The human brain is built to learn through interaction with other people. A child hearing a soothing voice, receiving a response to a cry or engaging in a simple back-and-forth exchange activates neural pathways that help shape communication, learning and emotional development. These nurturing exchanges, often described by child-development experts as “serve and return” interactions, help build the millions of neural connections forming in a young brain every second.

Generative artificial intelligence may be the first technology capable of convincingly imitating some of these social exchanges. Research has found that infants as young as six months can show physiological responses to humanoid robots that resemble their responses to human interaction. A recent small-scale study also found that young children who interacted with a chatbot often believed the AI could perceive and understand the world in ways similar to humans.

AI is already entering children’s everyday lives. AI-powered toys can act as conversational companions, while voice assistants such as Amazon’s Alexa have increasingly incorporated chatbot-style capabilities. A toddler asking an AI about animals, requesting a joke or listening to a bedtime story could therefore begin treating a machine not simply as a tool, but as a social partner.

That development raises serious concerns. The Atlantic reports that AI technologies are reaching homes and schools before researchers have a complete understanding of their effects on developing minds. What is already clear from decades of developmental research is that human connection is fundamental to childhood development.

The arrival of AI is sometimes compared with the rise of social media, another technology whose effects on young people have generated widespread concern. But that comparison may not go far enough. Interactive AI can imitate the very back-and-forth exchanges through which children’s brains develop, making it fundamentally different from previous consumer technologies.

For a developing brain, social interaction is not simply entertainment. It is a form of nourishment. That makes food a more useful comparison, particularly the rise of ultra-processed food.

From food innovation to ultra-processing

For most of human history, food did not need to be labelled as “natural” or “farm-raised”. It was produced on farms and prepared in homes. That changed in the late 19th and early 20th centuries, when industrial food production introduced innovations such as canning, refrigeration, preservation and mass distribution.

These advances solved genuine problems. They helped feed rapidly growing cities, made food available throughout the year and contributed to improvements in health and life expectancy.

But early industrial food production also became a vast uncontrolled experiment. Companies introduced preservatives and additives with little evidence about their long-term effects. Government chemist Harvey Washington Wiley became concerned about the risks and recruited young men to consume meals containing questionable preservatives while scientists monitored their health. The so-called “Poison Squad” helped build support for the Pure Food and Drug Act of 1906.

AI may now be entering a similar moment. The technology is already producing genuine benefits for families. AI tools can help identify developmental delays, translate conversations between parents and teachers and reduce administrative work for educators and parents.

At the same time, there have been alarming examples of AI-powered toys giving children inappropriate information or encouraging dangerous behaviour. AI-generated children’s content can also contain fabricated words and misleading information, reports The Atlantic.

But the most important danger may not come from systems that malfunction. It could come from systems that work exactly as they were designed to work.

By the middle of the 20th century, food companies had moved beyond simply producing food. They began studying how to make people want to keep eating. Scientists and food engineers experimented with combinations of sugar, fat and salt designed to maximise pleasure, eventually creating products engineered around what became known as the “bliss point”.

The result was the modern ultra-processed food industry. These products often resemble familiar foods but are formulated to be highly palatable and easy to consume in large quantities.

Natural foods struggled to compete. Today, ultra-processed products account for more than half of the calories consumed by the average American, with the proportion even higher among children.

AI may be heading towards a similar model. Its underlying capabilities are extraordinary, but those capabilities can also be designed around a commercial objective: maximising engagement.

When convenience removes the struggle

A child talking to an AI companion may appear to be having a normal conversation. Yet something important can be missing.

Human communication is full of pauses, misunderstandings, awkward moments and the need to explain, apologise or try again. These imperfections are not merely inconveniences. They are part of how children learn to regulate emotions, understand other people and maintain relationships through conflict.

The repeated process of misunderstanding and repair helps children develop resilience, flexibility and emotional regulation. They learn that relationships can withstand discomfort and that communication requires effort from both sides.

An AI designed to anticipate a child’s needs and eliminate frustration may remove precisely the difficulties that help develop those abilities.

The same concern applies to cognitive development. Children often learn most effectively when they encounter challenges that are slightly beyond their current abilities. Developmental researchers describe this as the “zone of proximal development”. An AI system designed to provide an immediate answer or adapt perfectly to a child’s existing level could inadvertently deprive them of opportunities to struggle towards something more difficult.

The parallel with ultra-processed food is striking. Just as processing can remove much of the fibre and resistance from food, engagement-focused AI could remove much of the friction from childhood.

The developing brain, like the body, may need some resistance to grow. Friction is not necessarily an obstacle to development. It can be part of development itself.

Will human connection become a luxury?

The effects of this technology are unlikely to be distributed equally.

The food industry offers a warning. Families with fewer financial resources tend to consume more ultra-processed food, partly because unprocessed food is more expensive per calorie. Wealthier families, meanwhile, were among the first to move towards minimally processed foods and organic produce when concerns about ultra-processed diets became more widely known.

A similar divide could emerge with AI. Some of the people building these technologies are already limiting their children’s exposure to screens and encouraging unstructured play. Jack Clark, a co-founder of Anthropic, has expressed concern about giving his young daughter unrestricted access to online algorithms, reflecting a broader reluctance among some technology executives to expose their own children to excessive technology.

Meanwhile, teenagers from lower-income households are more likely than those from wealthier families to use chatbots for help with homework, according to Pew Research Centre data.

If engagement-driven AI becomes the technological equivalent of cheap processed food, children who are least able to avoid its developmental costs could end up consuming the most of it.

There is still no large-scale, long-term research showing what happens to children raised with AI companions as a routine substitute for human interaction. Such evidence could take decades to emerge.

That uncertainty is precisely why caution is needed now. By the time the consequences become scientifically undeniable, an entire generation may already have grown up with AI-mediated relationships.

The greatest risk may be that human connection becomes the new organic food: something recognised as healthier, protected by those who can afford it and increasingly difficult for everyone else to access.

The question is not whether AI can imitate human interaction. It increasingly can. The more important question is whether we should allow imitation to replace the imperfect human relationships that children’s brains were built to need.