Either way you'd have to be watching what the kid was asking and making sure what they found was accurate and appropriate. I don't think I'd trust an AI trained on random crap found on the internet to teach my children unattended any more than I'd trust youtube's algorithm to feed them video suggestions all day without oversight.

Not only would I be risking my kids being exposed to things that were outright wrong or entirely inappropriate, but I'd also be missing the opportunity to discuss their questions and the answers with them, provide context, and learn new things myself in the process.