“If we are good friends, AI doesn't spy so much”: Children’s knowledge and misconceptions of AI safety

Progress in Artificial Intelligence (AI) is transforming learning landscapes. However, children largely lack the necessary knowledge and skills to interact safely with AI systems. To empower children to interact safely with such systems, we must first establish a baseline of their current AI knowledge. This paper presents a mixed-methods study, using surveys (n = 71) and focus groups (n = 36) to systematically investigate 10-16-year-olds’ understanding of AI. Quantitative results suggest that children’s AI knowledge increases with age. We did not find evidence that children’s attitudes towards AI safety change with age nor increased AI knowledge. Our thematic analysis contextualizes these findings by revealing children’s nuanced beliefs and misconceptions. This paper provides a comprehensive overview of children’s pre-existing AI knowledge and beliefs. Additionally, we offer concrete guidelines for developing AI literacy curricula to foster safe and responsible human–AI interaction.