Shaping early childhood in the age of AI: holistic competencies, pedagogical practices, and ecological support

Synthesis: Pei (2026) asks what children aged 0 to 8 should learn, and how they should be taught, once AI is present in early childhood settings, and answers with a systematic conceptual mapping review of 86 resources searched between 27 May 2024 and 20 June 2026 and coded against Bronfenbrenner's ecological systems. Four themes carry the answer: a five-domain set of holistic competencies in which AI Literacy sits alongside executive functions, physical and mental Well-Being, learning dispositions, and foundational cognitive and social-emotional skills; AI Governance frameworks that treat young children as a distinct population rather than small adults; professional development for a pre-primary workforce that is 94% female worldwide; and play-based, human-centered Pedagogies and Teaching Strategies for the early years of K-12. The central claim is that AI literacy "cannot be standalone" — it draws on working memory, inhibitory control, curiosity, and persistence that early childhood is uniquely positioned to build — and that the binding constraint on integration is infrastructure rather than curriculum content, since classroom practice, educator confidence, and policy safeguards have to move together.

Key Findings

1. Four themes, five domains. The 86 resources yield holistic competencies, governance, educator professional development, and play-based human-centered practice, and the competencies themselves resolve into five interdependent domains rather than a skills checklist.
2. AI literacy is nested, not foundational. AI literacy sits alongside executive functions, well-being, learning dispositions, and foundational skills; it "cannot be standalone," the review argues, because knowledge, attitudes, and capability depend on those domains.
3. Most of the corpus is not empirical. Of 86 resources, 26 were empirical studies (30.2%), 28 conceptual or review papers (32.6%), 14 scholarly books (16.3%), and 18 reports and policy documents (20.9%).
4. Executive functions do the coordinating work. Early inhibitory control was linked to adult health, income, criminal behavior, and subjective well-being in a 32-year study of over 1,000 children, with ages 3 to 5 a sensitive period of rapid EF development.
5. Prohibition backfires. Governance frameworks from Australia, China, the U.S., and the OECD converge on balancing innovation with safeguards, and the review reports that banning generative AI may be counterproductive by increasing unregulated use.
6. The workforce is 94% female. Educators' dual burden of integration and digital literacy gaps shapes the professional development theme, where co-design and design-based research are proposed over superficial technical training.
7. Pedagogy decides what AI adds. In an 8-week curriculum, direct instruction with scaffolding improved knowledge and skills, cooperative play better supported positive attitudes, and children as young as four grasped basic AI concepts.

AI literacy inside a wider competency set

The review's answer to its first research question is a pentagon, not a ladder. Foundational literacy and numeracy, social-emotional strengths such as resilience and adaptability, physical and mental well-being in digital childhoods, executive functions, and AI Literacy are presented as mutually influential domains, held together by learning dispositions — curiosity, persistence, creativity — that the paper calls a mediating layer between internal capacity and external environment. Executive functions carry the most developmental weight: working memory, inhibitory control, and cognitive flexibility form an "air traffic control system," biologically grounded yet shaped by interaction, play, and contextual learning, underpinning the Critical Thinking that lets a child question fluent but biased output. AI literacy itself is summarized as three competencies — knowledge, skills, and attitudes — with a rival five-dimension framing covering safety, identity, attitude, cognition, and capability. Both sit downstream of the rest, which is why the review insists that Curriculum Design for young children start with development, not with the tool.

Governance for a population that treats the robot as a person

National and international frameworks now frame the governance question. Australia's Department of Education (2023) set out principles for generative AI in schools covering pedagogy, well-being, transparency, and equity; China's Ministry of Education (2025) folded AI literacy for school leaders into its Digital Education Strategic Action Plan; and the OECD (2025) called for child-centered AI Governance addressing privacy, bias, and equity. The review's distinctive addition is developmental: young children often treat robots and AI characters as social agents, turning a design question into a calibration problem, since over-trust and blurred human-machine boundaries are harder to correct at five than at fifteen. Add age bias and children's thin representation in AI systems, and the remedies converge on child-specific datasets, algorithmic transparency, and enforceable Privacy protection: instruments aimed at a population that cannot consent, complain, or opt out.

Ninety-four percent: the workforce the policy assumes

The third theme is the least glamorous and the most concrete. Female educators make up 94% of pre-primary teachers worldwide and carry the integration agenda alongside digital literacy gaps and workplace stress; the review names the gendered shape of that burden without pursuing it far. AI may reduce workload in curriculum design and material preparation, but the literature keeps returning to educators' digital confidence and critical engagement as what decides whether a tool is used well — more than technical training does — and research on teachers' AI literacy remains limited even as policy attention grows. The proposed remedy is participatory: involving educators as co-designers, and using design-based research to build interventions with teachers rather than for them, which the reviewed studies report strengthens Self-Efficacy and Learner Agency while keeping local values in view. The role shift that follows — from content expert to facilitator, co-designer, and ethical overseer — is where over-automation does its damage, narrowing the professional judgment that made the role worth having.

Play as the delivery mechanism

If the first three themes describe what to build, the fourth describes where it has to fit. The review treats play not as a reward for finished work but as the medium through which young children encounter AI at all: block-building apps for sustainable city design that connect STEM content to creative Problem Solving, Computational Thinking developed through embodied, tangible, and unplugged activities, and embodied play with guided dialogue aligned to the AI4K12 "Big Ideas." Robots recur as the practical entry point — PopBots showed young children can build AI understanding through interaction and programming, though with varying conceptual difficulty — and the 8-week curriculum sharpens the pedagogy: scaffolding and direct instruction carry knowledge and skills, cooperative play carries attitudes. The conclusion drawn is augmentative rather than substitutive. AI should support the relational, exploratory core of early childhood practice, including the empathy and Creativity that no system reproduces, and the review adds a practical caution for stretched teachers: unplugged activities cost less to adopt than another platform to learn.

What this means for practice

• Instructors. Keep AI inside play and hands-on work: the reviewed evidence favors tangible, embodied, and unplugged activity, with direct instruction and Scaffolding reserved for the conceptual parts that need it.
• Faculty developers. Design early childhood professional development as co-design, not tool training: the reviewed educators gained agency and confidence when they shaped the intervention, and technical training alone did not move practice.
• Administrators. Write policy for the youngest children as its own case: child-specific data rules, transparency requirements, and a stance on robot and character-based tools that anticipates over-trust rather than banning the category.
• Policymakers. Fund family-school alignment and participatory research together. The review locates the leverage in the mesosystem, where consistent guidance on ethical AI use either reinforces or undermines what happens in the classroom.
• Researchers. Move from mapping to empirical and design-based work across diverse sociocultural contexts, including low- and middle-income settings. That is the review's own recommendation for strengthening the evidence base.

Limitations

• The evidence base is conceptual: only 26 of the 86 resources were empirical studies (30.2%), against 28 conceptual and review papers, 14 scholarly books, and 18 reports and policy documents, so most claims on this page summarize positions rather than measured outcomes.
• Coding was done by a single researcher, and inter-coder reliability was not assessed; the author refined the framework with verbatim excerpts and coding memos and recommends multiple coders or an intra-coder check.
• Asia-Pacific, North American, and European sources dominate, with thin representation from the Middle East, Africa, and South America; the author names the predominance of high- and middle-income contexts as a limitation.
• The search window ran from 27 May 2024 to 20 June 2026, and the author flags the rapid evolution of AI technologies as a limit on how long the mapping will hold, since the tools it describes change faster than the literature about them.
• Themes are interpretive constructs from thematic analysis, not frequency counts, so the prominence of a theme on the page reflects the author's judgment about salience rather than how many sources supported it.

Citation

Pei, L. (2026). Shaping early childhood in the age of AI: holistic competencies, pedagogical practices, and ecological support. Frontiers in Psychology, 17, 1995020.