Position: AI Assistance Should Support Children’s Development and Mental Health in Navigating Uncertainty

Children ask AI assistants to check their answers, explain things, and settle their worries. They do so while their capacity to notice what they do not know is still developing. Every AI reply makes a choice. It can find out what the child is unsure about, or it can decide for the child and do the thinking itself. An assistant that always decides removes the child's opportunity to stay with the unknown and work through it. Repeated, this may feed intolerance of uncertainty, a trait associated with anxiety in children. We argue that AI assistance for children should treat the developing capacity to navigate uncertainty as a design objective alongside immediate usefulness, in everyday conversation and not only in educational tools. We make the position concrete with a framework that analyzes AI responses to a child's questions that carry uncertainty. An AI assistant can either ask the child what they mean or interpret the message itself. It then chooses the type, amount, and format of support to provide. To illustrate the framework, we ask three things of any AI reply: whether it merely answers or also explains the child's error, who does the thinking, and whether it checks what the child understands. We collected 60 responses from the current default model for free ChatGPT users under direct, hint, and Socratic instructions. We found that no reply asked the child what they were unsure about. Who did the thinking was set by our instruction, not by the child's message. Every reply assumed a reading of the child, and such assumptions may close the opportunity to navigate uncertainty. Supporting that capacity begins with asking what the child is unsure about.