1. Introduction and problem

Early childhood education—kindergarten, preschool, or nursery education, depending on the country—faces growing institutional pressure to “integrate artificial intelligence,” often in the form of tool workshops, app catalogues, and promises of personalization. UNESCO’s AI competency framework for teachers defines fifteen competencies across five dimensions (human-centred mindset, ethics of AI, foundations and applications, AI pedagogy, and AI for professional learning) and three progression levels (acquire, deepen, create) (Miao & Cukurova, 2024). The companion framework for students proposes twelve competencies in four dimensions, with levels of understand, apply, and create (Miao, Shiohira, & Lao, 2024). Both documents are normative reference frameworks, not clinical trials or impact evaluations of development in children aged three to six. Confusing those categories is precisely the problem this article examines.

The 2023 Global Education Monitoring Report warns that technology should complement, not replace, face-to-face interaction with teachers, and that decisions about its use must be appropriate, equitable, evidence-based, and sustainable (UNESCO, 2023). Volume VII of OECD’s Starting Strong series locates digitalisation of early childhood education and care as a challenge of quality, equity, and workforce development, not as a race to adopt devices (OECD, 2023). UNESCO’s guidance on generative AI proposes a human-centred approach, data protection, and an age limit for independent conversations with generative-AI platforms (Miao & Holmes, 2023). None of these instruments claims that AI, by itself, improves child development in kindergarten.

The thesis of this paper is restrictive. UNESCO’s teacher-competency framework does not transfer to kindergarten as a tools workshop. It must be translated into languages the profession already possesses: play as a medium of learning, observation and documentation as forms of pedagogical knowledge, care as an ethical relation, and inclusion as a criterion of justice. That translation is not a rhetorical exercise: it requires anchoring each dimension of the framework in empirical evidence from interventions and expert frameworks, and marking clearly what is a finding, what is a norm, and what is the author’s inference. Where a source does not claim an effect on development, this article does not promise one.

The problem is compounded because the AI-literacy literature has concentrated on primary, secondary, and higher education (Su, Ng, & Chu, 2023; Ng et al., 2021). In early childhood education, scoping reviews document a small corpus, methodological heterogeneity, and a recurring gap: lack of teachers’ knowledge, skills, and confidence; absence of curriculum design; and scarcity of teaching guidelines (Su & Yang, 2022; Su, Ng, & Chu, 2023). A perception study with fifteen kindergarten teachers confirms those obstacles in professional voice: lack of school support, of teacher knowledge of AI, of curricular guidelines, and, in interviewees’ interpretation, of children’s “comprehensive ability” (Su, 2024). That last statement is a finding about teacher perceptions, not evidence of a cognitive limit in children; the distinction matters.

This article offers three contributions: to reconstruct the state of the art of that competence in kindergarten; to analyse three verified cases—an AI-literacy intervention, an expert framework on ChatGPT, and teacher perceptions—distinguishing levels of evidence; and to offer an operational translation of UNESCO’s five dimensions into the kindergarten lexicon, compatible with developmentally appropriate practice (NAEYC, 2020), with WHO and AAP screen limits (World Health Organization, 2019; Council on Communications and Media, 2016), and with child-centred AI requirements (UNICEF, 2021).

2. State of the art

It is useful to separate three strata that the literature often mixes. The first is the empirical stratum: interventions, case studies, interviews, and classroom observations in early childhood settings. The second is the framework stratum: definitions of AI literacy, curricular models, expert frameworks, and policy frameworks. The third is the synthesis stratum: scoping and systematic reviews that aggregate heterogeneous studies. Each stratum authorises different claims.

In the conceptual-framework stratum, Ng, Leung, Chu, and Qiao (2021) proposed four aspects of AI literacy from thirty articles: know and understand, use and apply, evaluate and create, and ethical issues. Yang (2022) formulated, from early childhood education, a pedagogical model of why, what, and how: AI literacy as an organic part of digital literacy; an accessible conceptual core (with large volumes of data, algorithms can be trained to identify patterns, make predictions, and recommend actions, with limitations); and an embodied, culturally responsive approach based on learning-by-making and pedagogy-as-relation. That text is cited here as a curricular framework, not as a report of a particular conversational-robot intervention. Luo, He, Gao, and Li (2024b) constructed, through interviews with seven Chinese experts and grounded theory, the SIACC framework—safety, identity, attitude, cognition, and capability—as a culturally situated proposal for early childhood AI literacy. These frameworks guide design; they do not measure development.

In the empirical stratum, Su and Yang (2024) evaluated the eight-week AI4KG programme with 26 children of average age four in a Hong Kong kindergarten and reported learning of basic AI concepts, thematic creativity, and differentiated perceptions of activities. Yang, Hu, Yeter, Su, Yang, and Lee (2024) documented, over six weeks with six five-year-olds (mean age 62.83 months), that children could learn about AI through intelligent agents in embodied-learning contexts. Both studies describe mediated learning in small samples; neither establishes that AI improves socioemotional, linguistic, or motor development. Su (2024), interviewing fifteen teachers, found four challenges (lack of school support, a perception of insufficient comprehensive ability in children, insufficient teacher knowledge, and lack of curricular guidelines) and three enablers (government, school, and social demand); more than half considered AI literacy crucial. Luo, He, Liu, Berson, Berson, Zhou, and Li (2024a) interviewed six experts from China and the United States on ChatGPT: they identified two roles (conversational agent for children and on-call facilitator for educators and caregivers), classified risks with the 3A2S framework (accessibility, affordability, accountability, sustainability, and social justice), and projected the tool as intelligence augmentation, not as a replacement for educators and caregivers.

In the synthesis stratum, Su and Yang (2022) reviewed 17 studies from 1995 to 2021 and concluded that most reported improvements in concepts of AI, machine learning, computing, and robotics, and in skills such as creativity, emotion control, collaborative inquiry, literacy, and computational thinking. That claim is a synthesis of a small, heterogeneous corpus, not a meta-analysis of effects. Su, Ng, and Chu (2023) reviewed 16 studies from 2016 to 2022 and underlined three challenges: lack of teachers’ knowledge, skills, and confidence; lack of curriculum design; and lack of teaching guidelines. Berson, Berson, and Luo (2025) examined 42 ethics studies and grouped concerns into data privacy, developmental impacts, algorithmic bias, and regulation; they found gaps in protecting children’s data and warned that emotional AI (social robots, emotion recognition) may offer learning opportunities and, at the same time, risks of overreliance, manipulation, or loss of autonomy if design is not developmentally appropriate. Zhang, Binti Halili, and Zainuddin (2026) analysed 21 studies of generative AI in preschool with a SWOT frame: opportunities for personalisation, teacher collaboration, and equity; threats of reliability and age-appropriateness, digital competence, possible reduction of creativity, privacy, and unequal access. The abstract of that review infers transformative potential; this article retains the SWOT findings and does not adopt the inference of “revolution.”

Policy and professional-practice frameworks complete the map. UNESCO’s Recommendation on the Ethics of Artificial Intelligence (UNESCO, 2021) anchors human rights, dignity, and human oversight. Miao and Holmes (2023) specify ethical and pedagogical validation, data protection, and an age threshold for independent use of conversational platforms (the Guidance indicates a minimum of 13 years, with reference to the GDPR threshold of 16). That normative threshold has a direct consequence: an independent conversation between a four-year-old and a generative model is not aligned with the Guidance. UNICEF (2021) formulates requirements for child-centred AI: support development and well-being, ensure inclusion, prioritise equity and non-discrimination, protect data and privacy, ensure safety, and prepare children for a world with AI. NAEYC and the Fred Rogers Center (2012) hold that technology can support learning when used intentionally within developmentally appropriate practice, and that it should not replace creative play, outdoor activity, or face-to-face interaction. NAEYC (2020) redefines that practice as a strengths-based, play-based approach, with observation, documentation, equitable care, and professionalism. WHO recommends, for children aged 3 to 4 years, no more than one hour of sedentary screen time, and preferably less (World Health Organization, 2019). The AAP limits, for ages 2 to 5, screen use to one hour per day of high-quality content, with adult co-viewing, and advises against screen media—other than video-chatting—under 18 months (Council on Communications and Media, 2016). These instruments are not AI studies; they are environmental constraints that kindergarten teacher competence must internalise.

The gap organising this article is twofold. UNESCO’s teacher-competency framework is written for the education system in general and does not specify the translation to kindergarten. Available empirical evidence comes from small samples, with East Asian urban contexts in several of the most cited studies, and with designs that do not authorise causality about development. Professional competence cannot be filled with technological enthusiasm.

3. Review method

A critical narrative review was conducted, not a meta-analysis. The purpose was not to estimate an effect size—none is comparably available in this field—but to articulate an argument of professional translation with verified sources. Inclusion criteria were: (a) publication between 2021 and 2026, with a justified exception for still-current practice frameworks (NAEYC & Fred Rogers Center, 2012; Council on Communications and Media, 2016; World Health Organization, 2019); (b) focus on early childhood education (ages 0–8, priority 3–6) or on teacher competence applicable at that level; (c) document type of peer-reviewed journal, international-organisation report, or professional position from NAEYC, AAP, WHO, UNESCO, OECD, or UNICEF; (d) access to a DOI page, institutional repository, or publisher site confirming authors, year, title, and findings. Sources whose DOI did not open to verifiable metadata, unreviewed opinion essays, and cases already used as the axis of the 2026-08-18 article (PopBots-type interventions, Australian robotic toys in 2020, and Druga et al. studies on Alexa) were excluded, so as not to repeat that angle.

The search was executed in August 2026 on DOI pages, ERIC, repositories (EdUHK, UCL Discovery), UNESCO, OECD, UNICEF Innocenti, ScienceDirect, Springer Nature, and Contemporary Educational Technology. Each cited source was verified against at least one of those pages. The corpus was organised into core cases, policy and professional frameworks, and syntheses. Cases were chosen to complement types of evidence: an intervention with children (Su & Yang, 2024), an expert study (Luo et al., 2024a), and a teacher study (Su, 2024). For saturation, the embodied-learning case (Yang et al., 2024), the SIACC framework (Luo et al., 2024b), and the reviews by Su and Yang (2022), Su, Ng, and Chu (2023), Berson et al. (2025), and Zhang et al. (2026) were retained.

The analysis distinguished three enunciative statuses. Empirical finding: what was observed or measured in the sample. Normative framework: what an organisation or expert framework prescribes. Pedagogical inference: the translation this article proposes for kindergarten, marked as such. The limits are those of any narrative review: there is no complete PRISMA protocol, there is an English-language bias, and Latin America, Africa, and rural contexts are under-represented (section 9).

4. Case 1. An AI-literacy intervention in a Hong Kong kindergarten (AI4KG)

Su and Yang (2024) published in Interactive Learning Environments (Vol. 32, No. 9, pp. 5494–5508; advance publication in 2023) an eight-week intervention with 26 children of average age four enrolled in a Hong Kong kindergarten. The stated aim was to evaluate the impact of the AI4KG programme on young children’s AI literacy, AI-related creativity, and perceptions of the programme. According to the authors, (1) young children were capable of learning basic AI concepts and knowledge; (2) in AI-related creativity, younger children designed a chatting robot via imagination, while older children created an AI robot to assist people in their drawings; (3) regarding perceptions of the programme, older children who enjoyed the activity were able to train AI, while younger children preferred to draw the future AI city and participate in the AI story activity. The authors interpret the whole as a positive benefit of AI-literacy education in preparing children for an AI-driven future.

Status of the evidence. Empirical finding: in that sample, with that programme and duration, learning of basic concepts, thematic creative productions, and age-differentiated activity preferences were observed. It is not an empirical finding, in this article, that AI improves global development, language, self-regulation, or school readiness: the study does not measure those outcomes. The phrase “preparing for an AI-driven future” is the authors’ inference, not a result of longitudinal follow-up. The sample is small, from a single centre, with no control group reported in the verified abstract, and is situated in a high-connectivity context. Its value for teacher competence is not that of an exportable recipe, but of an existence proof: with adult mediation and activity design (drawing, storytelling, imagining, “training”), children around four can take part in incipient literacy. That proof authorises teachers not to underestimate children’s inquiry; it does not authorise them to install a laboratory of generative models in the classroom.

Pedagogical inference of this article, marked as such: the competent teacher does not “deliver AI,” but designs play situations in which children explore data, pattern, error, and automatic help, and documents what they understand and confuse. AI4KG is an exemplar of mediation, not evidence of technological superiority.

5. Case 2. Experts from China and the United States: ChatGPT, roles, and the 3A2S framework

Luo et al. (2024a) published in Early Education and Development (Vol. 35, No. 1, pp. 96–113) an inductive and deductive content analysis of interviews with six expert professors from China and the United States. The empirical finding—here, empirical of expert discourse, not of the classroom—is twofold. As to roles, ChatGPT is profiled as (1) a conversational agent for young children and (2) an on-call facilitator for educators and caregivers. As to risks, experts organised them with the 3A2S framework: Accessibility, Affordability, Accountability, Sustainability, and Social Justice. Participants suggest that as knowledge, information retrieval, and integration skills become less critical, human agency and inquiry ability gain weight. Consequently, they project that ChatGPT’s future trajectory in early childhood education should be that of a resource for intelligence augmentation, not a replacement for the actions of educators and caregivers. They add that AI opens new domains, such as AI literacy and AI social interaction.

Status of the evidence. This is a study of six experts, not an evaluation of ChatGPT with children. The 3A2S framework is an analytic categorisation proposed in the article, useful as a language of policy and institutional oversight, not as a validated scale. The thesis of augmentation versus replacement coincides with the human-centred approach of Miao and Cukurova (2024) and of Miao and Holmes (2023), and with the GEM Report’s warning not to supplant human connection (UNESCO, 2023). It is therefore a convergence of frameworks and expert discourse, not proof of ChatGPT’s pedagogical efficacy in kindergarten.

Pedagogical inference of this article: the second role—on-call facilitator for the teacher—is the only one that, in light of Miao and Holmes (2023) and age limits, is a priority in kindergarten. Using a generative model to draft a note to families, explore variations of a play project, or anticipate inclusion questions may belong to teachers’ professional learning (dimension 5 of the UNESCO framework), provided that identifiable child data are not uploaded and that the teacher retains judgement. The first role—conversational agent with the child—collides with the age threshold for independent conversations and with co-viewing and screen-time recommendations (Miao & Holmes, 2023; Council on Communications and Media, 2016; World Health Organization, 2019). A competent teacher distinguishes the two roles and does not fuse them into a “classroom assistant.”

6. Case 3. Kindergarten teachers’ perceptions of AI literacy

Su (2024) published in the International Journal of Technology and Design Education (Vol. 34, No. 5, pp. 1665–1685) a qualitative study based on individual interviews with fifteen teachers. The aim was to identify their views on the importance of AI literacy in kindergarten and on challenges and enablers of promoting it. Empirical findings (perceptions): four main challenges—lack of school support, lack of children’s “comprehensive ability,” insufficient teacher knowledge of AI, and lack of curriculum guidelines; three enablers—government support, school support, and social needs; and that more than half of the teachers in the study considered that literacy crucial for kindergarten children. The article proposes an AI-literacy policy framework for young children with three dimensions: governance, pedagogical, and operational and management.

Status of the evidence. This is a perception study, not classroom observation or programme evaluation. The statement about children’s “comprehensive ability” is data about professional beliefs; it partly contradicts the findings of Su and Yang (2024) and of Yang et al. (2024), where four- and five-year-olds did take part in learning basic concepts under mediation. That tension is itself an object of teacher competence: underestimating the child and overestimating the tool are two faces of the same training deficit. The three-dimension policy framework (governance, pedagogy, operations) is a proposal by the author, not an international standard; it can be placed in dialogue with the UNESCO framework without replacing it.

This case is the closest to the problem of this article. Su, Ng, and Chu (2023) had already identified, in a review of sixteen studies, lack of knowledge, curriculum, and teaching guidelines as systemic challenges. Su (2024) shows that those gaps are verbalised in the kindergarten room. OECD (2023) includes workforce development as a lever of response to digitalisation in early childhood education and care. The pedagogical inference is direct: a “ChatGPT workshop” does not answer what teachers say they need. They need school support, guidelines that speak the language of play and care, and enough AI knowledge to decide, not to operate an application.

7. Translating the UNESCO framework into kindergarten

Miao and Cukurova (2024) organise fifteen competencies in five dimensions and three levels (acquire, deepen, create). The student framework (Miao, Shiohira, & Lao, 2024) does not apply literally to kindergarten children: its levels of “creating AI systems” exceed, in their school formulation, what Yang (2022) considers the accessible conceptual core in early childhood. The translation that follows is a pedagogical inference of this article, anchored in the cases and verified frameworks. It is not a new standard; it is a working hypothesis for continuing professional development.

7.1. Human-centred mindset → care, agency, and non-substitution. In kindergarten, this dimension translates as defence of the adult–child relation as irreplaceable. The GEM Report formulates it as a system principle: technology should not supplant human connection (UNESCO, 2023). Luo et al. (2024a) formulate it as augmentation, not replacement. NAEYC (2020) formulates it as a caring, equitable community of learners. To acquire, at this level, is to be able to name what is not delegated to a system: comfort, conflict mediation, observation of play, and decisions about well-being. To deepen is to argue before families and management why a “conversational tutor” is not a classroom aide. To create is to design centre protocols that make non-substitution explicit.

7.2. Ethics of AI → privacy, consent with families, data minimisation, and inclusion. Berson et al. (2025) document gaps in safeguarding children’s data (breaches, profiling, misuse) and algorithmic bias that reproduces inequities. UNICEF (2021) requires data protection, inclusion, and non-discrimination. Miao and Holmes (2023) require ethical and pedagogical validation of tools and an age limit for independent conversations. In kindergarten, to acquire is to know that photographs, voices, and traces of children are sensitive data and are not uploaded to generative-AI platforms. To deepen is to talk with families in the language of care. To create is to take part in lists of banned and permitted tools, with data minimisation. The 3A2S framework (Luo et al., 2024a) supplies justice questions: who accesses, who pays, who is accountable, what is sustained, who is excluded?

7.3. AI foundations and applications → a conceptual core in the service of observation, not an engineering syllabus. Yang (2022) offers the core: data, patterns, predictions, recommendations, and limits. Ng et al. (2021) recall that literacy is not only use. Su and Yang (2024) show that this core can be explored at age four through drawing, story, and imagining robots. To acquire, for the teacher, is to understand at a functional level what a model is, what a bias is, and why a system “gets it wrong.” To deepen is to recognise AI embedded in toys, documentation apps, and translators, and to decide its place. To create does not mean to programme: it means to design a play sequence in which children experience that a machine “learns” from examples and can fail, and to document children’s hypotheses. The SIACC framework (Luo et al., 2024b) recalls that safety and identity precede cognition and capability: one does not begin with the technical skill.

7.4. AI pedagogy → play, observation, and documentation, not an “AI hour.” NAEYC and the Fred Rogers Center (2012) require intentional use within developmentally appropriate practice. NAEYC (2020) places observation, documentation, and assessment at the centre of the craft. Yang et al. (2024) illustrate embodied learning mediated by intelligent agents, with classroom observation, teacher interviews, and artefacts, in a minimal sample. The translation is: AI pedagogy in kindergarten is play pedagogy with one more object—a system that responds—not a subject. To acquire is to know when to switch off. To deepen is to scaffold questions (“who taught the machine that?”, “what can it not do?”) without yielding the speaking turn to the system. To create is to integrate those questions into existing projects (the market, the hospital, the forest), not to invent a technology corner disconnected from the play curriculum. Zhang et al. (2026) warn, as a weakness reported in their corpus, of the reliability and age-appropriateness of generative content and the risk of reduced creativity: that SWOT warning reinforces the primacy of children’s making over machine-generated content.

7.5. AI for professional learning → communities of practice and judgement, not consumption of tools. Miao and Cukurova (2024) include this dimension as professional growth. OECD (2023) locates it as a system lever. Su (2024) shows that teachers ask for school support and guidelines, not only apps. Inference: legitimate use of generative AI in kindergarten is, first of all, teacher back-office—planning, communication with families, exploration of materials—with a veto on identifiable data, and critical evaluation of outputs. To acquire is to know risks of hallucination and bias. To deepen is to contrast an activity suggestion with knowledge of the concrete group. To create is to produce, with colleagues, local quality criteria (does this displace free play? does this respect WHO screen time? does this include children with disabilities and the languages of the classroom?).

The student framework (Miao, Shiohira, & Lao, 2024) translates, in kindergarten, in a still more restrictive way. “Understand” may amount to distinguishing person and machine in play. “Apply” and “create” do not mean that a four-year-old should design systems; they mean, if anything, that the child can imagine, draw, and dictate rules for a “helper robot,” as in AI4KG (Su & Yang, 2024). Forcing school levels onto kindergarten is a bad translation.

8. Discussion

Three tensions organise the discussion. The first is the tension between empirical feasibility and a promise of development. It is a finding that, under brief, mediated conditions, kindergarten children can learn basic AI concepts and produce imaginative artefacts (Su & Yang, 2024; Yang et al., 2024). It is not a finding that this improves their development. Reviews that report “improvements” in creativity, emotion control, or literacy (Su & Yang, 2022) aggregate heterogeneous studies spanning almost three decades; transferring that synthesis as evidence that “AI develops the child” is an unwarranted leap. Teacher competence includes resisting that leap in staff meetings, innovation fairs, and licence purchases.

The second tension is between the child as AI learner and the teacher as the professional who decides. Su (2024) and Su, Ng, and Chu (2023) agree that the bottleneck is the teacher: knowledge, confidence, curriculum, support. Luo et al. (2024a) shift the centre of gravity toward human agency. The UNESCO framework (Miao & Cukurova, 2024) is written, precisely, for that professional. Bad implementation consists in skipping the teacher and handing the room to the system. Good implementation consists in forming judgement: when not to use, how to mediate, how to document, how to talk with families. That judgement is exercised in the languages of the craft—the observation note, the documentation panel, the inclusion meeting—not in a certificate of prompts.

The third tension is between equity and adoption. The GEM Report (UNESCO, 2023) and OECD (2023) recall that the most disadvantaged are typically left outside the benefits and inside the costs. 3A2S (Luo et al., 2024a) and Berson et al. (2025) turn that intuition into criteria: accessibility, affordability, bias, and fragmented regulation. Zhang et al. (2026) add unequal access and privacy risks. A competence that ignores inclusion is not competence under the UNESCO framework, which declares inclusivity, human agency, non-discrimination, and linguistic and cultural diversity (Miao & Cukurova, 2024). In kindergarten, inclusion means not introducing a system that only “understands” one language, not photographing without family consent, and not replacing shared play with individual turns in front of a screen.

There is, in addition, a regulatory consequence. If Miao and Holmes (2023) set an age threshold for independent conversations with generative AI, the four-year-old classroom is not the place for that conversation. The teacher may use generative AI outside direct interaction with the child; the child remains under human mediation, co-viewing, and screen limits (Council on Communications and Media, 2016; World Health Organization, 2019). That distribution of roles is the most austere—and the most faithful—translation of the framework.

9. Limits

This review is narrative. It does not apply a complete PRISMA protocol or estimate effects. The empirical corpus of core cases is small and geographically skewed: Hong Kong and interviews with teachers and experts with a strong presence of China and the United States. Controlled Latin American or African trials of teacher AI competence in kindergarten were not located with the same degree of DOI openness; that absence is a gap, not proof of non-existence. Several Taylor & Francis and Springer articles restricted full text; work proceeded from publisher abstracts, ERIC, institutional repositories, and DOI pages, which limits methodological granularity. Scoping reviews (Su & Yang, 2022; Su, Ng, & Chu, 2023; Berson et al., 2025; Zhang et al., 2026) inherit the heterogeneity of their corpora. UNESCO, OECD, and UNICEF frameworks are prescriptive: their authority is normative, not empirical. The pedagogical inferences of this article are hypotheses for continuing professional development, not classroom evidence.

10. Conclusions

The AI competency framework for teachers (Miao & Cukurova, 2024) is necessary and insufficient. It is necessary because it names dimensions—humanity, ethics, foundations, pedagogy, professional learning—that kindergarten cannot improvise. It is insufficient because, without translation, it degrades into a tools workshop. Verified empirical evidence shows three modest things. First: children around four and five can learn basic AI concepts in mediated, brief, situated programmes (Su & Yang, 2024; Yang et al., 2024). Second: experts reject replacement of the educator and require accessibility, affordability, accountability, sustainability, and social justice (Luo et al., 2024a). Third: teachers identify as the obstacle not the lack of an app, but the lack of knowledge, institutional support, and guidelines in their professional language (Su, 2024; Su, Ng, & Chu, 2023).

The proposed translation anchors each UNESCO dimension in the kindergarten lexicon: care and non-substitution; privacy and consent with families; a conceptual core in the service of observation; play, documentation, and intentional switch-off; and professional learning with a veto on children’s data. That translation is compatible with developmentally appropriate practice (NAEYC, 2020; NAEYC & Fred Rogers Center, 2012), with screen limits (World Health Organization, 2019; Council on Communications and Media, 2016), with child-centred AI (UNICEF, 2021), and with the global warning not to let technology dictate the terms of education (UNESCO, 2023; OECD, 2023). Where sources do not demonstrate improved development, this article does not claim it. Teacher competence in artificial intelligence for early childhood education is not prompt skill: it is professional judgement about play, care, and inclusion in an environment saturated with systems that predict, recommend, and, often, get it wrong.

Ingeniero Mitre / Laboratorio Editorial de NEXTECH.IA

References

  1. Berson, I. R., Berson, M. J., and Luo, W. (2025). Innovating responsibly: Ethical considerations for AI in early childhood education. AI, Brain and Child, 1, Article 2. https://doi.org/10.1007/s44436-025-00003-5
  2. Council on Communications and Media. (2016). Media and young minds. Pediatrics, 138(5), e20162591. https://doi.org/10.1542/peds.2016-2591
  3. Luo, W., He, H., Gao, M., and Li, H. (2024b). Safety, identity, attitude, cognition, and capability: The ‘SIACC’ framework of early childhood AI literacy. Education Sciences, 14(8), 871. https://doi.org/10.3390/educsci14080871
  4. Luo, W., He, H., Liu, J., Berson, I. R., Berson, M. J., Zhou, Y., and Li, H. (2024a). Aladdin’s genie or Pandora’s box for early childhood education? Experts chat on the roles, challenges, and developments of ChatGPT. Early Education and Development, 35(1), 96–113. https://doi.org/10.1080/10409289.2023.2214181
  5. Miao, F., and Cukurova, M. (2024). AI competency framework for teachers. UNESCO. https://doi.org/10.54675/ZJTE2084
  6. Miao, F., and Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  7. Miao, F., Shiohira, K., and Lao, N. (2024). AI competency framework for students. UNESCO. https://doi.org/10.54675/JKJB9835
  8. National Association for the Education of Young Children. (2020). Developmentally appropriate practice (DAP) position statement. NAEYC. https://www.naeyc.org/resources/position-statements/dap
  9. National Association for the Education of Young Children, and Fred Rogers Center for Early Learning and Children’s Media. (2012). Technology and interactive media as tools in early childhood programs serving children from birth through age 8 [Joint position statement]. NAEYC. https://www.naeyc.org/sites/default/files/globally-shared/downloads/PDFs/resources/position-statements/ps_technology.pdf
  10. Ng, D. T. K., Leung, J. K. L., Chu, S. K. W., and Qiao, M. S. (2021). Conceptualizing AI literacy: An exploratory review. Computers and Education: Artificial Intelligence, 2, 100041. https://doi.org/10.1016/j.caeai.2021.100041
  11. OECD. (2023). Empowering young children in the digital age (Starting Strong). OECD Publishing. https://doi.org/10.1787/50967622-en
  12. Su, J. (2024). Kindergarten teachers’ perceptions of AI literacy education for young children. International Journal of Technology and Design Education, 34(5), 1665–1685. https://doi.org/10.1007/s10798-024-09876-8
  13. Su, J., Ng, D. T. K., and Chu, S. K. W. (2023). Artificial intelligence (AI) literacy in early childhood education: The challenges and opportunities. Computers and Education: Artificial Intelligence, 4, 100124. https://doi.org/10.1016/j.caeai.2023.100124
  14. Su, J., and Yang, W. (2022). Artificial intelligence in early childhood education: A scoping review. Computers and Education: Artificial Intelligence, 3, 100049. https://doi.org/10.1016/j.caeai.2022.100049
  15. Su, J., and Yang, W. (2024). Artificial intelligence (AI) literacy in early childhood education: An intervention study in Hong Kong. Interactive Learning Environments, 32(9), 5494–5508. https://doi.org/10.1080/10494820.2023.2217864
  16. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137
  17. UNESCO. (2023). Global education monitoring report 2023: Technology in education — A tool on whose terms? UNESCO. https://doi.org/10.54676/uzqv8501
  18. UNICEF. (2021). Policy guidance on AI for children 2.0. UNICEF. https://www.unicef.org/innocenti/reports/policy-guidance-ai-children
  19. World Health Organization. (2019). Guidelines on physical activity, sedentary behaviour and sleep for children under 5 years of age. WHO. https://iris.who.int/handle/10665/311664
  20. Yang, W. (2022). Artificial intelligence education for young children: Why, what, and how in curriculum design and implementation. Computers and Education: Artificial Intelligence, 3, 100061. https://doi.org/10.1016/j.caeai.2022.100061
  21. Yang, W., Hu, X., Yeter, I. H., Su, J., Yang, Y., and Lee, J. C.-K. (2024). Artificial intelligence education for young children: A case study of technology-enhanced embodied learning. Journal of Computer Assisted Learning, 40(2), 465–477. https://doi.org/10.1111/jcal.12892
  22. Zhang, Y., Binti Halili, S. H., and Zainuddin, Z. (2026). Generative AI in preschool education: A systematic review with SWOT analysis. Contemporary Educational Technology, 18(1), ep626. https://doi.org/10.30935/cedtech/17866