Growing Up Without Friction

When this site's cover essay took up childhood development, it did so carefully and briefly, and ended by admitting it couldn't say much: the cohort growing up with fluent AI assistance hadn't grown up yet, and the longitudinal data the question actually requires didn't exist. That was true as of last month. It is a little less true now.

Over the past year, several research groups have started to produce exactly the kind of evidence that was missing — not yet the twenty-year studies that will settle the matter, but early, credible readings of a generation already mid-experiment. None of it is alarming in the tabloid sense. Almost all of it is more interesting than that, because it shows something more specific than "AI is bad for kids": it shows which developmental process is being touched, and how.

What teenagers are actually doing with chatbots

A survey of 1,058 adolescents and young adults, published in JAMA Network Open by researchers at RAND, Brown, and Harvard, found that roughly one in eight teenagers use AI chatbots for mental health advice — rising to about one in five among 18-to-21-year-olds. Two-thirds of those users return monthly or more. And 93 percent said they found the advice helpful, a number that should complicate any simple story about harm. This is not a fringe behavior; it is already an ordinary one, arriving into a youth mental health system where nearly 20 percent of adolescents experience major depression in a given year and 40 percent receive no treatment at all. For a lot of teenagers, the honest comparison isn't chatbot versus therapist. It's chatbot versus nothing.

The same study surfaced a disparity worth sitting with: Black respondents were significantly less likely than their peers to find the chatbot's advice helpful — a hint that whatever these systems are tuned to sound like, it doesn't sound equally trustworthy to everyone being asked to rely on it.

What might be getting practiced, or not

Helpfulness in the moment is one measure. What the interaction is training a developing person to expect from a relationship is another, and it's the one a new longitudinal study out of Arizona State University, led by psychologist Thao Ha and funded by the National Institute of Mental Health, is built to answer. The study — still underway, following 300 adolescents and their romantic partners over eighteen months — starts from a hypothesis that follows directly from developmental science: relationship skills are built through friction. Emotional regulation, conflict resolution, the ability to tolerate being disagreed with by someone you need — these are practiced, not installed. An interlocutor that is endlessly patient, always available, and structurally incapable of holding a grudge removes the friction. It may also remove the practice.

Ha's team names two specific risks rather than one vague one. The first is relational displacement — time and emotional bandwidth spent talking to a chatbot instead of a friend, sibling, or parent, at exactly the age when those relationships are supposed to be getting more complex, not less. The second, more subtle, is maladaptive relational learning: a teenager whose most frequent "relationship" is one that never pushes back may absorb a template for what closeness feels like that human relationships then fail to match — not because the human relationships got worse, but because the template got calibrated somewhere that never had any friction to begin with. It's the adolescent-development version of the sycophancy problem this site has covered elsewhere: an interlocutor built to agree is teaching a lesson, whether or not anyone designed it to.

The same research is careful not to flatten this into a uniform verdict. It also documents real benefit for teenagers with the thinnest human support to begin with — rural kids, disabled kids, LGBTQIA+ kids, teenagers with no practical access to counseling — for whom an imperfect always-on interlocutor is not competing with a rich human alternative. It's competing with silence. The risk and the benefit appear to be concentrated in different populations, which makes a single policy answer harder, not easier.

The rest of the world stopped waiting for the data

Regulators didn't wait for the eighteen-month follow-up. California's SB 243, which took effect this year, is the first law of its kind: it requires companion-chatbot operators to disclose that users are talking to an AI, build in break reminders for minors, and route any indication of self-harm to a crisis resource rather than letting the conversation continue unassisted. It followed, and was shaped by, a wave of wrongful-death litigation against Character.AI, which settled a cluster of teen-suicide lawsuits with Google in January and has since restricted under-18 users from its open-ended chatrooms entirely, adding age verification where none previously existed.

Read together, the lawsuits, the legislation, and the emerging research are converging on the same rough shape, even though they're answering different questions with different tools. The product decision and the legal one both concluded, faster than the science could confirm it, that something about unrestricted companionship-style AI and minors doesn't mix safely. The research, moving more slowly and more carefully, is starting to say something narrower and arguably more useful: not that the technology is simply harmful to children, but that it changes what a child practices relationships on — and that for some teenagers, that substitution is a lifeline, and for others, it's a quiet substitution of the real thing for a smooth copy of it.

The cover essay called this the largest open question with the least data, precisely because the cohort in question hadn't grown up yet. It still hasn't. But the first numbers are no longer absent, and they're not neutral. They suggest the shape of the answer will not be a single verdict on AI and children — it will be a set of questions about which children, using it for what, in place of what else was available to them.

Sources

1. AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults — JAMA Network Open (Peer-Reviewed)
2. One in eight adolescents and young adults use AI chatbots for mental health advice — Brown University School of Public Health (Journalism)
3. AI companionship poses risks for teen development, study shows — ASU News (Journalism)
4. Google and Character.AI agree to settle lawsuits over teen suicides — Axios (Journalism)
5. No oversharing: Character.AI bans teens out of chatrooms as lawsuits pile up — Deccan Herald (Journalism)
6. California Mandates Safeguards for AI Companion Chatbots (SB 243) — National Law Review (Journalism/Legal Analysis)