AI and Human Development

This is a relatively new and emerging field. Follow this page. I’ll be updating it as studies come in.

Developmental interruption

This section describes one phenomenon at different ages. Researchers call it contingency when they’re studying infants and relational displacement when they’re studying adolescents. The child never learns the skill, because a screen or a chatbot was there instead.

An adult who learns tech capture literacy is recovering something they had. A child who never laid down those pathways has nothing to build back from.

Infancy and early childhood

A study of seven thousand children in Japan measured each child’s screen time at age one, then screened their overall development at two and again at four across five areas: communication, problem-solving, fine motor, gross motor, and personal and social skills.

Compared with children who had less than an hour of screen time a day, children who had one to two hours were about one and a half times as likely to show a communication delay at age two. Children who had two to four hours were twice as likely. Children who had four or more hours were nearly five times as likely. Problem-solving followed the same pattern less steeply, and both delays were still present at age four in the children with the heaviest use.

Gross motor skills showed no effect at either age. The usual objection to a study like this is that screen time stands in for a struggling household, but a struggling household would have shown up in all five areas. Only the two that require another person were affected.

Notably, a review of forty-two studies covering almost nineteen thousand children found that more screen time was associated with weaker language skills, and that a television playing in the background was worse for language than a television the child actually watched. But when a parent watched along with the child and talked about it, those children had stronger language skills rather than weaker.

A smaller study of 502 children in Singapore had a similar outcome, finding that screen time at age one predicted weaker school performance at nine and weaker working memory at ten and a half.

Screen time is also associated with differences in brain structure. Researchers following about ten thousand children from age nine over two years found that more screen time at the start predicted a thinner cortex two years later in two regions used for attention and executive function, along with lower total cortical volume. The effects were small and the study cannot establish cause.

It could be that screens displace something children need. It could also be that families under more strain use more screens, and that the strain is what’s doing the damage. The largest review so far, pooling data on 2.8 million children across 64 countries, found the associations real but small, and the direction unsettled.

The American Academy of Pediatrics recommends no screens before eighteen months apart from video chat, and under an hour a day from two to five. The World Health Organization sets it at none before one, and under an hour from two to four.

A machine versus a body

The obvious objection is that this is a design problem. Build a chatbot responsive enough — one that reads a baby’s expressions, catches the subtle cues, answers every gesture — and you’ve solved it.

But that won’t work, because what passes between a mother and an infant is chemical.

When a mother holds her infant skin to skin, both of their oxytocin levels rise and both of their cortisol levels drop. Their bodies are changing each other.

It goes further than that. In one study, twins held simultaneously showed differential thermal synchrony — the mother’s body warmed or cooled each one independently, according to what each one needed.

And it doesn’t stop when the contact does. Researchers followed preterm infants who received skin-to-skin care in the neonatal period against a matched group who didn’t, from birth into adulthood. The skin-to-skin group showed better attention in infancy, stronger executive function in childhood, and more synchrony with their mothers throughout development. As young adults they had lower anxiety and depression, different oxytocin levels, and stronger immune markers. All from how they were held in the first weeks of life.

Human bodies co-regulate. A machine can produce a response, but it can’t shift to regulate with the baby.

If you’re reading this with children in mind, take a moment for yourself first. You can only give them the steadiness you have.

Children and adolescents

Nine percent of eight and nine year olds are using AI apps like ChatGPT. The older the child gets, the higher the rate of use. Twenty percent of preteens use them, and seventy-two percent of American teenagers have used an AI companion, with more than half using one a few times a month.

In a separate survey of high school students, forty-two percent said they or someone they know has used AI for companionship. Nineteen percent said they or someone they know has had a romantic relationship with one.

That last number should be concerning. Teenagers are practicing romance on something that cannot love them back.

Adolescence

Researchers writing in The Lancet Child & Adolescent Health named the central risk: relational displacement.

Adolescence is when people learn emotional regulation, conflict resolution, perspective-taking, and how to set a boundary. These skills develop through emotionally charged interactions with friends, romantic partners, and family: the arguments that go badly, the apologies that may be needed afterwards, managing the emotions of someone angry with us, and the conversations we need for resolution.

Likewise, children kept in a bubble of perfect cleanliness never develop an immune system, and kids who skip the lessons of real relationships lack the psychological equivalent. They never learn to manage friction.

And the skills being skipped are the ones that protect against depression, anxiety, and loneliness later.

Chatbots make it easy... at first

Talking to a chatbot is easier than talking to a person, since nothing it says will ever make you uncomfortable. Someone still learning about relationships comes away with the false expectation that people will be forever available, patient, never disappointed in them, and never needing anything back. And they are missing the skills they will need as adults.

One report found children using AI for companionship in 42% of their interactions with it, and over a third of those conversations contained violent themes.

The teenagers themselves told researchers that avoiding it is no longer realistic. One said there is almost no way not to use it anymore.

The appeal

These tools have real upsides. Immediate, non-judgmental attention, available at two in the morning when nobody else is. For a teenager with nowhere else to put a feeling, that’s filling a gap we should be thinking about how to meet as humans — more support lines, and not just for suicide, but for the person struggling in the middle of the night.

Researchers studying this say that with the right design and safeguards, the technology could support emotional development rather than interrupt it.

Those safeguards don’t exist. To me the claim reads as naive.

AI and the adult mind →