1. Introduction and problem

In the 3-to-6-year span—kindergarten, preschool, CENDI, infant school—a package of three artefacts has been installed as if it counted as a STEM experience. The first is a “programmable” robot kit: a Bee-Bot, a KIBO, a Matatalab or a buttoned floor device. The second is a block app: a screen that chains icons or virtual pieces and declares that the child “already programs STEM.” The third is a model that suggests STEM challenges: a generative system that produces tasks or “missions” and delivers them as if the output were inquiry. All three are visible and cheap in coordination time. They allow a setting to exhibit that it “already does STEM with artificial intelligence and robotics.” The leap—from owning an artefact or executing a sequence to claiming that there is a STEM experience—is authorized neither by the robotics evidence in the early years nor by what classrooms measure when there is scientific, design or mathematical inquiry.

The thesis of this article is restrictive. A “programmable” robot kit, a block app or a model that suggests STEM challenges do not constitute a STEM experience in early childhood education. At this stage STEM is bodily, material inquiry shared with adults; not interface training. Trapero-González, Romero-Rodríguez, Fernández-Martín, and Alonso-García (2025), with fifteen early-years robotics programmes, show that the most cited resource is the Bee-Bot, that almost all STEM competences are worked except engineering, and that the meta-analyses measure experimental gains. That does not authorize translating “there is a robot” as “there is STEM.” It authorizes asking what was measured: often computational thinking, sequencing or vocabulary, not the cycle of asking, trying with matter and asking again with an adult. NAEYC requires developmentally appropriate, intentional practice anchored in materials and in other people, not in a module detached from the room (NAEYC, 2022). An algorithm that generates a “STEM challenge” does not inquire. It administers a prompt that the profession did not interpret in the body or in the material.

The problem is worsened by a professional reason that product sheets do not mention. STEM at ages 3 to 6 is not the anticipation of a primary-school programming course, nor a “robot hour” event, nor an accumulation of button presses. The scientific, the technological, the design-related and the mathematical are not properties of the artefact, but of the use of the body, of matter and of conversation with an adult who holds the question (Cho, 2024; Hu, Huang, and Li, 2024; Pavlou, Zacharia, and Papaevripidou, 2024; OECD, 2021). Confusing the product—kit, app, challenge prompt—with the process is the category error this paper names. Darmawansah, Hwang, Chen, and Liang (2023), in thirty-nine robotics-and-STEM articles, find that the predominant discipline is Technology understood as programming, and that the most used resource is LEGO. Inference, marked as such: the market of “AI and robotics for early STEM” inherits that distribution and automates it. It turns the T of programming an interface into the whole acronym.

This paper does not recycle the axes already treated in this series. The question is one of pedagogical category: what counts as a STEM experience when an early-years setting “does AI and robotics.” The contributions are three: to reconstruct the state of the art that separates bodily and material inquiry from interface training; to examine three families of cases; and to offer four tests for deciding when a kindergarten may claim that there is STEM, and not only a kit, an app or a feed of challenges.

2. State of the art: from STEM as inquiry to the artefact on display

Four strata that the market of “AI and robotics for early STEM” usually mixes should be kept apart. The first is STEM at ages 3 to 6 as inquiry: the body, matter and the adult who shares the question (Cho, 2024; Hu et al., 2024; Pavlou et al., 2024; Wilmes and Siry, 2024; Byrne, Jensen, Thomsen, and Ramchandani, 2023). The second is the evidence on robotics and computational thinking in the early years: kits, Bee-Bot, KIBO, Matatalab, STEAM programmes and meta-analyses (Trapero-González et al., 2025; Alonso-García, Rodríguez Fuentes, Ramos Navas-Parejo, and Victoria-Maldonado, 2024; Sung et al., 2023; Zhang et al., 2025; Yang, Ng, and Gao, 2022; Levinson and Bers, 2025; Angeli and Georgiou, 2023). The third is AI that interacts, personalizes or proposes (Chen, 2024; Su and Yang, 2022; Ljungcrantz, 2026). The fourth is the framework of rights, systems and developmentally appropriate practice, which treats the 3–6-year-old as a subject of process interactions, not as the user of an interface (OECD, 2021, 2023; NAEYC, 2022; UNESCO, 2021; Miao and Holmes, 2023; European Commission, 2022).

In the inquiry stratum, Cho (2024) shows that children aged 3 to 5 investigate a phenomenon when story, body and guided play place them as protagonists of the question. Hu, Huang, and Li (2024) find that unplugged computational thinking emerges in perception–action loops, extended to the body and the material. Pavlou, Zacharia, and Papaevripidou (2024) compare, with 132 children aged 5 to 6, physical and virtual manipulatives: haptic feedback matters in some domains and not in others. Wilmes and Siry (2024) locate science in interactions with materials, body, peers and teachers. Byrne et al. (2023) review 102 physical-manipulative interventions, most with 4-to-6-year-olds: reported benefits in mathematics, space and science, and also nulls. Status: finding that early STEM runs through the body, matter and an adult. Inference: a kit or an app does not cover, by existing, that profession.

In the robotics stratum, Trapero-González et al. (2025) synthesize fifteen programmes: the Bee-Bot stands out; science, technology and mathematics are worked; engineering remains at the margin; thirteen studies enter the meta-analysis with experimental gains. Alonso-García et al. (2024) estimate, in ten studies, a standardized mean difference of 0.93 of robotics on computational thinking. Sung, Lee, and Chun (2023), with 450 children aged 5 to 6, find increases in computational thinking and expressive vocabulary, not in numeracy. Zhang et al. (2025) randomize 198 children: robot and unplugged outperform the control on computational thinking; the robot also outperforms unplugged on executive functions. Yang, Ng, and Gao (2022), with 101 children, find that the robot beats blocks on sequencing, not on global self-regulation. Levinson and Bers (2025) pilot KIBO in a Boston preschool: coding rises 4.60 points on the Coding Stages Assessment. Status: finding that robotics repeatedly moves computational thinking, sequencing or coding. Inference: moving coding is not, by itself, moving STEM as inquiry into a phenomenon with matter and with an adult.

In the AI stratum, Su and Yang (2022) review seventeen studies from 1995 to 2021. Chen (2024) maps eighteen articles, eleven countries and 15 081 children aged 2 to 8; among other affordances, AI as a tool for interactive learning and AI that adapts and personalizes. Ljungcrantz (2026) updates 2020–2024: thirty-nine studies, a peak in 2024, a focus on ages 4 to 6. Status: finding about the AI-in-ECE field, not that a model that suggests challenges constitutes STEM. Inference: personalizing and proposing missions is exactly what the third artefact sells as “STEM with AI is already happening.” Proposing a challenge is not inquiring into a phenomenon.

In the systems stratum, OECD (2021) anchors ECEC quality in process interactions. OECD (2023) locates digitalization in staff, protection and meaningful uses, not in a feed of missions. NAEYC (2022) requires developmentally appropriate practice. UNESCO (2021) requires human oversight. Miao and Holmes (2023) set a threshold of 13 years for independent conversations with generative platforms and require pedagogical validation. The European Commission (2022) and the U.S. Department of Education (2023) agree on not replacing professional judgement. Inference: a four-year-old is not the user of a kit, an app or a model that “throws STEM challenges.” The child is the subject of an inquiry that a responsible adult exercises with the body, with matter and with the group in view.

3. Review method

A critical narrative review was conducted, not a meta-analysis. The purpose was not to estimate a homogeneous effect size, but to articulate an argument of pedagogical category with verified sources. Inclusion criteria: (a) 2021–2026; (b) STEM, science, educational robotics, computational thinking or AI in early childhood education, or explicitly marked transfer when the sample is not ages 3–6; (c) relevance to kindergarten, preschool, CENDI or ages 3–6; (d) peer-reviewed journal, DOI, or a NAEYC, UNESCO, OECD, European Commission or education-department report; (e) verifiable DOI or publisher page. Axes already used in this series were excluded as a central object, although some appear as a limit.

The search was run on 27 August 2026 on DOI pages, Springer, Elsevier, Wiley, SAGE, Frontiers, ACM, MDPI, OECD iLibrary, UNESDOC, ERIC, NAEYC and publisher sites. Each source was checked against at least one of those pages. Empirical finding, conceptual or normative framework, and pedagogical inference marked as such were distinguished.

4. Case 1. Owning a kit, snapping blocks or receiving a generated challenge is not doing STEM

Trapero-González, Romero-Rodríguez, Fernández-Martín, and Alonso-García (2025) publish in Knowledge Management & E-Learning the synthesis that best names the first artefact when it is presented as STEM. They analyse fifteen early-years robotics studies (fourteen articles and one conference paper, 2013–2023; ages three to seven, mean 5.15). The standout resource is the Bee-Bot. Science, technology and mathematics are covered to different degrees; engineering remains at the margin. Thirteen studies enter the meta-analysis and show significant results in favour of experimental groups on STEM competences. Status of the evidence. Empirical finding of synthesis: educational robotics in the early years is associated, in programmes that pass the filters, with gains against a control. It is not a finding that owning the kit constitutes a STEM experience, nor that the acronym is covered. The datum the market does not cite is absent engineering and a T that, in practice, reduces to programming a floor device. Pedagogical inference, marked as such: this is the gesture a kindergarten copies when it “does STEM with a kit.” The robot is bought, button presses or a computational-thinking test are measured, and the acronym is declared fulfilled. What there is is an artefact. STEM, in Cho (2024), in Pavlou et al. (2024) and in NAEYC (2022), asks for inquiry with matter and with an adult. A kit does not observe it.

Sung, Lee, and Chun (2023) saturate the portrait with the STEAM package that best illustrates the category leap. In thirty centres in Seoul, Busan and Gyeonggi, 450 children aged 5 to 6 (334 treatment, 116 control) receive, for five weeks, a programme integrated into the Nuri curriculum that includes robotic kits (KIBO) or an equivalent curriculum on the same topic. Computational thinking (Bebras and TACTIC–KIBO) and expressive vocabulary increase. The same pattern is not replicated in numeracy, self-regulation or social behaviour as a main effect. Status: finding from a STEAM-with-robotics programme in a dense early-years N, not that the kit sufficed or that the acronym was evenly covered. Inference, marked as such: if a programme called STEAM moves coding and vocabulary and does not move numeracy, a kindergarten cannot treat the kit as STEM by naming it so. Zhang et al. (2025) randomize 198 children aged 5 to 6: robot programming and unplugged programming outperform the control on computational thinking; the robot outperforms unplugged on that indicator and on inhibition, working memory and flexibility. Status: finding on computational thinking and executive functions, not on scientific or design inquiry. Inference: gaining on a CT test does not authorize declaring a STEM experience. It authorizes declaring that a form of sequencing was trained, with or without a robot.

Levinson and Bers (2025) publish the pilot that best names coding when it is presented as preschool STEM. In a Boston centre serving children experiencing homelessness, with Head Start and universal prekindergarten classrooms, they adapt Coding as Another Language with KIBO. Eighty-five children enrol; forty-nine complete a post-test. Those who were in the UPK classrooms and received the curriculum rise 4.60 points on the Coding Stages Assessment (p < 0.0001). Status: finding of coding knowledge in a situated pilot, not of STEM as inquiry into a phenomenon. The authors do not claim that the coding score equals science, engineering or mathematics. Inference: this is the second gesture the setting copies. Coding is measured and STEM is declared. Yang, Ng, and Gao (2022) compare, in four classrooms and 101 children (mean 64.78 months), Matatalab and a block circuit: the robot wins on sequencing (F = 5.09, p < 0.05); self-regulation shows no global effect; computational thinking improves more among older children. Status: finding of sequencing against block play, not that the robot replaces material inquiry. Inference: the block, which the market treats as “the usual,” is precisely one of the material poles of early STEM. Beating blocks on sequencing does not prove that the kit covers the acronym. It proves that a sequence was trained.

The third artefact—the model that suggests STEM challenges—has, in this corpus, no kindergarten trial that measures it as inquiry. That is inference, marked as such. Chen (2024) maps interaction and personalization. Su and Yang (2022) and Ljungcrantz (2026) saturate the growth of the field. Miao and Holmes (2023) exclude children under 13 as independent interlocutors of generative platforms. Inference: a system that produces “STEM missions” personalizes a path; it does not inquire into a phenomenon. A CENDI that delivers the “AI-generated challenge of the week” has made a product. STEM is verified if the group, with the body, matter and an adult, asked, tried and asked again. It is not verified in the prompt.

5. Case 2. What the kindergarten does do when there is STEM: bodily, material inquiry shared with adults

Cho (2024) publishes in Early Childhood Education Journal the study that, in this corpus, best names early STEM as a craft of the body and of the question, not as an interface. She designs a “story-driven embodied play” environment and iterates the activity with children aged 3 to 5 in a science museum and in local preschools. The analysis, from a sociocultural perspective, shows that story, movement and guided play place the child as a protagonist who investigates a phenomenon. The body is not an ornament: it is the medium with which one interacts and understands. Status of the evidence. Empirical finding of process in science at ages 3 to 5. It is not a finding about AI, nor that a kit reproduces that craft. The museum or local preschool is not generalized to a Latin American CENDI. Pedagogical inference, marked as such: this is the object an early-years setting may properly call a STEM experience. It is bodily inquiry. A button kit does not place the child as protagonist of a phenomenon. A block app does not move the body over the question. A model that suggests challenges does not investigate: it prescribes.

Hu, Huang, and Li (2024) saturate the portrait from embodied cognition. In unplugged activities, algorithmic thinking and debugging emerge most frequently, and computational thinking extends to the material and the body, in perception–action loops. Status: finding of process, not of a robot and not of AI. Inference: if what the market sells as “the T of STEM” emerges without a screen, a kindergarten does not need a block app in order to sequence or debug. It needs an adult who holds those loops. Pavlou, Zacharia, and Papaevripidou (2024) cut the equivalence between the virtual and the material. One hundred and thirty-two children aged 5 to 6, forty-four per domain, experiment with physical or virtual manipulatives on the balance beam, sinking/floating and springs. They improve in all three domains. There is no difference on the beam. On sinking/floating the virtual condition performs better. On springs, the physical. Status: haptic feedback is domain-dependent. Inference: a kindergarten cannot treat an app or a model that simulates the phenomenon as equivalent to a trial with matter. In some domains, the hand matters. In all of them, an adult conducts the experimentation. Early STEM is not the screen that “already brings the experiment.” It is the trial, sometimes with the hand, always with a shared question.

Wilmes and Siry (2024) analyse, in an early-childhood classroom in Luxembourg, how a child investigates with a digital microscope, soil, worms and compost: science is done with materials, body, peers and teachers. Status: finding of process, not of robotics. This article does not convert it into a multilingualism study: the object retained is shared material and bodily inquiry. Byrne, Jensen, Thomsen, and Ramchandani (2023) review 102 physical-manipulative interventions, most with samples aged 4 to 6, in twenty-six countries. They group mathematics, literacy and science. They report benefits in mathematics, space and science, and also nulls and design limits. In science, children experiment with slope and speed, gravity, floating, magnetism and simple machines. Status: scoping finding, mixed evidence, not that the manipulative suffices. Inference: early STEM, when it appears in this literature, appears as a trial with objects and with a guiding adult. OECD (2021) anchors quality in process interactions. NAEYC (2022) requires intentionality. Inference: evidence of STEM is verified if the group, with the body and with matter, and with an adult, asked again. It is not verified in the kit inventory.

6. Case 3. A robot enters early STEM only if it serves inquiry, not if it trains an interface

Angeli and Georgiou (2023) publish in Frontiers in Education the trial that, in this corpus, best names the ceiling of the robot when it is left alone. One hundred and seventy children aged 5 to 6, from ten preschools, are assigned to model-based scaffolding, code-based scaffolding or a no-scaffolding control, with Bee-Bot. In the phase with scaffolding, type of support moves computational thinking in favour of the experimental groups; the control, which “explores” the robot without support, performs worse. When scaffolding is withdrawn, a gender effect in favour of boys appears. Status of the evidence. Empirical finding that the Bee-Bot is not enough: without an external memory system and without an adult who holds the sequence, the child does not complete the task under the same conditions. It is not a finding of STEM as science or engineering, nor of AI. The context is a European country; it is not generalized to a CENDI. This article does not convert it into an inclusion study: the object retained is adult scaffolding of the device, not the gender difference as a thesis. Pedagogical inference, marked as such: this is the category criterion. A “programmable” kit without an adult who holds the question is interface training, and incomplete training. Angeli and Georgiou’s control is, exactly, what a kindergarten does when it “leaves the robot on the carpet because it is already STEM.”

Bakala, Pires, Tejera, and Hourcade (2023) saturate the portrait from the profession. In Montevideo, between March and May 2023, they run five focus groups with two preschool computing teachers on Qobo, Ozobot, KIBO and Botley. The title quotes the phrase: “it will surely fall.” The teachers name size, fragility, distance between the programme and the robot, and the need to link the device to the curriculum. Status: perceptions of two teachers in a private centre, not a trial. Inference: the market sells the kit as ready-made STEM. Whoever lives the room names that the robot falls and that without a curricular link there is no craft. Bourha, Hatzigianni, Sidiropoulou, and Vitoulis (2026) observe, over three months, 37 children aged 3 to 4 in four Greek settings, with Bee-Bot, Coko Robot and other technology-enhanced toys. They code problem-solving, computational thinking and collaboration, more clearly with open-ended programmable toys. Status: qualitative observation. This article does not convert it into a free-play study: the object retained is that, at ages 3 and 4, STEM-related behaviours appear in the shared handling of an open object, not in fidelity to an interface. Inference: a model that suggests challenges does not add that behaviour. It adds a prompt.

The distribution of roles is confirmed by contrast. Yang, Ng, and Gao (2022) show that the material block produces gains; the robot adds sequencing, not an acronym. Zhang et al. (2025) show that the unplugged condition also moves computational thinking. Trapero-González et al. (2025) show absent engineering. Darmawansah et al. (2023) show that programming predominates. Inference: AI and the robot enter early STEM, if at all, on the side of the adult who inquires with the group—as support for recording a trial or as one more material among materials—subject to pedagogical validation (Miao and Holmes, 2023). They do not enter as an interface to be trained or as an emitter of challenges. OECD (2023) locates digitalization in meaningful uses. The European Commission (2022) and the U.S. Department of Education (2023) require not replacing judgement. Inference: a system that generates “the STEM challenge of the day” and counts clicks is not a STEM innovation. It is a category error. The robot that serves is the one the adult puts at the service of a question the group already has in its hands.

7. Inferential framework: four tests for claiming that there is STEM, not an artefact

The framework that follows is this article’s pedagogical inference, anchored in the cases and in the verified instruments. It is not a new international standard. It distinguishes four tests. If a kindergarten, preschool, CENDI or infant school does not pass them, it cannot declare that a “programmable” robot kit, a block app or a model that suggests STEM challenges constitutes a STEM experience.

7.1. The test of inquiry, not of the displayed artefact. Trapero-González et al. (2025) document kits and gains without covering engineering. Sung et al. (2023) move computational thinking and vocabulary, not numeracy. Levinson and Bers (2025) measure coding. Chen (2024) maps interaction and personalization. Inference: evidence of STEM is verified if the group asked, tried and asked again about a phenomenon or a design problem. If the “evidence” that the setting does STEM is a photo of the kit, the app log or the file of generated challenges, the setting has done inventory, not inquiry.

7.2. The test of the body and of matter, not of the interface. Cho (2024) locates the body as the medium of science. Hu et al. (2024) locate computational thinking in perception–action loops. Pavlou et al. (2024) show that haptic feedback matters in some domains. Byrne et al. (2023) map trial with objects. Wilmes and Siry (2024) locate materials and body with peers and teachers. Inference: the 3–6-year-old is not the user of a block app or the addressee of a prompt. The child is the subject of an inquiry that runs through the hands, through displacement, through weight, through water, through the ramp, through assembly. If AI enters, it enters as the adult’s workshop—for example, to help record a trial—subject to pedagogical validation. It does not enter as a mission board or as a screen that replaces matter.

7.3. The test of the adult who shares the question, not of interface training. Angeli and Georgiou (2023) show that the Bee-Bot without scaffolding performs worse. Bakala et al. (2023) hear that the robot falls and that it must be linked to the curriculum. Yang, Ng, and Gao (2022) show that the material block also teaches. Zhang et al. (2025) show that the unplugged condition moves computational thinking. Miao and Holmes (2023) exclude children under 13 as independent interlocutors. Inference: early STEM is not a child who “already programs alone.” It is an adult who holds the question while the group tries. A model that suggests challenges inverts the craft: it delivers the question ready-made and measures fidelity to the interface.

7.4. The test of rights and professional judgement, not of the product catalogue. UNESCO (2021) requires human oversight. Miao and Holmes (2023) require pedagogical validation. The European Commission (2022) and the U.S. Department of Education (2023) require not replacing the teacher. NAEYC (2022) requires developmentally appropriate practice. OECD (2021, 2023) anchors quality in process interactions and in meaningful digital uses. Inference: an early-years setting cannot treat the child as the operator of a kit, an app or a feed of challenges. STEM is not fulfilled by programming the interface better. It is fulfilled by inquiring with the body, with matter and with adults who share the question.

The framework admits robotics when the device is subordinated to bodily, material inquiry shared with others (Angeli and Georgiou, 2023; Bourha et al., 2026; Yang, Ng, and Gao, 2022). It admits the unplugged and the haptic trial (Hu et al., 2024; Pavlou et al., 2024; Cho, 2024). It refuses to declare STEM by a kit, an app or a model that suggests challenges (Trapero-González et al., 2025; Sung et al., 2023; Zhang et al., 2025; Levinson and Bers, 2025; Chen, 2024).

8. Discussion

Three tensions organize the discussion. The first is between displaying an artefact and inquiring into a phenomenon. It is a finding that early-years robotics is associated with gains, above all in computational thinking (Trapero-González et al., 2025; Alonso-García et al., 2024; Zhang et al., 2025); that a STEAM programme with KIBO moves CT and vocabulary, not numeracy (Sung et al., 2023); that a KIBO pilot moves coding (Levinson and Bers, 2025); and that the AI-in-ECE field grew to thirty-nine studies in 2020–2024 (Ljungcrantz, 2026; Su and Yang, 2022; Chen, 2024). It is a framework that early-years quality is played out in process interactions (OECD, 2021; NAEYC, 2022). It is not a finding that a kit, an app or a model that suggests challenges produces the craft Cho (2024), Hu et al. (2024), Pavlou et al. (2024) and Wilmes and Siry (2024) observe. The three artefacts measure what engineering knows how to deliver and declare what only inquiry would authorize.

The second is between the T of programming and the acronym. Darmawansah et al. (2023) show that the literature leans toward programming. Trapero-González et al. (2025) show absent engineering. Sung et al. (2023) and Zhang et al. (2025) measure CT. Inference: insisting that the kindergarten “already does STEM” because there is a Bee-Bot or an app is inverted pedagogy. The interface is taken and made to stand for science, design and mathematics. The child is left as an operator; the adult, as a supervisor of fidelity to the kit.

The third is between the adult who shares the question and the model that prescribes it. Angeli and Georgiou (2023) show that without scaffolding the robot does not suffice. Bakala et al. (2023) show a robot that falls. Miao and Holmes (2023) exclude children under 13. Inference: the only use of AI and robotics that does not contradict STEM at ages 3–6 is the one that remains on the side of the adult who inquires with the group, subject to pedagogical validation. A model that suggests challenges because the system needs a product is not that use. A kit that is pressed to fill the technology hour is not that use.

9. Limits

This review is narrative. It does not apply PRISMA or estimate combined effects. Trapero-González et al. (2025) and Alonso-García et al. (2024) synthesize robotics and CT, not inquiry into a phenomenon. Sung et al. (2023) are from South Korea and five weeks. Zhang et al. (2025) and Yang, Ng, and Gao (2022) measure CT, sequencing and executive functions, not the full acronym. Levinson and Bers (2025) are a coding pilot in Boston. Angeli and Georgiou (2023) are Bee-Bot and scaffolding; the gender finding is not turned into a thesis here. Bakala et al. (2023) are two teachers. Bourha et al. (2026) are a qualitative N = 37. Cho (2024), Hu et al. (2024), Pavlou et al. (2024) and Wilmes and Siry (2024) are not about AI. Byrne et al. (2023) mix pre-primary and primary, with mixed evidence. Darmawansah et al. (2023) cover mostly K–12. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) map AI in ECE, not kindergarten STEM. NAEYC, UNESCO, OECD and the 2021–2023 guidelines are framework sources. No Latin American trials of AI that measure bodily and material inquiry against a feed of STEM challenges in kindergarten were located. The inferences in section 7 are hypotheses of pedagogical category, not implementation evidence.

10. Conclusions

A “programmable” robot kit, a block app or a model that suggests STEM challenges do not constitute a STEM experience in an early childhood education setting. The verified evidence does not authorize that declaration. Fifteen programmes associate early-years robotics with gains, with the Bee-Bot dominant and engineering at the margin (Trapero-González et al., 2025). Ten studies estimate a large effect on computational thinking (Alonso-García et al., 2024). Four hundred and fifty children in a STEAM programme with KIBO gain in CT and vocabulary, not in numeracy (Sung et al., 2023). One hundred and ninety-eight children gain in CT with a robot and also without one (Zhang et al., 2025). A KIBO pilot moves coding (Levinson and Bers, 2025). One hundred and one children gain in sequencing against blocks, not in global self-regulation (Yang, Ng, and Gao, 2022). Eighteen and thirty-nine reviews saturate the growth of AI in ECE without equating it to kindergarten STEM (Chen, 2024; Su and Yang, 2022; Ljungcrantz, 2026). By contrast, when there is STEM in the early years, there is body, matter and adult: children aged 3 to 5 who investigate a phenomenon (Cho, 2024); computational thinking extended to the body (Hu et al., 2024); domain-dependent haptics (Pavlou et al., 2024); science with materials, body, peers and teachers (Wilmes and Siry, 2024); and a Bee-Bot that, without scaffolding, does not suffice (Angeli and Georgiou, 2023). Current guidance requires human oversight, pedagogical validation and not replacing professional judgement (UNESCO, 2021; Miao and Holmes, 2023; European Commission, 2022; U.S. Department of Education, 2023).

Where the sources do not measure a kindergarten, this article does not assert it. Where they measure a kit, an app or a generated challenge, it does not translate them into a STEM experience. Accompanying three- to six-year-olds in STEM is to exercise a bodily, material inquiry shared with adults. The rest is interface training. It is not STEM, and it should not be presented as what it is not.

Laboratorio Editorial de NEXTECH.IA / Ingeniero Mitre.

References

  1. Alonso-García, S., Rodríguez Fuentes, A. V., Ramos Navas-Parejo, M., and Victoria-Maldonado, J. J. (2024). Enhancing computational thinking in early childhood education with educational robotics: A meta-analysis. Heliyon, 10(13), e33249. https://doi.org/10.1016/j.heliyon.2024.e33249
  2. Angeli, C., and Georgiou, K. (2023). Investigating the effects of gender and scaffolding in developing preschool children’s computational thinking during problem-solving with Bee-Bots. Frontiers in Education, 7, Article 757627. https://doi.org/10.3389/feduc.2022.757627
  3. Bakala, E., Pires, A. C., Tejera, G., and Hourcade, J. P. (2023). “It will surely fall”: Exploring teachers’ perspectives on commercial robots for preschoolers. In Proceedings of the 2023 ACM Conference on Information Technology for Social Good (pp. 477–486). Association for Computing Machinery. https://doi.org/10.1145/3582515.3609570
  4. Bourha, D., Hatzigianni, M., Sidiropoulou, T., and Vitoulis, M. (2026). From blocks to bots: The STEM potential of technology-enhanced toys in early childhood education. Behavioral Sciences, 16(1), 161. https://doi.org/10.3390/bs16010161
  5. Byrne, E. M., Jensen, H., Thomsen, B. S., and Ramchandani, P. G. (2023). Educational interventions involving physical manipulatives for improving children’s learning and development: A scoping review. Review of Education, 11(2), e3400. https://doi.org/10.1002/rev3.3400
  6. Chen, J. J. (2024). A scoping study on AI affordances in early childhood education: Mapping the global landscape, identifying research gaps, and charting future research directions. Journal of Artificial Intelligence Research, 81, 701–740. https://doi.org/10.1613/jair.1.16882
  7. Cho, K. (2024). Story-driven embodied play: Empowering young children’s agency in science learning. Early Childhood Education Journal, 52, 1577–1585. https://doi.org/10.1007/s10643-023-01570-z
  8. Darmawansah, D., Hwang, G.-J., Chen, M.-R. A., and Liang, J.-C. (2023). Trends and research foci of robotics-based STEM education: A systematic review from diverse angles based on the technology-based learning model. International Journal of STEM Education, 10, Article 12. https://doi.org/10.1186/s40594-023-00400-3
  9. European Commission. (2022). Ethical guidelines on the use of artificial intelligence (AI) and data in teaching and learning for educators. Publications Office of the European Union. https://doi.org/10.2766/153756
  10. Hu, W., Huang, R., and Li, Y. (2024). Young children’s experience in unplugged activities about computational thinking: From an embodied cognition perspective. Early Childhood Education Journal, 52, 769–782. https://doi.org/10.1007/s10643-023-01475-x
  11. Levinson, T., and Bers, M. (2025). Robotics in universal prekindergarten classrooms. International Journal of Technology and Design Education, 35, 3–24. https://doi.org/10.1007/s10798-024-09905-6
  12. Ljungcrantz, L. (2026). The interaction of AI and early childhood education. A state-of-the-art review 2020–2024. Early Childhood Education Journal, 54, 3565–3581. https://doi.org/10.1007/s10643-025-02079-3
  13. Miao, F., and Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  14. NAEYC. (2022). Developmentally appropriate practice in early childhood programs serving children from birth through age 8 (4th ed.). NAEYC. https://www.naeyc.org/resources/pubs/books/dap-fourth-edition
  15. OECD. (2021). Starting Strong VI: Supporting meaningful interactions in early childhood education and care. OECD Publishing. https://doi.org/10.1787/f47a06ae-en
  16. OECD. (2023). Empowering young children in the digital age (Starting Strong). OECD Publishing. https://doi.org/10.1787/50967622-en
  17. Pavlou, Y., Zacharia, Z. C., and Papaevripidou, M. (2024). Comparing the impact of physical and virtual manipulatives in different science domains among preschoolers. Science Education, 108(4), 1162–1190. https://doi.org/10.1002/sce.21869
  18. 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
  19. Sung, J., Lee, J. Y., and Chun, H. Y. (2023). Short-term effects of a classroom-based STEAM program using robotic kits on children in South Korea. International Journal of STEM Education, 10, Article 26. https://doi.org/10.1186/s40594-023-00417-8
  20. Trapero-González, I., Romero-Rodríguez, J. M., Fernández-Martín, F. D., and Alonso-García, S. (2025). Educational robotics and STEM competence in early childhood education: Systematic review and meta-analysis of programmes and outcomes. Knowledge Management & E-Learning, 17(1), 71–87. https://doi.org/10.34105/j.kmel.2025.17.003
  21. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137
  22. U.S. Department of Education, Office of Educational Technology. (2023). Artificial intelligence and the future of teaching and learning: Insights and recommendations. U.S. Department of Education. https://www.ed.gov/sites/ed/files/documents/ai-report/ai-report.pdf
  23. Wilmes, S. E. D., and Siry, C. (2024). Engaging with materials and the body: Young plurilingual children’s resource-rich interactions in science investigations. Research in Science & Technological Education, 42(1), 114–132. https://doi.org/10.1080/02635143.2023.2298353
  24. Yang, W., Ng, D. T. K., and Gao, H. (2022). Robot programming versus block play in early childhood education: Effects on computational thinking, sequencing ability, and self-regulation. British Journal of Educational Technology, 53(6), 1817–1841. https://doi.org/10.1111/bjet.13215
  25. Zhang, X., Chen, Y., Hu, L., Hwang, G.-J., and Tu, Y.-F. (2025). Developing preschool children’s computational thinking and executive functions: Unplugged vs. robot programming activities. International Journal of STEM Education, 12, Article 10. https://doi.org/10.1186/s40594-024-00525-z