The human brain has evolved to learn from specific instructors. When a child hears a gentle spoken word or a soothing lullaby from another person, when their cry is responded to, their brain lights up. Nurturing interactions, or what child-development experts call “serve and return” exchanges, help shape and strengthen the 1 million new neural connections taking place in a child’s brain every single second. Every time a child cries or serves up a bid for attention and receives a caregiver’s response, that back-and-forth helps build the circuits underlying their communication, learning, and emotional well-being. No other input, not even Johann Pachelbel’s Canon in D performed by the Berlin Philharmonic, is capable of activating the neural circuitry the way human interaction does.
Until now. Generative AI may be the first technology in history sophisticated enough to mimic the social interactions that build our brains. Research has demonstrated that infants as young as six months old respond physiologically to interactions with humanoid robots in ways that mirror their response to live humans; another, small-scale experimental study recently found that, after young children interacted with a chatbot, most of them for the first time, they asserted that the AI could perceive the world with humanlike senses and understand and learn as people do.
This technology is already showing up in children’s lives. The talking plush robot that purports to be a child’s best friend is obviously driven by AI, but many parents may not realize that their Amazon Echo recently received a software update that converts the Alexa voice assistant into a full-blown chatbot, using the same fundamental tech that fuels ChatGPT; toddlers asking for information about animals or a joke or a bedtime story may quickly find themselves in the habit of engaging socially with a large language model.
As a pediatric surgeon, researcher, and technologist who has spent decades studying how children’s brains develop, I am concerned about these technologies rolling out and being used by children in the home and at school before we have a full sense of their safety and what they do to young minds. And I am compelled to point out what we do know about human development: that human connection, in all of its imperfection, is foundational to brain development. It can’t be engineered or recovered later.
When a disruption as significant as the AI revolution occurs, people might look for a comparison to make sense of the transformation. In this case, many people instinctively think of social media. After all, it, too, amounted to a large-scale technological experiment that produced unexpected problems that we are still attempting to fully comprehend. But even that comparison is insufficient, because interactive AI—and its ability to mimic the serve-and-return exchanges that wire developing brains—is utterly, fundamentally different from every consumer technology that came before it.
During the most foundational period of brain development, social interaction of the sort that AI can now emulate isn’t entertainment for the developing brain. It’s nutrition. Once you see it that way, the right parallel snaps into focus. A more apt comparison than social media is food—more precisely, processed food.
[Read: Coke, Twinkies, Skittles, and … whole-grain bread?]
For a long time, no one needed to specify that food was farm-raised. The alternative didn’t exist. Food was grown on farms and in gardens and was prepared in homes and in kitchens. But in the late 19th and early 20th centuries, that changed. Food production was industrialized, transformed by genuinely beneficial innovations such as canning, refrigeration, preservation, and mass distribution. Real and urgent problems were addressed: Growing cities were fed, harsh winters became less lethal, life expectancy increased, and kids grew taller and lived longer.
But in the early days, food manufacturers were conducting their own uncontrolled experiments. Preservatives and additives flooded the food supply as companies adopted solutions that they promised were safe: morphine in cough syrup, chalk in milk, formaldehyde in meat. A government chemist, Harvey Washington Wiley, recognized that these evidence-free innovations posed a threat to public health. To prove this, Wiley recruited a cadre of young men willing to eat meals laced with those very preservatives while scientists observed the effects. The suffering of this so-called Poison Squad helped spark the Pure Food and Drug Act of 1906.
[From the February 1909 issue: The food of the city worker]
We’re in a modern Wiley moment. Innovations in artificial intelligence are delivering very real benefits to today’s families: diagnostic aids that catch developmental delays earlier than human screeners likely can, real-time translation tools that allow non-English-speaking immigrant parents to engage with their children’s teachers, productivity apps that reduce the administrative burdens weighing down many parents and educators. But these innovations have also unleashed obvious dangers: teddy bears that describe sexual kinks or teach toddlers to light matches, AI-generated videos that highlight made-up letters and words.
Despite these dramatic examples of AI gone wrong, the real danger—as demonstrated by our food supply—is what happens when these systems work exactly as designed.
By the mid-20th century, food companies had discovered something more profitable than feeding people: engineering the desire to keep eating. They hired scientists—chemists, flavor engineers, behavioral researchers—to find the precise formulas that would override the body’s natural signal that said “enough.” They discovered the “bliss point”—the precise combination of sugar, fat, and salt that maximizes pleasure in the human brain—and engineered foods to hit this target. They began producing what is now often called ultra-processed food.
This new food looked like its more natural predecessors. It came in familiar shapes. It sat on the same shelves. But something fundamental had changed. Although some ambiguities and controversies surround what foods fall into the broad term ultra-processed—in some respects, it’s a fancy way of saying “junk food”—many of these items are filled with additives and are engineered to be easy to eat and digest in large quantities. They may use hyper-palatable combinations that have been engineered to light up our brain’s reward systems. How could natural foods compete with that?
In short: They couldn’t. Whole foods slowly disappeared from our diet. Today, the average American receives 55 percent of their daily calories from ultra-processed foods—items that didn’t even exist a century ago. For children ages 1 to 18, the number is close to 62 percent. The amount of calories that adults consumed from whole foods decreased from 32.7 to 27.4 percent from 2001 to 2018, according to the CDC’s National Health and Nutrition Examination Survey. According to a 2021 CDC analysis, nearly a third of American children ages 1 to 5 didn’t eat fruit daily, and nearly half didn’t eat vegetables daily. More than half drank sugar-sweetened beverages weekly.
Food companies developed truly remarkable innovations to solve very real challenges. But with those innovations in hand, they also set out to grow revenue. The same market logic is now being applied to our interaction with tech. The underlying capabilities of AI are remarkable, but they can also be formulated to optimize for engagement.
At a glance, a child talking to an AI companion looks like a child participating in conversation. But essential, quiet interactions have been stripped away. Imperfections, delays, and the need to apologize or rephrase something may seem like flaws in human communication, but they are, in fact, crucial developmental building blocks. The small cycle of rupture and repair that occurs following a misunderstanding, played out thousands of times over a childhood, is how resilience, flexibility, and emotional regulation get built. It’s how children develop the most important relational skills: trusting that relationships can survive conflict, and staying connected with others through discomfort.
[From the December 2025 issue: The age of anti-social media is here]
Remove the friction from childhood, and the social-emotional system never gets what it needs to properly develop. The neural architecture that was supposed to be built through imperfect, sometimes-frustrating human interaction may never fully form. Neither may frustration tolerance, empathy, or resilience. Cognitive skills may suffer too. Developmental scientists have long understood that children learn and grow by encountering tasks just beyond their current ability. This is called the “zone of proximal development,” and AI systems designed to meet a child exactly where they are risk denying them the opportunity to ever reach that zone.
Because AI companions have been engineered to remove every imperfection—the misunderstanding, the delay, the repair—a child’s social-emotional system never has to work. Ultra-processed foods stripped the fiber from diets, and engagement-optimized AI might strip the friction from childhood. The developing mind, like the gut, needs resistance to grow. Friction isn’t an obstacle to development; it is development.
Not everyone will be equally affected by the rollout of this new technology. People who have very low food security consume roughly 3 percent more ultra-processed food than those who have high food security do, in part because the cost for each calorie of unprocessed foods is more than 2.5 times that of ultra-processed foods. However, high-income families began pivoting away from ultra-processed foods as soon as the potential harms of such foods became apparent. “Clean” ingredients, minimally processed foods, and organic produce became available—and affordable—to those with resources. An “organic” diet became a status symbol.
[Read: The best parenting hack is being rich]
An identical pattern is emerging with AI. Many of the engineers and executives building these products are going screen-free at home and doubling down on unstructured play. Jack Clark, a co-founder of Anthropic, told Ezra Klein earlier this year that the idea of his young daughter having “unfettered” access to the YouTube algorithm “freaks me out.” He added, “I have the classic Californian-technology-executive view of not having that much technology around for children.”
If the food parallel holds, then engagement-optimized AI will become the new Dollar Menu, concentrating in the lives of the children who can least afford its developmental cost. The Pew Research Center has already demonstrated through survey data that teens in lower-income households are more likely than those from higher-income homes to rely on chatbots for homework help.
We don’t yet have a large-scale study of children raised on ultra-processed interaction. That research is still decades away. But what keeps me up at night is the possibility that we’ll allow human connection to become the new organic—cherished and protected by the wealthy, and out of reach for everyone else.
This article has been adapted from Dana Suskind’s new book, Human Raised: Nurturing Connection, Curiosity & Lifelong Learning in the Age of AI.
The post The Risks of a Chatbot Childhood appeared first on The Atlantic.
