Our Take
The argument is sound: passive screens harm development; interactive AI could restore the serve-and-return loop that builds language and cognition. But no such product exists yet, and the piece offers no path to making business incentives align with developmental outcomes.
Why it matters
Millions of children are already using AI-powered apps optimized for engagement, not learning. Pediatricians have remained silent on AI while issuing detailed screen-time guidance—a gap that widens as these tools move faster than clinical evidence can follow.
Do this week
Pediatricians: document the AI tools your patients are using at well-child visits this month, so you can build an evidence agenda before the products solidify.
A pediatrician's case for AI with guardrails
Dua Hassan, a physician at Boston Children's Hospital, argues that AI designed for children should be built around developmental milestones—vocabulary, turn-taking, emotional labeling—and tested in randomized controlled trials measuring real outcomes, not engagement metrics. She contrasts today's AI products, optimized for screen time like their predecessors (autoplay, recommendation algorithms), with the model set by Sesame Street: educational media built by educators and developmental psychologists, tested with real children, and revised based on learning gains.
Hassan acknowledges the gap between current pediatric guidance and family reality. The American Academy of Pediatrics recommends limited screens and co-viewing, but parents cannot co-view during dinner preparation or after a 10-hour workday. Rather than counsel avoidance of a tool that is already embedded in children's lives, Hassan proposes designing AI that mimics the back-and-forth interaction—serve and return—that research shows drives language acquisition and emotional regulation.
She calls on technology companies to optimize for growth rather than engagement, on pediatricians to stop issuing guidance for an avoidable medium and start demanding clinical evidence, and on the medical field to apply the same rigor to AI for children that it applies to pediatric pharmaceuticals.
The passivity problem is real; the solution isn't yet
Early childhood passive screen exposure has been linked to language delays, attention problems, and social difficulties. A study in JAMA Pediatrics found higher screen time in early childhood associated with poorer developmental screening scores by age 2. Co-viewing with parental narration and interaction consistently improves outcomes—the medium itself is not the problem. Passivity is.
Hassan's insight is sound: an AI that waits for a response, asks prediction questions, and models emotional language could simulate serve-and-return at scale. But the critical gap remains unfilled. No pediatrician-vetted, developmentally optimized AI for children currently exists in the market. Products available today are built to the same engagement-maximization playbook as YouTube and TikTok, because that is the business model. Hassan identifies the design choice but does not address the economic force that sustains it: engagement is measurable and monetizable; developmental gain is not.
The silence from pediatrics is also structural. The American Academy of Pediatrics has detailed guidance on screen time and co-viewing. It has issued almost nothing on AI. This lag reflects the newness of the technology, but it also leaves a vacuum in which commercial incentives alone shape what children use.
What pediatricians can do now
Document. At your next 10 well-child visits, ask parents what AI tools their children are using and for how long. Note the products and the claimed purposes. This data will do two things: ground your own clinical observations in what families actually deploy, and begin to build the evidence base that currently does not exist.
Speak up in your professional networks. The AAP's silence on AI is not permanent. Pediatricians who have seen language delays or attention problems correlate with specific AI products have the standing to demand that the organization commission evidence reviews and issue guidance before the tools metastasize.
Do not counsel avoidance. If you tell a parent to avoid AI, you are issuing guidance you cannot realistically enforce and that ignores the real constraint: parents are not failing; they are adapting to impossible schedules. Offer specificity instead: "If you use [product], here is what to watch for in your child's language and social play."