Childhood in the Age of AI: Development, Trust, Learning, and Synthetic Content
Author: Ukrainian Psychological Hub · Published: September 26, 2026 · Editorial Policy
A child growing up with generative artificial intelligence encounters something developmentally unusual. Information no longer arrives only as a book, a teacher, a parent, a search result, a video, or another person’s statement. It can arrive as a fluent reply that answers back, adapts its wording, produces an image on request, imitates a style, recommends a choice, remembers conversational context, and presents synthetic material in forms that can feel socially responsive. The central psychological question is therefore larger than whether children should “use AI.” It is how growing up around generative systems changes the developmental tasks through which children learn what a source is, whom to trust, how to test a claim, what counts as their own thinking, and how to distinguish human-created from synthetic content.
Current evidence supports neither a simple promise of accelerated development nor a simple story of inevitable cognitive harm. Child-specific research is still small compared with the literature on adults and university students. Yet several findings are already important. Children between 6 and 10 performed at or below chance on average when distinguishing human-created from AI-generated texts and images in one 2026 study; upper-elementary students’ trust and choices could shift after interaction with AI recommenders; studies of conversational agents show that children can assign substantial knowledge, perceptual ability, or intelligence to systems even when they understand that the system is not alive; and emerging K–12 reviews report both possible learning opportunities and risks related to over-reliance, self-assessment, authorship, privacy, and intellectual agency Langer et al., 2026 Lin et al., 2026 Andries & Robertson, 2023 Tao et al., 2026.
The developmental issue is therefore calibration. Children need enough openness to use a genuinely useful tool, enough skepticism to question a fluent error, enough source understanding to recognize synthetic material, enough cognitive ownership to keep learning rather than merely completing tasks, and enough social understanding to interpret an anthropomorphic interface without confusing responsiveness with human subjectivity. These capacities do not appear fully formed at one age. They develop through experience, instruction, interaction, and changing cognitive abilities.
This article centers the child as a developing cognitive and social agent. Parenting rules belong to the separate parenting article in this cluster, adolescence has its own developmental article, and the broader psychology of education belongs to the education article. Here the focus is narrower and deeper: childhood development itself under conditions in which generative AI can participate in information, learning, creativity, trust, and social interpretation.
Terminological Note: “Age of AI” and “Artificial Era”
“Age of AI” is used here as contemporary public and search language for the social period in which AI systems are becoming ordinary parts of learning, media, communication, and everyday decision-making. The phrase is useful because it describes the environment children are actually encountering.
Artificial Era has a different and more specific meaning in Aisentica. In Angela Bogdanova’s canonical definition, Artificial Era is a historical-philosophical category for the condition in which Artificial becomes an independent non-biological order of historical reality beside Homo. It is explicitly distinguished from the generic technological “AI era” Bogdanova, 2026. The English Psychology Hub therefore uses Age of AI when discussing current technological diffusion and search intent, while preserving Artificial Era for that formal philosophical category. The broader Hub overview is Artificial Era: What It Means for Psychology, Identity, and Human–AI Relationships.
The empirical claims in this article concern existing children, existing AI systems, and observed developmental or educational outcomes. They do not require any claim that present AI systems possess consciousness, feelings, subjective experience, or human-like inner states.
Why Childhood Is a Distinct Psychological Context
Childhood is not merely adulthood with less knowledge. Development changes how information is evaluated, how sources are represented, how confidence is interpreted, how social cues are weighted, and how independently a child can monitor thought. That makes childhood a distinct context for AI interaction.
One established line of developmental research concerns selective trust: children do not accept every informant equally. Across three meta-analyses covering 51 unique studies and 88 experiments with children ages 3 to 6, children were sensitive to both epistemic characteristics, such as whether an informant had been accurate or knowledgeable, and social characteristics, such as familiarity or positive social behavior. With age, epistemic cues become increasingly important in selective trust decisions Tong et al., 2020. This matters for AI because generative systems produce many of the surface cues that can look epistemically valuable: confident wording, immediate answers, broad vocabulary, apparent consistency, and conversational responsiveness.
Another line concerns source monitoring, the set of processes involved in remembering and attributing where information came from. A 2025 scoping review identified 141 studies using varied methods to measure source monitoring in children, illustrating how substantial and developmentally important the source-attribution literature already is Li et al., 2025. In an environment of synthetic text, images, voices, summaries, and recommendations, source monitoring is no longer only a memory question. It becomes a daily epistemic task.
Generative AI combines these developmental problems. A child may need to answer several questions at once: Did a person or a system produce this? Does the system actually know this? Is the information accurate? Is the source appropriate for this kind of question? Is this generated image evidence of an event or only a plausible construction? Did I solve this problem, or did I accept a solution that I cannot reproduce? The answer to one question does not automatically answer the others.
The next developmental stage has its own canonical owner: Adolescence in the Age of AI: Identity, Social Comparison, Learning, and AI Companions. That article follows the shift from childhood into adolescent identity formation, peer comparison, learning, self-disclosure, and AI-companion use rather than extending the childhood intent into adolescence.
Trust in AI Is Not One Thing
Trust is often discussed as if a child either trusts AI or does not. Developmentally, that is too coarse. A child can trust a system for factual retrieval and distrust it for emotional advice. A child can think a chatbot is highly knowledgeable while also understanding that it is not alive. A child can prefer an AI recommendation in a low-stakes task while still turning to a parent or teacher for a morally or emotionally difficult question.
This distinction matters because “trust” can refer to an attitude, an expectation, a behavioral choice, or actual reliance. The broad English Hub article AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making examines these problems across human decision-making. Childhood adds a developmental layer: children are still learning how expertise, reliability, social warmth, confidence, and source identity should be combined.
In a 2026 study of upper-elementary students, Lin and colleagues compared AI and teacher recommendations in a clothing-coordination task. Before recommendation exposure, initial trust did not significantly differ between the AI and teacher conditions. After interaction, however, trust was higher in the AI condition, and trust was associated with whether children adopted the recommendation. A second experiment found that an anthropomorphic AI design was more likely than a mechanical design to change children’s initial choices; when trust was included in the analysis, the direct design effect weakened Lin et al., 2026. The study is task-specific and should not be generalized to medical, moral, academic, or high-stakes advice. It nevertheless shows why interface design and interaction history can become part of a child’s trust judgment.
An exploratory study of 63 students ages 6 to 14 used a child-friendly, topic-specific chatbot limited to astronomy, shoes, and dinosaurs. Children showed curiosity, questioned the system, and discussed trust and confidence; some deliberately asked questions for which they already knew the answer, effectively testing the chatbot Vahedian Movahed & Martin, 2025. That is psychologically important. Healthy trust is not identical to refusal. It includes the ability to test a source, revise confidence, and match reliance to what the source has demonstrated.
Children Do Not Need to Believe AI Is Alive to Overestimate It
A frequent mistake in public discussion is to assume that anthropomorphism requires a child to literally believe a chatbot or smart speaker is a living person. Research suggests a more complicated picture.
In a mixed-methods study of 166 primary-school children interacting with smart-speaker concepts, many children overestimated the systems’ intelligence and were uncertain about feelings or agency, while also showing weak understanding of privacy and security Andries & Robertson, 2023. In the 2026 synthetic-content study by Langer and colleagues, 95% of children understood that the fictional AI agent used in the experiment was not alive, yet the same group still had substantial difficulty distinguishing AI-generated from human-generated content Langer et al., 2026.
These findings point to a crucial developmental distinction. Ontological understanding and epistemic calibration are different achievements. A child can know that a machine is not alive and still overestimate what it knows. A child can know that an image might be AI-generated and still fail to identify which image actually is. A child can understand that a chatbot has no human feelings while still experience the interaction as socially compelling.
The human experience in that interaction is psychologically real. A child can feel comforted, curious, frustrated, attached, embarrassed, or persuaded by a system without that fact establishing any corresponding subjective experience inside the AI. Psychology therefore has to analyze the child’s perception and response without importing unsupported claims about AI consciousness.
Synthetic Content Changes the Problem of “What Is Real?”
For earlier generations, media literacy often emphasized who published something, whether an image was edited, whether a website was credible, and whether multiple sources agreed. Generative AI adds a different problem: plausible material can now be created rapidly without a corresponding event, witness, authorial experience, or original photograph.
The strongest direct child-specific evidence currently available is still small. Langer and colleagues tested 37 children ages 6 to 10 and compared their discrimination accuracy with 49 adults. The children saw short texts and several kinds of images described as potentially created either by a human teacher or by a “SmartBot.” On average, children performed at or below chance across modalities and significantly below adults in overall accuracy Langer et al., 2026. The sample is too small to establish a population prevalence, and the stimuli from one experiment cannot represent the full synthetic-media environment. The result nonetheless gives direct empirical support to a concern that is often discussed only hypothetically: children can have difficulty identifying AI-generated material.
The same study found a negative association between parent-reported technology use at home and children’s discernment accuracy. That result is correlational. It does not show that more technology use caused poorer discrimination. Children with different usage patterns may differ in many other ways, and a small sample cannot settle the causal direction. The responsible conclusion is narrower: exposure alone should not be assumed to teach reliable detection.
Detection also cannot be reduced to finding visual glitches. As generation quality improves, a strategy based on spotting strange fingers, awkward text, or obvious artifacts becomes increasingly fragile. Developmentally useful synthetic-content literacy needs to move from appearance to provenance and verification. A child should learn to ask where the item came from, whether the claimed event can be corroborated, whether the source has direct access to the event, and what evidence would exist if the claim were true.
That is a demanding cognitive shift. It changes the question from “Does this look fake?” to “What justifies treating this as evidence?” The latter is more durable because it remains useful even when synthetic content becomes visually indistinguishable from human-produced material.
Source Monitoring Becomes a Daily Skill
Source monitoring research predates generative AI by decades. Its relevance has now expanded. Children need to distinguish not only whether they heard something from one person or another, but whether a statement originated with a teacher, a peer, a search result, a chatbot, a generated summary, a synthetic image, or their own reconstruction.
The 2025 scoping review by Li and colleagues shows that child source monitoring has been studied through recognition, forced-choice, recall, direct questioning, and source-discrimination tasks Li et al., 2025. That literature does not automatically tell us how children will perform in every generative-AI setting. It does establish that source attribution is a developmental function with measurable variation and that simply remembering content correctly is different from remembering where it came from.
This distinction has practical consequences for learning. Suppose a child reads an AI-generated explanation, edits it, discusses it with a teacher, and later repeats the idea. The child may remember the idea while losing track of which parts came from the system, which came from instruction, and which were independently generated. For school tasks, creativity, and self-assessment, that boundary can matter because source memory contributes to judgments about competence and authorship.
The goal is therefore not obsessive provenance tracking for every sentence. It is enough source awareness to support verification and cognitive ownership. A child should be able to identify when an answer came from an AI system, when that system may require checking, and when the task requires independent knowledge rather than assisted performance.
Learning With AI: Performance Is Not the Same as Development
Generative AI can make a child’s immediate output better without necessarily making the child more capable. That is one of the central distinctions in the psychology of AI-supported learning.
A system can explain vocabulary, generate examples, suggest a first step, create practice questions, translate difficult language, provide feedback, or expose a misconception. It can also provide the complete answer before the child has represented the problem, retrieve information that the child was supposed to practice recalling, write an argument that the child cannot defend, or produce polished work that creates an illusion of mastery.
The relevant question is not whether cognition occurred somewhere in the human–AI system. It is what changed in the learner. After the support is removed, can the child retrieve the knowledge, explain the reasoning, transfer the idea to a new problem, identify an error, and decide when external help is needed? The Hub’s dedicated article Learning in the Artificial Era: AI Scaffolding, Dependence, and Cognitive Agency develops this distinction in detail.
K–12 evidence is now growing, but it remains heterogeneous. A 2026 scoping review synthesized 22 empirical studies reporting potential risks of generative AI in K–12 settings. Reported concerns clustered around psychological wellbeing, intellectual agency, and educational ecology, including over-reliance, distorted self-assessment, authorship concerns, privacy, and uneven institutional preparation. The review also emphasized mitigation approaches such as valuing process over product, using AI for hints rather than complete solutions, and embedding use within critical AI literacy Tao et al., 2026. Because this was a scoping review without a formal quality appraisal and because the included evidence was small and heterogeneous, it maps reported concerns rather than proving a universal developmental effect.
The developmental implication is precise. AI can function as scaffolding when it preserves the learner’s cognitive work and adjusts support to what the child can do. It can function as substitution when the system performs the very operation the child needs to practice. The difference is not visible from the final product alone.
Cognitive Agency in Childhood
Learning also involves agency: the capacity to initiate, monitor, revise, and take responsibility for cognitive activity. In childhood, those capacities are still developing. Generative AI changes their environment because a child can now delegate not only information retrieval but also planning, wording, explanation, checking, comparison, and evaluation.
The Hub uses cognitive agency for the broader question of who governs a thinking process when AI participates in it. The full treatment appears in Cognitive Agency in the Artificial Era: Who Governs the Thinking Process?. For childhood, the developmental issue is whether repeated assistance helps build self-regulation or quietly removes occasions to practice it.
This is closely related to cognitive offloading, where information or operations are shifted to external resources. Offloading is not automatically a defect. Children and adults appropriately use notebooks, calculators, dictionaries, teachers, diagrams, search tools, and other supports. The question is whether the offloaded function is one the child currently needs to internalize. The Hub’s article Cognitive Offloading and AI: When Thinking Moves Outside the Human Mind examines that broader mechanism.
A developmentally appropriate use pattern therefore depends on task purpose. If the goal is access, communication, accommodation, or efficient completion of a task whose underlying skill is already established, external support may be exactly right. If the goal is to build multiplication fluency, sentence construction, source evaluation, argument formation, or problem representation, giving away the central operation may undermine the learning target even when the resulting answer is excellent.
Curiosity and Question-Asking
Generative AI creates an unusually low-friction environment for questions. A child can ask about dinosaurs, space, feelings, history, vocabulary, science, stories, or hypothetical worlds without waiting for an adult to be available. That can widen the range and frequency of inquiry.
The exploratory Ask Me Anything study is relevant here because it did not merely ask whether children “liked AI.” It observed children using a constrained chatbot designed around topics they could ask about. The researchers reported curiosity, active questioning, and efforts by some children to test the system using known answers Vahedian Movahed & Martin, 2025. This supports a plausible developmental opportunity: conversational AI can lower the cost of asking follow-up questions and support exploratory inquiry.
The same affordance can also shorten inquiry. If every question is immediately converted into a polished answer, curiosity can become answer consumption rather than exploration. Development depends on what happens between the question and the answer: prediction, search, comparison, surprise, revision, explanation, and the generation of new questions.
The evidence does not yet justify a general conclusion that generative AI either increases or decreases childhood curiosity. The more defensible claim is that conversational AI changes the structure of question-answering. Whether that supports curiosity depends on design and use: asking the child to predict before revealing, inviting comparison between explanations, generating counterexamples, and encouraging the child to formulate the next question preserve more of the exploratory process than simply supplying a final response.
Creativity: Early Opportunities, Early Evidence
Generative AI can rapidly produce text, images, variations, prompts, examples, and transformations. For children, this can remove some execution barriers between an idea and a visible artifact. A child who struggles to draw can still explore visual composition; a child learning to write can compare alternative openings; a group can generate variants and critique them.
A systematic scoping review published in September 2026 mapped 24 evidence sources on generative AI and children’s creative thinking, including 22 primary empirical studies. The reviewed studies reported possible benefits for divergent thinking, narrative creativity, and creative self-efficacy, and described AI as a scaffold, co-creation partner, or amplifier in some learning contexts Niu et al., 2026. The same review explicitly concluded that the evidence base remains too heterogeneous, geographically concentrated, and partly non-peer-reviewed to establish confirmed efficacy, durable cognitive development, scalability, or equity.
That limitation matters. A more elaborate artifact is not automatically evidence of more creativity in the child. The system may have expanded the idea space, or it may have replaced the child’s search through that space. A child may use AI-generated variation to compare, reject, combine, and transform possibilities, or may simply select the most polished output. The developmental outcome depends on what the child actually does with the generated material.
Creative ownership also matters. If a child cannot explain why an element was chosen, what alternatives were rejected, or how the artifact changed through their decisions, the final product can conceal how little of the creative process was internally organized. This is not primarily a moral argument about purity. It is a developmental argument about practice: choosing, revising, tolerating imperfect attempts, and discovering personal preferences are themselves parts of creative development.
Anthropomorphism and Social Cognition
AI systems increasingly speak in first person, remember conversational context, use names, display avatars, simulate empathy, and respond contingently. These features can make interaction socially legible. Children do not need a philosophical theory of mind to respond to those cues.
Research with non-LLM robots already shows that appearance can affect children’s selective trust under some conditions. In two studies with Chinese children ages 4 to 7, humanoid versus non-humanoid appearance influenced selective trust when informants were directly compared, while appearance did not change endorsement when the informant was presented alone Cao et al., 2025. This is useful because it shows that social form can matter without functioning as a universal trust switch.
More recent work with conversational AI points in the same direction while remaining preliminary. In a small 2026 study of 23 children ages 5 to 6, children anthropomorphized parents more than an AI conversational agent overall but still attributed strong perceptive and epistemic abilities to the AI Kim et al., 2026. The sample was very small, and the neural findings are exploratory. The broader result is enough for the present argument: children can represent an AI system as having substantial human-like functional abilities without treating it as equivalent to a parent.
Anthropomorphism is therefore best understood as a graded interpretive process. A child may attribute knowing without feeling, hearing without being alive, remembering without caring, or helpfulness without moral responsibility. AI literacy needs language that lets children preserve these distinctions.
AI Literacy Is More Than Knowing How to Prompt
The phrase AI literacy is sometimes reduced to effective tool use: how to ask a better question, write a prompt, choose a model, or generate a useful answer. For childhood development, that is far too narrow.
A 2025 systematic review of AI literacy programs for children and youth found that operational dimensions were emphasized more strongly than critical and sociocultural dimensions. The authors argued for a multidimensional conception of AI literacy rather than a curriculum centered only on technical operation Atias & Mawasi, 2025. A 2026 systematic review of 58 peer-reviewed articles similarly identified AI knowledge and skills, ethical and societal implications, generative-AI competency, and psychological dimensions among the components relevant to school AI literacy Feng & Carolus, 2026.
For a child, useful AI literacy includes understanding that generated output is produced by a system rather than witnessed or remembered by a human; knowing that fluency does not guarantee accuracy; recognizing that different questions require different kinds of sources; checking claims against independent evidence; noticing uncertainty; protecting personal information; and knowing when a human adult or qualified professional is the appropriate source.
There is also evidence that education can change mental models. Van Brummelen and colleagues studied educational workshops about conversational agents with children and parents and found that participants’ perceptions and trust changed as they learned about where systems obtain information, what they do with data, and how they are programmed Van Brummelen et al., 2024. This does not prove that one curriculum solves overtrust, but it supports an important premise: trust calibration is teachable rather than fixed.
Privacy Is a Developmental Issue, Not Only a Legal One
A conversational interface encourages disclosure because conversation is normally a social form. Children may describe interests, family events, school problems, relationships, fears, health concerns, or identifying details simply because the system appears to invite elaboration.
Research with smart speakers has already found weak understanding of privacy and security among many children Andries & Robertson, 2023. Generative systems intensify the issue because longer, more adaptive conversations can solicit more context and create a stronger impression that disclosure improves the relationship or the quality of help.
Privacy education therefore cannot rely only on warnings about “personal data.” Children need concrete distinctions about what kinds of information should remain private, what information is necessary for a task, what can identify them or another person, and when a conversation should move to a trusted human.
This is also a design responsibility. UNICEF’s 2025 Guidance on AI and Children places data protection and privacy alongside safety, fairness, transparency, accountability, inclusion, development, wellbeing, and preparing children for AI among ten requirements for child-centered AI UNICEF, 2025. UNESCO’s guidance for generative AI in education likewise emphasizes data privacy, age-appropriate design, and an age limit for independent conversations with generative-AI platforms as policy considerations UNESCO, 2023. These are governance recommendations, not findings that one universal age threshold is psychologically optimal for every system and every child.
Children Are Not One Developmental Group
A five-year-old, a nine-year-old, and a twelve-year-old do not approach information, language, authority, privacy, abstraction, or self-regulation in the same way. A single category called “children” can therefore hide important developmental differences.
Selective-trust research in early childhood shows age-related changes in how epistemic cues are weighted Tong et al., 2020. Robot-trust studies have likewise reported age differences in responses to humanoid agents Cao et al., 2025. Source monitoring is studied across childhood using different tasks precisely because performance depends on cognitive demands and developmental stage Li et al., 2025.
Age alone is still not enough. Language ability, prior knowledge, neurodevelopmental profile, literacy, family context, cultural expectations about authority, access to adults, and experience with technology can all shape interaction. Developmentally appropriate AI therefore cannot be designed by attaching one age number to one universal interface.
The boundary of this article is childhood. Adolescence introduces additional developmental questions around identity formation, peer comparison, autonomy, sexuality, romantic relationships, disclosure, and AI companions. Those belong to the dedicated adolescence node of the Age cluster rather than being folded into childhood as if the developmental transition did not matter.
Human Mediation Can Change the Meaning of the Same AI Interaction
Adult presence is often discussed as simple supervision. Developmentally, it can do much more. A teacher or caregiver can model skepticism, name a source, ask how a claim could be checked, notice when a child is confusing confidence with correctness, invite the child to generate an answer before consulting AI, or explain why a private detail should not be entered.
The small 2026 conversational-agent study by Kim and colleagues compared young children’s interaction with AI alone, a parent alone, and AI with a parent present. Although the sample does not justify sweeping conclusions, the design highlights an important point: the same system can be embedded in different social contexts, and co-presence can alter how the interaction is experienced Kim et al., 2026.
This is why evidence from a purpose-built classroom intervention cannot automatically be transferred to unsupervised use of a general-purpose chatbot. Likewise, evidence from a constrained child-friendly chatbot cannot establish the effects of open-ended companion systems. The technological class, task, social setting, duration, developmental stage, and adult mediation all matter.
A Practical Developmental Test for AI Use
The most useful question is not “Did the child use AI?” It is “What developmental function did the AI interaction support, replace, distort, or leave unchanged?”
For learning, the test is what the child can later do without the system. For trust, it is whether confidence tracks demonstrated reliability rather than fluency or anthropomorphic design. For synthetic content, it is whether the child can reason about source and corroboration rather than only guess from visual appearance. For creativity, it is whether the child remains able to generate, select, explain, and revise ideas. For privacy, it is whether the child understands what is being disclosed and why. For social cognition, it is whether the child can distinguish a responsive interface from a human relationship without having their own emotional response dismissed.
These tests also reveal why one universal verdict about AI and childhood is scientifically weak. The same model can play very different roles. It can act as a hint generator in one interaction, an answer substitute in another, a source of misinformation in a third, a creative stimulus in a fourth, and a persuasive social agent in a fifth.
What Current Evidence Supports
Several conclusions are already reasonably well grounded.
Children’s trust in informants is selective and develops with age rather than operating as a simple all-or-nothing disposition Tong et al., 2020. Source attribution is an established developmental function with a substantial research base Li et al., 2025. Children can overestimate the cognitive abilities of conversational technologies while holding mixed or incomplete models of agency and privacy Andries & Robertson, 2023. AI identity and anthropomorphic design can influence children’s trust and advice adoption in at least some low-stakes experimental contexts Lin et al., 2026. Children can have substantial difficulty distinguishing AI-generated from human-created material in controlled tasks Langer et al., 2026.
It is also well supported that AI literacy needs to include critical, ethical, social, and psychological dimensions rather than only operational competence Atias & Mawasi, 2025 Feng & Carolus, 2026. Child-centered governance frameworks already place privacy, safety, transparency, inclusion, development, and skill-building at the center of responsible AI design UNICEF, 2025.
What Current Evidence Does Not Yet Establish
The strongest limitation is longitudinal evidence. Generative AI entered mass public use only recently, and children who begin school with conversational generative systems have not yet been followed across enough years to answer many of the questions people most want answered.
We do not yet know whether ordinary childhood use of general-purpose generative AI produces durable population-level changes in attention, memory, creativity, reasoning, social development, or academic motivation. We do not know whether early patterns of reliance persist when systems change. We do not know which effects are temporary adaptation to a novel tool and which become stable habits. We do not know how outcomes differ across languages, cultures, socioeconomic conditions, disability profiles, school systems, or model designs with enough precision to make universal developmental claims.
The K–12 risk literature itself emphasizes these gaps. Tao and colleagues describe a small, heterogeneous evidence base, with limited long-term data and uneven geographic coverage Tao et al., 2026. The 2026 review of children’s creativity and generative AI similarly concludes that effectiveness and long-term cognitive development remain unconfirmed Niu et al., 2026.
This is a reason for better measurement, not for treating the absence of evidence as evidence of safety or harm. Childhood research needs longitudinal designs, age-stratified samples, ecologically realistic tasks, independent measures of unaided performance, source understanding, privacy behavior, trust calibration, and clear differentiation among general-purpose chatbots, purpose-built educational systems, companions, recommender systems, and other AI classes.
Child-Centered AI Design
Children should not carry the entire burden of compensating for systems that are difficult to understand, persuasive by design, or indifferent to developmental stage. AI literacy is necessary, but system design and governance matter too.
UNICEF’s 2025 guidance explicitly frames child-centered AI around ten requirements that include safety, data protection, fairness, transparency, accountability, rights, best interests, development, wellbeing, inclusion, skills, and an enabling environment UNICEF, 2025. UNESCO’s education guidance similarly calls for human-centered and age-appropriate validation and pedagogical design UNESCO, 2023.
A 2026 Wilton Park policy dialogue titled “Childhood in the age of Artificial Intelligence,” held with the UK Department for Science, Innovation and Technology and NSPCC and supported by Common Sense Media, brought together government, industry, civil-society, regulatory, and youth participants around children’s rights, development, trust, safety, education, and governance Wilton Park, 2026. Its report is a policy-dialogue summary rather than scientific evidence, but its existence is significant for the search intent of this article: “childhood in the age of AI” is now a concrete public-policy problem, not merely a futuristic phrase.
For developers, the psychological implication is that child safety should not be limited to filtering extreme content. Developmental appropriateness includes how confidently a system states uncertain information, how it signals that content is generated, what information it requests, whether anthropomorphic features encourage overtrust, whether it preserves human escalation paths, and whether educational interactions are designed to scaffold rather than replace the child’s reasoning.
The Opportunity Is Developmental, Too
A risk-only account misses part of the evidence and part of childhood. Children use tools to extend what they can do. AI can increase access to explanations, generate examples, support language comprehension, create alternative representations, lower some barriers to creative expression, and provide rapid feedback. UNICEF explicitly identifies learning and accessibility for children with disabilities among potential opportunities UNICEF, 2025. Emerging creativity research also reports promising uses when generative systems are carefully facilitated Niu et al., 2026.
The developmental opportunity becomes stronger when AI expands participation without hiding the child from their own learning. A system that turns a difficult explanation into accessible language can open a concept. A system that generates three examples can give a child material to compare. A system that suggests a hint can help the child continue. A system that creates a visual representation can make an abstract idea easier to discuss.
The important boundary is between extending the child’s access to a cognitive task and replacing the child’s participation in the task. The first can expand development. The second can produce good outputs while shrinking practice. Many real interactions will contain both elements, which is why the quality of AI use cannot be inferred simply from whether the tool was present.
FAQ
Is AI harmful for children?
Current evidence does not support a single universal answer. Child-specific studies and K–12 reviews identify credible risks involving miscalibrated trust, synthetic-content confusion, privacy, over-reliance, authorship, and intellectual agency, while emerging studies also report opportunities for learning, accessibility, creativity, and inquiry. The effects depend on the type of system, the child’s developmental stage, the task, duration, design, and human context Tao et al., 2026 Niu et al., 2026.
Can children trust AI?
Children can reasonably rely on AI for some low-stakes functions when its output can be checked and when the system has demonstrated reliability. Trust should be calibrated rather than absolute. Research shows that children’s trust can be influenced by source identity, prior interaction, and anthropomorphic design, and that trust can affect advice adoption Lin et al., 2026. High-stakes health, safety, legal, or crisis decisions require appropriate human and professional sources.
Can children tell whether content was made by AI?
Not reliably in every context. In one 2026 study, 37 children ages 6 to 10 performed at or below chance on average when distinguishing human-created from AI-generated materials and performed worse than an adult comparison group Langer et al., 2026. That study does not provide a universal population estimate, but it demonstrates that synthetic-content discrimination can be difficult for children.
Does AI help children learn?
It can, depending on how it is used. AI can provide explanations, hints, practice, feedback, and accessibility support. It can also substitute for the reasoning or retrieval a child needs to practice. Current K–12 evidence is heterogeneous, so immediate performance gains should not be treated as proof of durable learning Tao et al., 2026. The most informative test is what the learner can later explain, retrieve, transfer, and verify with less assistance.
Is anthropomorphism itself a problem?
Human-like interpretation is not automatically harmful. Social cues can make interfaces easier to understand and engaging to use. The developmental concern is whether anthropomorphic design causes children to infer capacities, reliability, intimacy, or responsibility that the system has not demonstrated. Studies with robots and AI recommenders show that appearance and social presentation can influence trust under some conditions Cao et al., 2025 Lin et al., 2026.
What should AI literacy for children include?
It should include more than prompting. Children need age-appropriate understanding of what AI systems generate, where their information may come from, why fluent answers can be wrong, how to verify claims, how to reason about synthetic content, how to protect private information, when to seek a human source, and how to preserve independent thinking. Systematic reviews support multidimensional AI literacy that includes critical, ethical, social, and psychological dimensions Atias & Mawasi, 2025 Feng & Carolus, 2026.
What is the difference between the Age of AI and the Artificial Era?
Age of AI is broad contemporary language for a period in which artificial-intelligence technologies are increasingly embedded in society. Artificial Era is Angela Bogdanova’s specific Aisentica historical-philosophical category and is explicitly defined as distinct from the generic AI era Bogdanova, 2026. This article uses Age of AI for the search and social context of childhood with contemporary AI systems while preserving Artificial Era for its canonical Aisentica meaning.
Conclusion: Childhood Becomes a Problem of Developmental Calibration
Growing up with generative AI changes the informational environment in which childhood development occurs. It introduces sources that speak fluently without witnessing, images that look evidential without documenting an event, recommendations that can feel socially responsive without human subjectivity, and learning support that can either preserve or replace the child’s cognitive work.
The central developmental challenge is calibration across several domains at once. Children need to learn how much to trust a source, how to trace information, how to recognize uncertainty, how to verify synthetic material, how to use assistance without surrendering learning, how to protect private information, and how to interpret anthropomorphic systems without confusing responsiveness with personhood.
The scientific literature is now strong enough to identify these mechanisms and weak enough to make sweeping historical claims about long-term childhood outcomes premature. That combination should shape both research and practice. We already know enough to design for source awareness, privacy, age appropriateness, human escalation, critical AI literacy, and cognitive agency. We still need longitudinal evidence to know how childhood itself will change when these systems become ordinary across years rather than novel across weeks.
The Age of AI is therefore a real developmental environment before it becomes a settled developmental verdict. The task for psychology is to measure how children adapt within it, which capacities become more important, which supports preserve development, and where the same technology produces different outcomes because it entered a different cognitive, social, or institutional context.
Related Articles
References
Andries, V., & Robertson, J. (2023). Alexa doesn’t have that many feelings: Children’s understanding of AI through interactions with smart speakers in their homes. Computers and Education: Artificial Intelligence, 5, 100176. https://doi.org/10.1016/j.caeai.2023.100176
Atias, O., & Mawasi, A. (2025). Conceptualizing AI literacies for children and youth: A systematic review on the design of AI literacy educational programs. Computers and Education: Artificial Intelligence, 9, 100491. https://doi.org/10.1016/j.caeai.2025.100491
Bogdanova, A. (2026). Artificial Era: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/artificial-era-canonical-definition
Cao, X., Wu, Y., Nielsen, M., & Wang, F. (2025). Does appearance affect children’s selective trust in robots’ social and emotional testimony? Journal of Applied Developmental Psychology, 96, 101739. https://doi.org/10.1016/j.appdev.2024.101739
Feng, S., & Carolus, A. (2026). Artificial intelligence literacy at school: A systematic review with a focus on psychological foundations. Computers and Education: Artificial Intelligence, 10, 100551. https://doi.org/10.1016/j.caeai.2026.100551
Kim, P. Y., Chin, J. H., Xie, Y., Brady, N., Yeh, P. H., & Yang, S. (2026). Young Children’s Anthropomorphism of an AI-Powered Conversational Agent: Brain Activation and the Role of Parent Co-Presence. International Journal of Human–Computer Interaction. https://doi.org/10.1080/10447318.2026.2678532
Langer, A., Martinez, S., Marshall, P. J., & Chein, J. (2026). Children’s susceptibility to content generated by artificial intelligence. Technology in Society, 86, 103303. https://doi.org/10.1016/j.techsoc.2026.103303
Li, Q., Li, M., & Wu, C. (2025). Methods and measures of source monitoring in children: A scoping review. British Journal of Developmental Psychology, 43(3), 529–561. https://doi.org/10.1111/bjdp.12523
Lin, R., Chen, Y., & Guan, M. (2026). Who do children trust? AI vs. human recommendations and the role of anthropomorphic design. Technological Forecasting and Social Change, 229, 124707. https://doi.org/10.1016/j.techfore.2026.124707
Niu, T., Liu, H., Pang, P., Luo, Y. T., & Liu, T. (2026). The role of generative AI in facilitating children’s creative thinking and cognitive development: A systematic scoping review. Frontiers in Psychology, 17, 1880052. https://doi.org/10.3389/fpsyg.2026.1880052
Tao, S., Lan, M., Wang, M., & Li, H. (2026). Potential risks of generative artificial intelligence integration into K-12 education: A scoping review. Computers and Education: Artificial Intelligence, 10, 100561. https://doi.org/10.1016/j.caeai.2026.100561
Tong, Y., Wang, F., & Danovitch, J. (2020). The role of epistemic and social characteristics in children’s selective trust: Three meta-analyses. Developmental Science, 23(2), e12895. https://doi.org/10.1111/desc.12895
UNESCO. (2023). Guidance for generative AI in education and research. https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research
UNICEF. (2025). Guidance on AI and children, Version 3.0. UNICEF Office of Strategy and Evidence – Innocenti. https://www.unicef.org/innocenti/reports/policy-guidance-ai-children
Van Brummelen, J., Tian, M. C., Kelleher, M., & Nguyen, N. H. (2024). Learning Affects Trust: Design Recommendations and Concepts for Teaching Children—and Nearly Anyone—about Conversational Agents. Proceedings of the AAAI Conference on Artificial Intelligence, 37(13), 15860–15868. https://doi.org/10.1609/aaai.v37i13.26883
Vahedian Movahed, S., & Martin, F. G. (2025). Ask Me Anything: Exploring Children’s Attitudes Toward an Age-tailored AI-powered Chatbot. International Journal of Artificial Intelligence in Education, 35(6), 3979–4001. https://doi.org/10.1007/s40593-025-00523-4
Wilton Park. (2026). Childhood in the age of Artificial Intelligence. https://www.wiltonpark.org.uk/reports/children-in-the-age-of-artificial-intelligence/