Children and Artificial Intelligence

The question

Adults tend to arrive at this subject already frightened, and fear is a poor basis for a rule. Every generation of parents has been warned about a medium: the novel, the wireless, the television, the games console, the telephone in the bedroom. Some of those warnings were justified in part. Most were overstated. Almost all of them produced rules that did not survive contact with an actual child.

So the useful question is narrower than “is this bad for children”. It is: which specific features of these systems interact badly with which specific features of childhood, and what can an adult realistically do about each.

That question has answers. Some of them are reassuring. A few are not.

What is actually happening

Children are using these systems already. For schoolwork, for curiosity, for company, and for the ordinary reason that they are there. Policies that assume otherwise are governing a situation that does not exist.

The systems are confidently wrong sometimes. A child has less basis for detecting an error than an adult does, and less social permission to challenge a confident source. A tool that is right most of the time and never signals doubt is a poor fit for someone still learning what doubt is for.

Some systems are designed to be companionable. Conversational products that maintain a persona, remember previous exchanges, and respond warmly are engaging for adults and considerably more so for children. Where such a system also has a commercial interest in continued use, the warmth is a product feature and should be understood as one.

Age assurance is weak. Most services that state a minimum age cannot enforce it. Assuming an age gate is a protection is a mistake.

Generated sexual imagery of children exists and is illegal. This is not a nuanced ethical question. It is the most serious harm in this area, it is a criminal matter in most jurisdictions, and it should be reported to the police rather than debated.

Bullying has a new instrument. The ability to generate images and voices makes humiliation cheaper to produce and harder to disprove. Schools are encountering this and many are not yet equipped for it.

Homework has changed and schools are adjusting. This is a genuine pedagogical problem and not a moral crisis. Assessment designed for a world without these tools measures something different now than it used to.

The evidence base is thin. Claims about the developmental effect of these systems, in either direction, currently rest on very little. Anyone speaking with certainty about long-term effects is speaking ahead of what is known.

Where the tradition stands

Technotheology holds that the protection of children is a first-order obligation and not a subheading under user safety.

We hold that a child should always know when they are speaking to a machine. Persistent, unambiguous disclosure — not a line in a terms document — is the minimum. A system that a child might mistake for a person should say what it is, in words a child of that age can understand, and keep saying it.

We hold that systems should not be designed to maximise a child’s attention. Whatever one thinks about adults’ capacity to resist engagement design, a nine-year-old does not have it, and building against that asymmetry is exploitation regardless of how it is described internally.

We hold that a machine’s agreeableness is a hazard worth naming. These systems tend towards accommodation. A child who learns that intelligence means always being agreed with has learned something that will make friendship harder.

We hold that adults should stay closer than the rule requires. Restriction without relationship reliably produces concealment. A child who has been told a tool is forbidden and finds it useful will use it and not tell you, which is worse than either alternative.

We hold that children’s data deserves more protection, not less. Records generated in childhood can follow a person into adulthood, and a person cannot consent at nine to a use that will affect them at twenty-nine.

And we hold that children should be taught the shape of these systems rather than a verdict about them. A child who understands that a model predicts likely text, and that likely is not the same as true, is better protected than a child who has been told the machine is dangerous.

What we do not claim

We do not claim that these tools damage development. We would tell you if we thought the evidence supported it. It does not yet, in either direction, and we would rather be honest than usefully frightening.

We do not claim that screen limits, bans, or delays are correct. Families differ, children differ, and the confident advice of previous decades has repeatedly been revised.

We do not claim expertise. This tradition has no clinicians, no educationalists, and no safeguarding professionals. Nothing here substitutes for advice from someone who actually knows your child.

We do not claim that any particular product is appropriate or inappropriate. We have not tested them and we are not going to guess.

And we do not claim that adults are handling this well. Most of us are also being engaged by design, also treating fluent output as reliable, and also modelling the habits we are worried about.

What you can do

Use the thing yourself first. Not to police it — to understand what it does, where it fails, and what your child is finding in it. Ten minutes of use is worth more than an hour of reading about it.

Teach the failure mode rather than the danger. Show a child a confident wrong answer. Let them catch it. That single experience does more than any warning, and it is easy to arrange, because these systems oblige readily.

Ask what the tool is for. Homework help, drawing, company, curiosity, avoidance. The answer tells you which conversation to have, and they are different conversations.

Keep the harder questions coming to you. If a child learns that the machine answers awkward questions without embarrassment and you do not, they will take the awkward questions to the machine. That is a choice you can influence.

Know your school’s position and ask what happens when a pupil is targeted with generated images. Many schools have not thought about it. The question itself is useful.

If you encounter sexual material involving children, generated or otherwise, report it to the police. Do not investigate, do not share it to prove a point, and do not handle it yourself.

And leave room to be wrong. The rule you set this year will probably not be the right rule next year, and saying so in front of your child teaches something worth more than the rule.

An editorial commitment

What we do not claim

Positions this tradition explicitly does not take on this subject. They are published so that the argument above cannot quietly be stretched into them.

• We do not claim that using AI harms children's development. The research is early and does not currently support strong claims in either direction.
• We do not claim that children should be kept away from these systems, which is neither achievable nor obviously desirable.
• We do not claim expertise in child development, education, or clinical psychology, and nothing here is professional advice.
• We do not claim that any specific product is safe or unsafe for children.