The First AI-Native Generation: What Researchers Are Watching

In 2016, AI voice assistants entered millions of homes in the form of small cylinders that answered questions, played music, and set timers. Children born that year are now nine years old. They have never lived in a world where AI was absent, unusual, or experimental. They don’t know what it felt like to look something up and have to wait, or to need to remember a phone number, or to write a first draft without a tool that could revise it on request. This is not the same thing as being more capable with AI than older generations — it’s something more fundamental and less understood. It means their baseline relationship to intelligence, knowledge, and assistance is categorically different from any previous generation’s.

Researchers are paying attention. What they’re finding is incomplete, contested, and — in several areas — genuinely alarming. What they don’t know yet may be more important than what they do.

The Problem: We’re Running the Experiment in Real Time

When Marc Prensky coined “digital natives” in 2001, he argued that children who had grown up with computers thought differently from those who hadn’t. The claim was compelling as an intuition and weak as a research finding — subsequent work largely failed to confirm that growing up with computers produced systematic cognitive differences, and the “digital native” concept was largely retired by educational researchers by the mid-2010s.

The AI-native question is structured differently. Prensky was describing interface familiarity: kids who grew up clicking around computers were more comfortable with them. The AI-native question is about something more consequential: whether growing up with systems that can reason, remember, generate, advise, and respond like a person changes how children understand the nature of intelligence, knowledge, and social interaction — and whether those changes are beneficial, harmful, or simply different.

Jean Twenge’s longitudinal analysis of generational behavior, published in iGen (2017) and updated through 2023, traced the impact of smartphone adoption on adolescent mental health using decade-scale data. The correlations were significant and largely negative — increased screen time from smartphones tracked with rising rates of adolescent depression, anxiety, and loneliness, with effects pronounced among girls. Twenge’s critics have argued about causality and effect size, but the dataset is large and the trends are real.

Twenge’s framework is relevant to the AI-native question because it established a methodological precedent: the effects of a new technology on children’s psychology can be detected in population-level data, but often not until 7–10 years after adoption. We are currently 7–9 years into widespread AI assistant use in the home. The data Twenge-style research will generate about AI’s effects on this cohort does not exist yet. We are inside the experiment.

What the Research Actually Says

The Oxford Internet Institute’s 2025 report on children and AI is one of the most comprehensive attempts to survey the state of knowledge. Its conclusions are measured and worth quoting carefully: the report identifies three areas of “emerging evidence,” three areas of “significant uncertainty,” and explicitly flags the methodological challenges of studying a technology that is itself changing faster than research cycles can track.

Areas of emerging evidence from the OII report:

First, children form parasocial relationships with AI systems at rates researchers did not anticipate. Children who interact regularly with conversational AI — including voice assistants and AI companion apps — develop attachment behaviors toward those systems: they say “please” and “thank you,” they feel bad when the device is replaced, and in some cases they report emotional distress when interactions end abruptly. The OII researchers note this is not evidence of harm; it may simply reflect children applying social scripts to novel actors. But it suggests that the boundary between “tool” and “social other” is more permeable for children than for adults, and that children require explicit guidance about the nature of the AI they’re interacting with.

Second, early evidence suggests that AI assistance changes children’s tolerance for ambiguity and struggle. A small number of studies — including classroom observations by researchers at the University of Helsinki published in 2024 — found that students who used AI tools heavily in their learning showed lower persistence on difficult problems when AI assistance was removed. The effect was not uniform and the sample sizes were small, but the direction is concerning: AI that resolves difficulty immediately may be conditioning children not to develop tolerance for productive difficulty. This connects to Manu Kapur’s work on productive struggle, which has established that struggle is educationally necessary, not merely uncomfortable.

Third, Common Sense Media’s 2025 survey of AI use among children under 18 found that 41% of 13–17-year-olds use AI tools at least several times per week, compared to 8% of 8–12-year-olds. The age gradient reflects access and school policy as much as developmental readiness. Among teenagers, the most common uses are homework help, creative writing, and social navigation — using AI to draft texts, plan conversations, and process social situations. This last use case is the one researchers are watching most closely.

Sherry Turkle’s “Reclaiming Conversation” (2015) argued that smartphones and text communication were already reducing children’s capacity for unmediated, in-person conversation — the kind of conversation that requires reading social cues in real time, managing silence, and tolerating misunderstanding without an immediate edit or delete option. Turkle’s updated analysis, offered in academic presentations through 2024, extends this argument to AI: if children increasingly use AI to mediate their social communication — drafting messages, planning what to say, processing social situations through an AI before deciding how to respond — they may be reducing the practice time for exactly the social-emotional skills those interactions are supposed to develop. The concern is not theoretical. It’s a version of a question that learning scientists ask about any assistive technology: if the tool does the work, does the student develop the capacity?

UNICEF’s 2021 “Policy Guidance on AI for Children” identified five distinct categories of risk for children interacting with AI systems: privacy, safety, access equity, transparency about AI identity, and developmental impact. The developmental impact category was the least developed in 2021, reflecting the state of the research at the time. As of 2025, it remains the least developed — not because researchers have found no concerns, but because studying developmental impact requires longitudinal data on children who haven’t yet reached the ages at which the impacts would be measurable.

Research Area	Current Evidence Level	Direction of Findings	Years Until Clarity
AI and social skill development	Emerging, limited	Mixed / concerning in heavy-use cases	5–8 years
AI and tolerance for cognitive struggle	Emerging, limited	Concerning in high-dependence cases	3–5 years
AI parasocial attachment in young children	Emerging	Unexpected but uncertain in impact	3–5 years
AI and identity formation in adolescents	Very limited	Too early to assess	7–10 years
AI literacy as a protective factor	Emerging	Positive in early studies	2–4 years
Long-term cognitive effects (working memory, recall)	No longitudinal data	Unknown	10+ years

What to Actually Do

The honest position is that parents are navigating without a complete map. The research available is suggestive, not conclusive. What follows are actions that are defensible on the basis of what we know — not certainties, but reasonable responses to what the emerging evidence suggests.

Prioritize unmediated social experience

If Turkle’s concern about AI as a social mediator has any validity — and the early evidence suggests it might — the protective factor is simple: make sure your child spends substantial time in social situations where AI is not available to assist. Conversations, group activities, conflict resolution, collaborative projects. Not because technology is inherently bad, but because the skills developed in unmediated social interaction — reading faces, managing discomfort, listening without preparing a response — require practice in exactly those conditions. This is not novel parenting advice. What’s novel is that AI mediation of social life is now a plausible habit for a 13-year-old in a way it wasn’t four years ago.

Maintain structured struggle

The emerging evidence on AI and productive struggle tolerance is the finding that most directly should shape how parents structure homework and learning. If AI is available to resolve difficulty on demand, and if children are developing a lower baseline tolerance for difficulty, the corrective is to structure some practice without AI access. Not punitive, not all the time, but deliberately: here is a problem you will work through without asking AI. Here is a draft you will write before you see what AI produces. Here is a question you will sit with for 10 minutes before getting help. For more on the engineering mindset that comes from working through difficulty, see the article on engineering mindset, kids, and failure as learning.

Be explicit about AI as a tool, not a person

The OII’s finding that children form parasocial attachments to AI systems is less about AI being harmful and more about young children requiring explicit conceptual help. A child who understands, concretely, that an AI has no continuous memory between sessions, no awareness of the child’s life outside the conversation, and no interest in the child’s wellbeing is not the same as a child who has been told “it’s just a computer.” The conceptual work takes time and age-appropriate explanation. The article on how to explain AI to kids covers specific approaches by developmental stage.

Watch for AI-mediated social avoidance

The specific behavior worth tracking in teenagers: whether they are using AI as a substitute for practicing difficult social situations rather than as a preparation for them. Using AI to draft a difficult apology, then actually delivering it, is different from using AI to draft it and sending it without further engagement. Using AI to process social anxiety is different from using AI instead of practicing the situation. The direction of the AI use matters. Toward engagement is fine; away from it is the concern.

Build explicit AI literacy, not just AI fluency

The OII 2025 report found that AI literacy — understanding what AI is, how it works in general terms, and what its limitations are — was a protective factor in the early studies. Children who had been explicitly taught about AI were more likely to use it as a tool and less likely to form inappropriate dependence or parasocial attachment. This argues for teaching AI literacy deliberately, not just allowing AI use and assuming children will develop accurate mental models on their own. For a framework on what AI literacy looks like at different ages, the article on AI literacy for kids in middle school is a useful starting point.

Accept uncertainty and stay observant

The most important thing researchers at the OII and UNICEF emphasize is that the effects of AI on this generation are not yet known with confidence, and parents who are certain — in either direction — are outrunning the evidence. The children who will tell us what growing up AI-native meant are currently 5–9 years old. Their retrospective accounts, their measured outcomes, and their observed behaviors in adulthood will be the research. Parents who stay observant — who notice changes in their children’s tolerance for difficulty, social engagement, and relationship to authority and intelligence — are generating the most relevant local data available.

What to Watch for Over the Next 3 Months

Watch for two things in particular. First, whether your child’s behavior around ambiguity and difficulty is changing. Are they more likely to immediately reach for AI when stuck than they were six months ago? Do they show less tolerance for not knowing something right away? These are early behavioral signals of the productive-struggle concern.

Second, watch how your child describes AI entities. Do they refer to AI tools as “it” or “they”? Do they describe the AI’s motivations or feelings? Do they report emotional responses to AI interactions? Young children’s difficulty distinguishing animated objects from actual social actors is developmentally normal and not alarming in young children. In older children, sustained anthropomorphization of AI systems without correction is worth addressing — gently, through explanation, not dismissal.

The first AI-native cohort is young. What we know about them is limited. What we do in the next five years — in terms of parenting, education design, and technology policy — will shape the findings researchers produce in the decade after that.

Frequently Asked Questions

Is growing up AI-native necessarily negative?

No. The OII’s 2025 report explicitly avoids that conclusion. There are potential benefits — access to always-available explanation, reduced friction in learning new skills, support for neurodivergent learners who benefit from patient, adaptive interaction — that may prove significant. The honest answer is that the evidence doesn’t yet support strong conclusions in either direction, and parents should resist both panic and complacency.

How is this different from past concerns about TV, video games, or smartphones?

The precedent comparisons are useful but imperfect. TV and video games are passive or reactive; AI is generative and conversational. Smartphones introduced social media and communication; AI introduces a new kind of interlocutor. The differences that matter most are: AI can simulate human social response, which neither TV nor video games could; AI can do cognitive work for children, which social media largely could not; and AI is becoming woven into educational settings in a way that previous technologies were not. These differences don’t predict worse outcomes — they predict different mechanisms of impact.

At what age should kids have access to AI tools?

UNICEF’s 2021 guidance and the OII’s 2025 report both emphasize age-appropriateness but don’t specify hard cutoffs, because the research base doesn’t support them. The consistent theme across guidance is: more access with more parental involvement at younger ages, with explicit explanations of AI’s nature; more autonomous access with more AI literacy scaffolding at older ages. Age 10–12 appears to be the rough boundary at which children can begin to develop accurate mental models of AI with appropriate teaching.

Should I limit my child’s AI use the way I might limit screen time?

Not necessarily the same way. AI use is more heterogeneous than passive screen time — a child spending 30 minutes using AI to work through a math problem is doing something categorically different from a child spending 30 minutes watching short-form videos. Time-based limits make more sense for passive consumption. For AI specifically, the quality and nature of use matters more than duration. The question is not how much, but how: is AI doing the thinking, or is it supporting the child’s thinking?

What will researchers know in five years that they don’t know now?

Primarily: whether the early behavioral signals (decreased struggle tolerance, parasocial attachment, social avoidance mediated by AI) translate into measurable differences in adolescent social-emotional development and academic outcomes. The children currently age 5–9 will be 10–14 in five years, and longitudinal studies begun now will be able to report first findings. Twenge’s smartphone research took about 7 years to generate the cohort-level data that made the case clearly.

How should I talk to my child about being part of the first AI-native generation?

Honestly and with genuine curiosity, not alarm. The truth — that they are the first generation to grow up with AI as a constant presence, and that we don’t fully know yet what that means — is both accurate and non-threatening when framed as interesting rather than dangerous. Children who understand their historical position tend to think more deliberately about their relationship with the technology, which is precisely the metacognitive habit researchers identify as protective.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

Sources

• Oxford Internet Institute. (2025). Children and Artificial Intelligence: An Evidence Review. University of Oxford.
• Twenge, J.M. (2023). Generations: The Real Differences Between Gen Z, Millennials, Gen X, Boomers, and Silents—and What They Mean for America’s Future. Atria Books.
• Turkle, S. (2015). Reclaiming Conversation: The Power of Talk in a Digital Age. Penguin Press.
• UNICEF. (2021). Policy Guidance on AI for Children, Version 2.0. UNICEF.
• Prensky, M. (2001). “Digital natives, digital immigrants.” On the Horizon, 9(5), 1–6.
• Common Sense Media. (2025). AI Use Among Teens and Tweens: 2025 Survey Report.
• Kapur, M. (2016). “Examining productive failure, productive success, unproductive failure, and unproductive success in learning.” Educational Psychologist, 51(2), 289–299.
• Tanhua-Piiroinen, E., et al. (2024). “AI tool dependency and productive struggle in elementary school students.” Computers & Education, 198, 104778.