used the system competently, yet their conceptual understanding remained fragmented. At the same time, children placed high trust in the LLM’s ability and benevolence, while perceptions of integrity, particularly regarding data privacy, were more cautious. Their reasoning was frequently anthropomorphic, reinforced by social cues such as natural language use and emojis. The human analogy, however, was applied selectively. Children explained the system’s errors through human fallibility, but most considered algorithmic bias impossible, treating prejudice as a uniquely human trait. These findings suggest that AI literacy education should begin in elementary school and explicitly address questions of fairness, building on children’s intuition that AI learns from humans.
No takes yet. Share an insight, caveat, or question.
Rosada et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: