While initiatives such as MIT’s DAILy curriculum, the AI4K12 initiative with its “Five Big Ideas in AI,” and ISTE’s hands-on AI projects have brought AI learning into K-12 settings, a new study argues that the youngest learners have been largely left behind. Published in the journal AI, Brain and Child, the research proposes a comprehensive pedagogical framework for introducing AI literacy to children aged four to six, a group that includes many preliterate learners for whom existing curricula were never designed.

The study, led by Yufeng Qian, responds to what the author describes as a pressing need: educators currently lack an instructionally actionable framework that specifies both what to teach and how to teach it for young children. A feature in The Hechinger Report has noted that, unlike in K-12 and higher education, AI has yet to get a foothold in early childhood education. An international scoping review by Su and colleagues, synthesizing sixteen empirical studies from 2016 to 2022, reached a similar conclusion, finding that AI literacy remains underexplored in early childhood settings and highlighting persistent barriers such as the absence of curriculum design guidelines and limited teacher AI knowledge and confidence.

What makes AI literacy distinct from the coding and digital literacy frameworks already in circulation is the nature of the technology itself. Many coding experiences emphasize rule-based, deterministic control, but AI literacy requires children to grasp that computers can learn from data through statistical inference, finding patterns in training examples, and that the choice and balance of training data can shape system behavior and introduce bias. AI literacy also foregrounds socio-technical perspectives, such as evaluating outputs and recognizing limitations, because AI systems can generate incorrect information and perpetuate bias, making human judgment central to deciding when and how AI should be used.

To build the framework, the study used a pedagogical framework-building approach, conducting a systematic synthesis of relevant learning theories, existing AI literacy frameworks, and empirical studies on early AI learning. A targeted scoping review of the literature, using keyword searches and backward and forward snowballing, identified twenty relevant AI literacy frameworks that met the inclusion criteria. Cross-framework thematic coding of these materials revealed three recurring emphases: core AI ideas to understand, what learners do with and around AI systems, and the perspectives and dispositions learners develop. From this analysis, a Concepts-Practices-Perspectives structure emerged as the organizing backbone of the framework.

The first pillar, AI concepts, distills the field into three interdependent ideas that young children can learn: patterns, decision-making, and feedback. Patterns mean that an AI system looks for sameness and difference across examples and uses those similarities to group or label new things, an idea that builds on familiar early-childhood practices such as sorting, matching, and categorizing. Decision-making covers how an AI system chooses what to do next, introduced through two accessible pathways: pattern-based decisions, where the system learned from examples, and rule-based decisions, where if-this-then-that logic applies. Feedback captures the understanding that AI performance can change through cycles of input, output, evaluation, and adjustment, so that children see AI as something that can improve when people try something, look at what happens, and adjust what they provide next.

These concepts were drawn from a focused synthesis of twelve empirical studies that implemented and evaluated AI-related learning experiences with young children, with each study screened for explicit evidence that children engaged with and were assessed on a given concept. Examples cited in the synthesis include the AI4KG intervention, an eight-week curriculum with kindergarteners in Hong Kong that used tools such as AI for Oceans, Teachable Machine, and Quick, Draw! to let children experience how AI outputs depend on training and testing data. The PopBots project, where children label and test examples with a robot learning companion, illustrates how learning from examples becomes visible and discussable, while a family-based study by Druga and colleagues showed children and parents training their own object-recognition models with a smartphone app and predicting how the model would classify new cases.

The second pillar, AI practices, is anchored in computational thinking, which the study frames as a transferable problem-solving approach that can be enacted with or without computers. Drawing on an umbrella review of thirteen review papers and nine empirical studies of young children’s computational thinking, the framework foregrounds four skills considered feasible for ages four to six: decomposing, abstracting, sequencing, and debugging. Decomposing means breaking a complex goal into smaller, manageable parts; abstracting means using a small set of symbols, such as an arrow block that stands for “move forward,” to represent actions; sequencing means ordering steps precisely to reach a goal; and debugging means identifying why a solution does not work as intended and fixing it. Evidence cited ranges from Bee-Bot navigation studies and Matatalab programming interventions to comparisons of the tangible KIBO robotics kit with the on-screen ScratchJr programming environment.

The third pillar, AI perspectives, addresses the dispositions children develop for relating to AI critically and responsibly. The instructional goal at this age is not formal ethics or policy reasoning but an accurate, non-magical stance built through talk, reflection, and guided inquiry. The AI4KG curriculum built perspective outcomes into its design with play-based modules titled “Is AI always right?” and “Prejudice of AI,” while PopBots deliberately included moments where the robot signaled uncertainty and made a guess, supporting the child-friendly insight that AI outputs are choices produced by a system and sometimes the system does not know. The study notes that developmental research shows children begin judging unfair treatment as wrong as early as around age three, providing a realistic entry point for early ethics learning, and suggests teaching bias as an unfair pattern of mistakes rather than through abstract terminology.

To specify how to teach, the framework draws on four learning theories: Piaget’s constructivism, which points to tangible, hands-on experiences and discovery learning for preoperational-stage children; Vygotsky’s sociocultural theory, which emphasizes scaffolding within the zone of proximal development and cultural relevance; Papert’s constructionism, which favors learning by building shareable artifacts; and embodied learning theory, which treats physical action and sensory experience as integral to cognition. These foundations are consolidated into the framework’s signature design lens, called 3C-3H plus DS: learning experiences that are cognitively appropriate, culturally responsive, and computational-thinking-focused; that engage children’s heads, hearts, and hands; and that are supported by dialogic teacher scaffolding, with five principles of teacher talk, including being collective, reciprocal, supportive, cumulative, and purposeful, each paired with sample prompts.

The framework also lays out an incremental progression, labeled U-T-H-S, that moves preliterate learners from concrete play toward early symbolic reasoning. It begins with unplugged mini-games using familiar materials such as cards, mats, and building bricks, extends to tangible smart toys that respond to voice, touch, or blocks, bridges to hybrid experiences connecting physical kits to an on-screen display, and culminates in screen-based tools that let children create, test, and revise representations, such as training simple classifiers with examples. The approach is cyclical rather than sequential, revisiting the same core ideas across stages to support durable mental models. The author is explicit that the framework is a synthesized design proposal, not a validated intervention, and outlines priorities for future research, including multi-site design-based research and pragmatic trials comparing the progression to shorter snapshot lessons, developmentally appropriate play-based assessments, and low-cost unplugged variants for low-resource settings. For now, the framework offers curriculum designers, teacher educators, and policymakers a roadmap for bringing AI literacy to learners too young to read about it.

Subject of Research: Development of a pedagogical framework for early AI literacy in early childhood education

Article Title: Building early AI literacy: developing a pedagogical framework for early childhood education

Article References: Qian, Y. (2026). Building early AI literacy: developing a pedagogical framework for early childhood education. AI, Brain and Child, 2(1), Article 6. https://doi.org/10.1007/s44436-026-00030-w

Image Credits: AI Generated

DOI: 10.1007/s44436-026-00030-w

Keywords: AI literacy, early childhood education, pedagogical framework, computational thinking, constructivism, sociocultural theory, embodied learning, teacher scaffolding, curriculum design, preschool, AI concepts, educational technology