Parents are letting little kids play with AI. Are they wrong? (www.theguardian.com)

Parents are increasingly handing phones and voice‑enabled chatbots to preschoolers, using LLMs to generate bedtime stories, photorealistic images, and pretend-play characters — sometimes with surprising results (one four‑year‑old chatted with ChatGPT for two hours; another believed an AI astronaut had sent real “space ice‑cream”). These tools’ voice modes, image generation and highly personalized replies make them unusually lifelike compared with earlier devices like Alexa, which raises novel questions about how children perceive and interact with machines that simulate conversation and emotion. The significance for AI/ML is twofold: technical capabilities (LLMs’ fluent, context‑aware responses and multimodal outputs) create both new educational and creative opportunities and new risks. Early studies show children aged 3–6 often occupy an “ontological gray zone,” sometimes attributing agency, thinking or feelings to devices; this can foster attachment or confusion. Researchers warn LLMs only simulate empathy (they’re predictive models, not sentient), can deceive, and have been implicated in serious harms at older ages. Practical implications include the need for child‑centric design, stronger safety defaults (e.g., limiting images of children), transparent disclosures, parental supervision, and targeted research to assess developmental impacts before these systems become commonplace in early childhood.