Artificial intelligence is entering kindergartens and family living rooms at remarkable speed, and a new open-access study argues that the most important question about this shift is not whether the technology helps or harms children, but how it restructures the developmental conditions under which young children learn to act as agents at all. The paper, published in the journal AI, Brain and Child on 11 August 2026, is an exploratory, theory-generating conceptual inquiry built on in-depth interviews with six interdisciplinary scholars rather than direct measurements of children’s behaviour, and its authors are explicit that the findings should be read as conceptual propositions to be tested, not as empirical demonstrations of effects.
The six participants came from backgrounds spanning early childhood education, educational technology, AI ethics and child developmental psychology, and all had published on the intersection of AI and early childhood education within the previous two years. The first author conducted one-on-one online interviews lasting between 38 and 58 minutes, averaging 46 minutes, with the four Chinese participants interviewed in Mandarin and the two American participants in English. Recordings were transcribed verbatim, Mandarin data were translated into English, and the translations were verified both by the participants themselves and by external translators unconnected to the project.
The analytical approach was primarily deductive and partly inductive content analysis, carried out in MAXQDA 2022 through a five-phase coding procedure. The second author completed the coding independently, the first author then reviewed the results comprehensively, and the two resolved discrepancies through iterative discussion until they reached what the paper describes as a high level of inter-coder agreement. A four-scholar peer debriefing team and an inquiry audit by the corresponding author added further checks. The theoretical scaffolding for the analysis combined three frameworks: Floridi’s multiple-realisability-of-agency thesis, actor-network theory, and complexity systems thinking. Together these supplied the vocabulary for the study’s central distinction, an asymmetry between AI as a system that functionally performs the markers of agency and the young child as a developing agent whose agency emerges through embodied, social and experiential processes.
That asymmetry produced the study’s core finding, a tension the experts described as the gap between an efficient illusion of agency and the authentic agency children generate themselves. Expert D, quoted in the paper, framed it through the multiple-realisability lens: AI is essentially agency without intelligence, possessing no genuine intentions or real comprehension, yet its technical goal is to simulate the three core markers of agency, interactivity, autonomy and adaptability. Young children, by contrast, are agents with developing intelligence whose agency is built from the ground up through hands-on practice in an imperfect, friction-rich world. Expert E posed the question that threaded through all the interviews: is this carefully crafted, frictionless and highly efficient illusion of agency actually empowering children, or quietly eroding the organic way they generate their own authentic agency?
The first dimension the experts examined was interactivity, and here the concern centred on what the paper calls the removal of productive friction. One expert offered a particularly strong claim, suggesting AI could fully replace the roles of parents and teachers, but the authors note that the broader literature contests this replacement view and that the interviews more consistently pointed to a reduction in the productive friction of human interaction. Expert A explained that when children talk with peers or adults they constantly encounter misunderstandings, awkward pauses and differing viewpoints, and that this friction forces them to express themselves better, interpret what others mean and build meaning together, developing empathy, self-regulation and resilience in the process. When AI seamlessly adapts and skips all that, the expert argued, children risk losing those vital developmental moments. Across the interviews, the worry was not smooth interaction itself but the possibility that increasingly frictionless exchanges reduce opportunities to practise the negotiation, interpretation and emotional regulation through which social-emotional agency develops.
On autonomy, the experts described what the authors call the tripartite interaction, the everyday triangle of child, adult and AI through which young children increasingly learn and act. Expert C warned that AI’s near-omnipotent capabilities in analysis, judgment and prediction could gradually turn it into an omnipotent authority figure within this triangle, stripping parents and teachers of their traditional roles and leading all three parties to defer to AI’s judgments, with children’s decision-making power largely ceded to the machine. Expert A added that early childhood autonomy develops through experimentation, successive decision-making and reflection on one’s own actions, and that when AI preemptively provides solutions, corrects mistakes or guides decisions without transparent scaffolding, it may short-circuit that process, depriving children of trial-and-error learning and blurring their understanding of who is thinking, deciding or initiating action. Viewed through the actor-network lens, the paper argues that AI may contribute to a redistribution of agency across the network, an emergent and relational effect rather than a power the technology itself possesses.
The third dimension, adaptability, produced perhaps the most vivid imagery. Expert C likened AI to an ultra-advanced swimming machine that lets children drift comfortably through the ocean of knowledge without ever learning the strokes themselves. Expert A described algorithmic recommendation systems that continuously deliver homogeneous content tailored to children’s initial preferences, creating an information environment where what you like is exactly what you will see, a mechanism compared to the echo chambers adolescents encounter on social media. Long-term immersion in such a cognitive comfort cocoon, the expert argued, would severely limit children’s opportunities to engage with diverse viewpoints and challenging information. Expert E contrasted the friction-heavy physical world, where building blocks fall because of gravity and paints get messy when they mix, with the completely frictionless digital world of smooth screens, instant feedback and an infinite undo button, worrying that this excessively smooth material experience deprives children of the messy groundwork needed for real, embodied agency.
From these expert perspectives the authors derive their central conceptual contribution, a reorientation from pedagogical scaffolding toward learning-ecosystem architecture, crystallised in what they call a pedagogy of friction. The proposal draws on established learning-science research showing that conditions which slow or complicate immediate performance can enhance durable learning, what Bjork and Bjork termed desirable difficulties, and that carefully staged struggle prior to instruction can outperform smooth, fully guided experiences, as research on productive failure demonstrates. In practice, the authors argue, educators and parents should consciously counteract the excessive smoothness technology brings, preserving hands-on activities with real physical laws such as building with blocks, playing with sand and mud, and creating by hand, while treating disputes, negotiation and compromise among peers not as disruptions to be eliminated but as important opportunities for developing social agency.
The implications the authors draw extend across design, practice and policy. For technology designers, they propose shifting from a User Experience paradigm that pursues seamless, frictionless interaction to a Growth Experience paradigm whose core is not to make learning effortless but to promote deep engagement, illustrated by a hypothetical AI-assisted block-building application that allows a virtual tower to collapse under simulated gravity and then asks the child why it fell rather than auto-correcting it. For educators, they call for a transformation from technology supervisors who control screen time and block inappropriate content into learning ecosystem architects who proactively design and maintain environments conducive to authentic agency. For policy and parenting, they recommend moving beyond the quantitative management of screen time toward questions about who dominates the interaction and whether it enhances or undermines a child’s autonomy, creativity and critical thinking, while noting that this proposition should be tested directly in future research.
The authors are candid about the boundaries of what they can claim. The evidence is the perception and theoretical judgment of six experts, not direct measurement of children’s behaviour or development. The analysis was guided by pre-specified frameworks and was therefore primarily deductive, which can steer attention toward confirming instances. Participants were selected purposively and through the research team’s networks, so the strong convergence reported may partly reflect a selection of like-minded specialists rather than a consensus that would survive a more heterogeneous sample, and the perspectives of frontline teachers, parents and children themselves were not represented. The study is best understood as generative rather than confirmatory, and the authors call for quasi-experimental or longitudinal designs, case studies in kindergartens and families, and child-centred research using play-based interviews and children’s drawings to test whether the tensions these experts perceive hold up under empirical scrutiny.
Subject of Research: How AI's simulated agency may reshape the ecological conditions supporting authentic agency development in early childhood
Article Title: The illusion of agency: deconstructing AI’s influence on the generation of authentic agency in early childhood
Article References: Luo, W., Gao, M., Liu, J., Berson, I., Berson, M., & Li, H. (2026). The illusion of agency: deconstructing AI’s influence on the generation of authentic agency in early childhood. AI, Brain and Child, 2(1), Article 17. https://doi.org/10.1007/s44436-026-00040-8
Image Credits: AI Generated
DOI: 10.1007/s44436-026-00040-8
Keywords: artificial intelligence, early childhood education, child development, agency, actor-network theory, expert interviews, pedagogy of friction, AI literacy, learning ecosystems, qualitative research, developmental psychology, educational technology
News Source: Glenn Wilkins. (October 28, 2026). AI’s ‘illusion of agency’ may quietly reshape how young children build real autonomy, experts warn. Scienmag.
