1. Introduction and problem

In the 3-to-6 age band—kindergarten, preschool, CENDI, early childhood school—a package of five artifacts has taken hold that purport to count as support for peer conflict negotiation / turn-taking negotiation / turn-taking / situated dispute resolution: a detector/CNN/computer vision or multimodal system that «detects conflicts» or scores social competence / peer engagement / conflict count / turn compliance / friendship behavior / SEC auto-score as pedagogical assessment; a GenAI that generates conflict resolution scripts, turn-taking lesson plans, sharing worksheets, or peace corner scripts by prompt as a closed recipe; a dashboard of conflicts per hour, minutes of dispute, or turn-taking compliance for administrative surveillance; a robot/app that replaces adult mediation and peer negotiation with a correctness score or an automated «say sorry» script; and a system that converts proximity, synchrony, or facial emotion into a ranking of social developmental level without pedagogical mediation. The five make it possible to display that the center «already does peer conflict / turn-taking with AI». The leap—from detection, score, generated script, count metric, or maturational ranking to claiming support—is not authorized by AI-in-ECE mappings or by the pedagogy of negotiation when there are disputes over turns, materials, roles, and space; verbal and nonverbal peer negotiation; adult mediation that names emotions, models request/wait/offer, and sustains children’s agency to repair; and co-presence that reads conflict as an opportunity for social learning, not as an infraction of a score.

The thesis is restrictive: those five artifacts do not constitute support for peer conflict negotiation / turn-taking in early childhood. In early childhood education that practice develops with real materials, adult co-presence that interprets meaning and process, request/wait/offer language, and children’s agency to repair (NAEYC, 2022; OECD, 2021, 2023); not a detector, a GenAI of scripts, a count dashboard, or a robot that replaces mediation. Kengesbay (2025) contributes DL/CV (RNN/3D CNN) to detect social conflicts in kindergarten: a detection finding, not a pedagogy of repair. Xiao and Li (2026)—N = 150; more than 120 hours—propose multimodal fusion and graph-temporal modelling for social competence assessment: score ≠ request/wait/offer. Li and Li (2026)—N = 500; 36–72 months; eight kindergartens—integrate face, voice, and movement for SEC/CASEL: SEC auto-score ≠ situated turn-taking negotiation. Markova et al. (2026)—N = 20—and Horn et al. (2024) track proximity/synchrony: spatial dynamics ≠ mediation that names and models. Yu (2021) monitors facial emotion: emotional score ≠ mediation. Nijssen et al. (2021) and Kim et al. (2024) situate sharing/friendship with a robot; Tykhonenko et al. (2026)—N = 28—measure turn-timing with a robot; Pinto and Belpaeme (2024) anticipate turn transitions in HRI (transfer); Araguas et al. (2025) compare human tutor versus robot. Taylor et al. (2023) contrast peer engagement with teacher quality at the classroom level (non-AI). That authorizes asking what was measured: conflict count, competence, SEC, proximity, or compliance—not the practice when two children negotiate a turn and an adult sustains repair.

The problem is aggravated by six category confusions. First: it is not generic SEL—Li and Li (2026) anchor SEC/CASEL; the axis is negotiating turns and repairing among peers—. Second: it is not child participation/voice as the axis. Third: it is not CLASS / adult–child interactions as the axis—Taylor et al. (2023) do not authorize reducing peer negotiation to a CLASS score—. Fourth: it is not joint attention—Markova et al. (2026); Horn et al. (2024)—. Fifth: it is not free play or symbolic play—Galbraith (2022) is a peripheral contrast—. Sixth: it is not clinical emotion monitoring—Yu (2021); Araguas et al. (2025)—nor adult/general HRI as classroom pedagogy (Pinto and Belpaeme, 2024, transfer). Mark-making, dialogic reading, literacy, orality, sand/water/playdough/loose parts/outdoor/risky/block/spatial, inclusion, wellbeing, formative assessment, documentation, STEM, environmental education, motor skills, creativity, risky play, and inquiry are not recycled as axes. There is an economy of coordination: the five artifacts fit on a slide; the episode of dispute, request/wait/offer, and repair does not. NAEYC (2022) and OECD (2021, 2023) require meaningful interactions and subordinate digitization. Contributions: separating craft and artifacts; examining marked cases; and offering four tests of support.

2. State of the art: from the situated practice of peer negotiation to the artifact on display

It is useful to separate four strata that the market for «AI for peer conflict / turn-taking in early childhood» tends to conflate. The first is the construct of peer conflict negotiation / turn-taking negotiation for ages 3 to 6 as a relational and discursive practice: disputes over turns, materials, roles, and space; verbal and nonverbal negotiation; repair with children’s agency (NAEYC, 2022; OECD, 2021, 2023; Taylor et al., 2023, read as a contrast of peer engagement, not as a CLASS axis). The second is the craft that cultivates it—real materials; an adult who names emotions, models request/wait/offer, and sustains peer repair; co-presence that interprets conflict as social learning, not as an infraction of a score—(NAEYC, 2022; OECD, 2021, 2023; Galbraith, 2022, peripheral contrast). The third is the evidence on AI affordances in ECE and on systems of detection, multimodal scoring, spatial tracking, emotion monitoring, and robotic mediation—without equating them to situated negotiation—(Chen, 2024; Ljungcrantz, 2026; Nikolopoulou, 2025; Kengesbay, 2025; Xiao and Li, 2026; Li and Li, 2026; Markova et al., 2026; Horn et al., 2024; Yu, 2021; Nijssen et al., 2021; Kim et al., 2024; Tykhonenko et al., 2026; Pinto and Belpaeme, 2024; Araguas et al., 2025). The fourth is the framework of rights and developmentally appropriate practice, which treats the young child as a subject who negotiates and repairs, not as a vector of conflict-count nor as the exclusive addressee of GenAI scripts or robots that «say sorry» (UNESCO, 2021; Miao and Holmes, 2023).

NAEYC (2022) and OECD (2021, 2023) anchor developmentally appropriate practice, meaningful interactions, and subordinate digitization: framework. The verified systems measure objects distinct from the craft. Kengesbay (2025) detects social conflicts in kindergartens with DL/CV: a video-detection finding, not peer request language. Xiao and Li (2026)—N = 150—contribute multimodal assessment of social competence and emotional expression. Li and Li (2026)—N = 500—contribute multimodal SEC aligned to CASEL: an SEC assessment finding; marked ≠ situated turn-taking negotiation. Markova et al. (2026)—N = 20—and Horn et al. (2024) contribute tracking/sensors of spatial and temporal dynamics: proximity/synchrony as indicators of affiliation, not mediation that repairs. Yu (2021) monitors facial emotion: emotional score. Nijssen et al. (2021) examine sharing with a robot and anthropomorphic attributions; Kim et al. (2024) design robotic mediation of friendship behaviors; Tykhonenko et al. (2026)—N = 28—relate shyness and contingent dialogical actions to a social robot’s instructions; Pinto and Belpaeme (2024) predict turn transitions in human–robot dialogue: marked HRI transfer. Araguas et al. (2025) contrast human tutor versus robot in preschool emotion recognition. Taylor et al. (2023) fix the pedagogical contrast: individual peer engagement and classroom-level teacher–child interaction quality, without AI. Inference: a conflict count neither names the emotion nor models «may I when you finish?»; an adult who sustains peer repair does. Chen (2024) and Ljungcrantz (2026) map AI in ECE without equating it to situated negotiation. Nikolopoulou (2025) and Miao and Holmes (2023) set caution and thresholds for GenAI lesson plans/scripts by prompt. UNESCO (2021) requires human oversight. Galbraith (2022) is a peripheral contrast of play pedagogies.

3. Review method

A critical narrative review was conducted, not a primary meta-analysis. The purpose was not to estimate a homogeneous effect size for the five artifacts, but to articulate an argument of pedagogical category with verified sources. Inclusion criteria: (a) 2021–2026, with marked transfer when the sample does not equate to ages 3–6 in a peer-conflict classroom, is adult/general HRI, clinical emotion monitoring, or the object is SEC/CASEL without equating it to situated negotiation; (b) peer conflict, turn-taking, sharing, friendship behaviors, multimodal social competence/SEC, spatial tracking, robotic mediation, or AI in ECE with artifact/craft relevance; (c) kindergarten, preschool, CENDI, or ages 3–6, or explicit transfer; (d) peer-reviewed journal, DOI, or NAEYC/UNESCO/OECD report; (e) verifiable DOI or editorial page. Axes already used in this series were excluded as central objects—generic SEL, participation/voice, CLASS/adult–child interactions as axis, joint attention, free play, symbolic play, mark-making, dialogic reading, literacy, orality, sand/water/playdough/loose parts/outdoor/risky/block/spatial, inclusion, wellbeing, formative assessment, documentation, STEM, environmental education, motor skills, creativity, risky play, and inquiry—. Galbraith (2022) only as a peripheral contrast of play pedagogies, not teacher education as axis. Pinto and Belpaeme (2024) = predictive HRI, transfer. Li and Li (2026) are read for multimodal SEC, not as SEL already published as an axis. Taylor et al. (2023) are read as a pedagogical contrast of peer engagement, not as a CLASS axis. Yu (2021) and Araguas et al. (2025) are read for emotion monitoring / robot tutor, not as a clinical or SEL axis.

The search was executed on 7 September 2026 (slot catch-up Monday 09:02 America/Mexico_City; MISSED, run NOW) on DOI pages, Crossref, Springer, Elsevier, Wiley, Taylor & Francis, MDPI, IEEE Xplore, Frontiers, Royal Society, JAIR, OECD iLibrary, UNESDOC, NAEYC, and editorial sites. Each source was verified against the fuentes.md corpus and crossref_check.json. Empirical finding, framework, and pedagogical inference were distinguished. No N, d, r, AUC, or DOI was invented: when an artifact lacks a verified study with metrics of a conflict-count dashboard, GenAI of peace corner scripts as a display package in Latin American CENDI, or a robot that replaces mediation with «say sorry» in a 3–6 classroom, it is discussed as a category ceiling supported by AI-in-ECE mappings (Chen, 2024; Ljungcrantz, 2026) and by verified detection/scoring/HRI systems (Kengesbay, 2025; Xiao and Li, 2026; Li and Li, 2026; Markova et al., 2026; Horn et al., 2024; Yu, 2021; Nijssen et al., 2021; Kim et al., 2024; Tykhonenko et al., 2026; Pinto and Belpaeme, 2024; Araguas et al., 2025). Twenty-one sources from the verified corpus were used.

4. Case 1. A CV/multimodal detector of conflicts / social competence / conflict count / turn compliance, GenAI of scripts/lesson plans/sharing worksheets, conflicts-per-hour dashboard, robot/app that replaces mediation with a score or «say sorry», or proximity/synchrony→social developmental ranking do not constitute support for peer conflict negotiation / turn-taking

Chen (2024) and Ljungcrantz (2026) saturate the artifact ceiling when AI in ECE is presented as if it were support for peer conflict negotiation. Chen (2024) maps global affordances: a scoping finding—tutoring, analytics, content generation, agents—not that a conflict count or social competence score cultivates request/wait/offer. Ljungcrantz (2026) reviews AI–ECE interaction 2020–2024: state of the art, not situated turn-taking negotiation. Inference: «low conflict count / high turn compliance = there was support» is an analytics ceiling. Kengesbay (2025), Xiao and Li (2026), and Li and Li (2026) measure detection or competence/SEC; an administrative score can coexist with the absence of mediation that interprets meaning and process.

Nikolopoulou (2025) and Miao and Holmes (2023) name the risk of GenAI conflict resolution scripts / turn-taking lesson plans / sharing worksheets / peace corner scripts by prompt that closes the episode: a caution framework, not a trial versus an episode with real materials and an adult who models requesting (NAEYC, 2022; OECD, 2021). Inference: producing a «say sorry» script by prompt may be subordinate preparation; support begins when there is a situated dispute, children who negotiate, and an adult who names without reducing to correctness. Kim et al. (2024) and Nijssen et al. (2021) show sharing/friendship with a robot under experimental conditions; they do not authorize replacing adult mediation. Tykhonenko et al. (2026) and Pinto and Belpaeme (2024) measure turn-timing or turn prediction in HRI: latency ≠ pedagogy of repair.

In the verified corpus, the five artifacts lack pedagogical equivalence trials with the craft in 3–6 CENDI; no d, r, or AUC of «support» is invented. Markova et al. (2026) and Horn et al. (2024) measure proximity/synchrony; Yu (2021) facial emotion; Araguas et al. (2025) human versus robot emotion recognition; Taylor et al. (2023) peer engagement without AI; Galbraith (2022) peripheral contrast. Restrictive inference: detector ≠ adult who names; GenAI of scripts ≠ open episode; dashboard ≠ observation; «say sorry» robot ≠ agency to repair; ranking by proximity ≠ mediation. UNESCO (2021) requires human oversight. The five share a grammar of substitution: the score speaks for the dispute; the GenAI closes; the dashboard replaces observation; the robot erases peer negotiation; the ranking converts process into level without mediation. When the score speaks for the child and the script speaks for the adult, the human-oversight threshold is breached even if the dashboard looks «green».

5. Case 2. What the kindergarten does do when there is support for conflict negotiation / turn-taking: relational-discursive craft and what Kengesbay, Xiao/Li, Li/Li, Markova, Horn, Yu, Nijssen, Kim, Tykhonenko, Taylor, Pinto/Belpaeme, and Araguas measure—with Galbraith as peripheral contrast

The floor of the craft is not a score: it is real materials in dispute (turns, objects, roles, space); verbal and nonverbal peer negotiation; adult mediation that names emotions, models request/wait/offer language, and sustains children’s agency to repair; co-presence that interprets conflict as an opportunity for social learning, not as an infraction of a «score» (NAEYC, 2022; OECD, 2021, 2023). Inference: there is support when that practice is protected; not when conflict count, GenAI of scripts, or social developmental ranking is displayed. The craft is recognized in the episode: two children dispute; the adult names, offers language, and sustains their repairing; the robot or app, if they appear, are means subordinate to the adult, not substitutes.

Kengesbay (2025) saturates the detection anchor: deep learning and computer vision to detect social conflicts in kindergartens with real video (RNN/3D CNN). Status: empirical detection finding. Inference: it measures presence/detection of conflict in video; it does not equate to a pedagogy of negotiation nor authorize presenting a CNN as an assessment of peer repair. Xiao and Li (2026)—N = 150; more than 120 hours—contribute a multimodal fusion and graph-temporal framework for developmental assessment of social competence and emotional expression in preschoolers. Marked distinction: it is cited for the competence-scoring artifact; this article is not converted into generic SEL. Inference: social competence score ≠ request/wait/offer language in a situated dispute. Li and Li (2026)—N = 500; 36–72 months; eight kindergartens—contribute a multimodal system (facial, voice, movement) for assessing social-emotional competence aligned to CASEL. Marked inference: SEC auto-score ≠ peer turn-taking negotiation; it does not certify mediation that repairs.

Markova, Arato, Ceral, Hofbauer, Quigley, and Horn (2026)—N = 20 preschoolers—decode social dynamics with automatic tracking of spatial and temporal patterns (position, proximity, orientation) to investigate interactions and relationships in peer groups. Horn, Karsai, and Markova (2024) formulate an automated data-driven approach to children’s social dynamics in space and time with sensors/ML: proximity and synchrony as indicators of affiliation. Inference: tracking of affiliation does not replace the adult who names emotion or the children who offer turns. Yu (2021) contributes emotion monitoring based on face/emotion recognition in preschool: transfer; emotional score ≠ conflict mediation. Nijssen, Müller, Bosse, and Paulus (2021) examine children’s sharing with a robot and the role of anthropomorphic emotional attributions: HRI of sharing; it does not authorize a robot that replaces adult mediation in peer disputes. Kim, Hwang, Lim, Cho, and Lee (2024) design robotic mediation to facilitate friendship behaviors: a CRI design finding; friendship behavior score ≠ the craft of situated repair between two children fighting over a material.

Tykhonenko, Tolksdorf, and Rohlfing (2026)—N = 28—relate children’s shyness to contingent dialogical actions when reacting to a social robot’s instructions, including turn-timing latencies. Inference: turn latency with a robot ≠ pedagogical turn-taking negotiation in a dispute over roles/space. Pinto and Belpaeme (2024) propose predictive turn-taking with language models to anticipate turn transitions in human–robot dialogue (RO-MAN): marked transfer; HRI ≠ a 3–6 peer-conflict classroom. Araguas, Blanchard, Derégnaucourt, Chopin, and Guellai (2025) compare human tutor versus robot on body knowledge and emotion recognition in preschool: the human/robot contrast does not authorize replacing conflict mediation with a robotic tutor. Taylor, Alamos, Turnbull, LoCasale-Crouch, and Howes (2023) examine individual peer engagement in pre-kindergarten in relation to classroom-level teacher-child interaction quality: non-AI pedagogical contrast; peer engagement ≠ conflict-count dashboard nor recycled CLASS axis. Galbraith (2022)—play pedagogies among preservice teachers—is a peripheral contrast, not a teacher-education axis. Coherent mediation (NAEYC, 2022; OECD, 2021, 2023): materials in dispute, observation of process, request/wait/offer language, and agency to repair without a correctness script.

Taken together: there are systems that detect conflicts (Kengesbay, 2025), score competence/SEC (Xiao and Li, 2026; Li and Li, 2026), track proximity (Markova et al., 2026; Horn et al., 2024), monitor emotion (Yu, 2021; Araguas et al., 2025), mediate friendship/sharing with a robot (Kim et al., 2024; Nijssen et al., 2021), measure HRI turn-timing (Tykhonenko et al., 2026; Pinto and Belpaeme, 2024), or contrast pedagogical peer engagement (Taylor et al., 2023). None—by itself—certifies verbal/nonverbal peer negotiation, mediation that names and models, and children’s agency to repair. The technical evidence is real in its domain; «score = support for conflict negotiation / turn-taking» is an illegitimate category inference.

6. Case 3. Peer conflict / turn-taking negotiation ≠ SEL as axis, participation-voice, CLASS-interactions, joint attention, free play, symbolic play, or clinical emotion monitoring

Six boundaries protect the axis. First: generic SEL—Li and Li (2026) measure SEC/CASEL; there may be emotional regulation IN the dispute; the axis is the relational-discursive practice of negotiating turns and repairing among peers—. Inference: «we work on SEL = there is conflict negotiation» does not certify; nor is an SEC auto-score pedagogical turn-taking negotiation. Second: child participation/voice—there is speech and agency in repair; this article is not converted into participation/voice already published—. Third: CLASS / adult–child interactions as axis—Taylor et al. (2023) relate peer engagement to classroom-level teacher quality; they do not authorize reducing peer negotiation to an already published adult–child interaction score—. Fourth: joint attention—Markova et al. (2026) and Horn et al. (2024) record orientation/proximity; they do not authorize rewriting the axis as joint attention—. Fifth: free play / symbolic play—Galbraith (2022) is a peripheral contrast of play pedagogies; the axis is not free play or symbolic play already published—. Sixth: clinical emotion monitoring—Yu (2021); Araguas et al. (2025)—fix emotion score/tutor; emotional screening ≠ mediation of disputes over materials in the kindergarten—. UNESCO (2021), Miao and Holmes (2023), and Nikolopoulou (2025) require human oversight and teacher mediation of GenAI. Inference: coherent AI for ages 3–6 remains on the adult’s side (subordinate preparation/documentation; Galbraith, 2022 peripheral; Taylor et al., 2023 as a contrast of pedagogical observation), not as a conflict detector, GenAI of peace corner scripts, conflicts-per-hour dashboard, «say sorry» robot, or social developmental ranking without mediation.

These boundaries protect the axis. Peer conflict / turn-taking negotiation may touch, laterally, SEL, voice, interaction quality, co-orientation, play, or emotion; it is not reduced to any of those already published axes. A center that «works on SEL», «gives voice», «improves CLASS», «works on joint attention», «does free/symbolic play» or «monitors emotions» has not, by that label alone, demonstrated situated negotiation with mediation that names/models and agency to repair. The distinction is one of pedagogical category, not a denial of lateral overlaps.

7. Inferential framework: four tests for claiming support for peer conflict negotiation / turn-taking, not an artifact

The framework that follows is pedagogical inference of this article, 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 early childhood school does not pass them, it cannot declare that the five artifacts constitute support for peer conflict negotiation / turn-taking negotiation.

7.1. Test of situated practice (disputes over turns/materials/roles/space; verbal and nonverbal negotiation; agency to repair), not of the CV/multimodal detector nor of the correctness or «say sorry» robot/app. NAEYC (2022), OECD (2021, 2023), and the detection/scoring/HRI cases (Kengesbay, 2025; Xiao and Li, 2026; Li and Li, 2026; Nijssen et al., 2021; Kim et al., 2024; Tykhonenko et al., 2026; Pinto and Belpaeme, 2024; Araguas et al., 2025) define the craft/artifact contrast. Chen (2024), Ljungcrantz (2026), and Nikolopoulou (2025) map agents/analytics as affordance, not as situated negotiation. Inference: evidence of support is verified in situated dispute, peer negotiation, and repair with agency. If the «evidence» is conflict count, social competence, friendship behavior score, or automated «say sorry», the center did behavioral analytics or HRI, not support.

7.2. Test of adult co-presence that names emotions, models request/wait/offer, and sustains repair—not merely «say sorry»—not of GenAI conflict resolution scripts / turn-taking lesson plans / sharing worksheets / peace corner scripts. NAEYC (2022), OECD (2021), and Galbraith (2022, peripheral) situate practice and interactions. Miao and Holmes (2023) and Nikolopoulou (2025) require validation and mediation of GenAI. Inference: a script by prompt does not certify the episode of peer negotiation.

7.3. Test of human mediation and of interpreting conflict as social learning, not of the conflicts-per-hour / minutes of dispute / turn-taking compliance dashboard nor of social developmental ranking by proximity/synchrony/facial emotion. OECD (2021, 2023) require meaningful interactions. Markova et al. (2026), Horn et al. (2024), and Yu (2021) fix tracking/monitoring; Taylor et al. (2023) anchor pedagogical observation of peer engagement without converting it into count surveillance. Inference: low conflicts-per-hour or high compliance do not by themselves elevate mediation that repairs. Subordinate adult preparation/documentation is legitimate; the scorer that converts the child into a conflict-count vector is not.

7.4. Test of category distinction and professional judgment, not of the product catalog. Peer conflict / turn-taking ≠ SEL as axis, participation-voice, CLASS-interactions, joint attention, free play, symbolic play, or clinical emotion monitoring. Li and Li (2026), Taylor et al. (2023), Markova et al. (2026), Horn et al. (2024), Yu (2021), Araguas et al. (2025), and Galbraith (2022) block those confusions. UNESCO (2021), Miao and Holmes (2023), NAEYC (2022), and OECD (2021, 2023) require human oversight and not replacing professional judgment. Inference: support is fulfilled with real materials in dispute, peer negotiation, co-presence that names/models, and agency to repair—not by detecting conflicts, scoring turn compliance, or surveilling minutes of dispute.

The framework admits digital subordinate to adult preparation/documentation (Galbraith, 2022 peripheral; Taylor et al., 2023 as a contrast of pedagogical reading of peer engagement; Kengesbay, 2025 and Xiao and Li, 2026 as detection/assessment tools, not substitution of mediation). It rejects declaring support via the five artifacts (Miao and Holmes, 2023; Nikolopoulou, 2025; UNESCO, 2021). The four tests are read together.

8. Discussion

Three tensions. First: displaying the five artifacts versus exercising support. Kengesbay (2025), Xiao and Li (2026; N = 150), Li and Li (2026; N = 500), Markova et al. (2026; N = 20), Horn et al. (2024), Yu (2021), Nijssen et al. (2021), Kim et al. (2024), Tykhonenko et al. (2026; N = 28), Taylor et al. (2023), Pinto and Belpaeme (2024, transfer), Araguas et al. (2025), and Galbraith (2022 peripheral) sustain the contrast; Chen (2024) and Ljungcrantz (2026) map AI without equivalence to situated negotiation. Second: automated assessment versus relational-discursive pedagogy—declaring support by conflict count, SEC auto-score, or turn compliance is inverted pedagogy—. Third: GenAI of scripts / «say sorry» robot / ranking by proximity versus craft—Miao and Holmes (2023), Nikolopoulou (2025), NAEYC (2022), and OECD (2021, 2023)—: selling peace corner scripts or robotic friendship mediation without human mediation confuses product with practice. Taylor et al. (2023) confirm peer engagement with interaction quality without authorizing surveillance dashboards; UNESCO (2021) subordinates AI to professional judgment.

The four tests in section 7 read these tensions. The empirical contrast—detection; competence N = 150; SEC N = 500; tracking N = 20; sensors/ML; emotion monitoring; sharing/friendship robots; turn-timing N = 28; pedagogical peer engagement; predictive HRI with transfer; human vs robot emotion; Galbraith peripheral—defines the floor that the artifacts do not reach alone. Inference: support erodes when they are treated as if they were the practice. Subordinate digital is legitimate toward the adult (language preparation; documentation; Taylor et al., 2023 contrast); inverting the sequence is not (NAEYC, 2022; OECD, 2021, 2023). If the center shows situated disputes, negotiation, an adult who named and modeled, and children who repaired, it may speak of support; if it only shows detectors, GenAI scripts, conflicts per hour, «say sorry» robots, or rankings without mediation, it speaks of artifacts. Chen (2024) and Ljungcrantz (2026): AI affordances in ECE ≠ situated turn-taking negotiation.

9. Limits

This review is narrative. It applies neither its own PRISMA nor estimates of combined effects. Kengesbay (2025) contributes conflict detection in kindergarten: no d or r is invented; it is read as a detector, not as a craft of repair. Xiao and Li (2026) measure N = 150 multimodal competence: assessment-construct limit. Li and Li (2026) measure N = 500 SEC/CASEL: limit; SEC ≠ situated negotiation. Markova et al. (2026) measure N = 20 tracking: small sample. Horn et al. (2024) contribute a sensors/ML framework. Yu (2021) is emotion monitoring: transfer to mediation. Nijssen et al. (2021) and Kim et al. (2024) delimit sharing/friendship with a robot. Tykhonenko et al. (2026) measure N = 28 turn-timing with a robot. Pinto and Belpaeme (2024) are predictive HRI: marked transfer. Araguas et al. (2025) contrast human/robot tutor in emotion recognition. Taylor et al. (2023) are non-AI pedagogical contrast. Galbraith (2022) is peripheral contrast, not an axis. Chen (2024), Ljungcrantz (2026), and Nikolopoulou (2025) map AI in ECE, not conflict-count dashboards or GenAI of peace corner scripts as a display package in Latin American CENDI. No verified trials of the five artifacts as a single «peer conflict / turn-taking with AI» package in a 3–6 classroom were located; they are discussed as a category ceiling. NAEYC, UNESCO, and OECD are framework sources. The inferences in section 7 are category hypotheses, not implementation evidence.

10. Conclusions

The five artifacts—CV/multimodal detector of conflicts or scores of social competence / peer engagement / conflict count / turn compliance / friendship behavior / SEC; GenAI of conflict resolution scripts / turn-taking lesson plans / sharing worksheets / peace corner scripts; dashboard of conflicts per hour / minutes of dispute / turn-taking compliance; robot/app that replaces mediation with a score or «say sorry»; social developmental ranking by proximity/synchrony/facial emotion without mediation—do not constitute support for peer conflict negotiation / turn-taking negotiation in early childhood education. Chen (2024) and Ljungcrantz (2026) confirm affordances without equivalence to situated negotiation. Nikolopoulou (2025) and Miao and Holmes (2023) set GenAI limits. Kengesbay (2025), Xiao and Li (2026), Li and Li (2026), Markova et al. (2026), Horn et al. (2024), Yu (2021), Nijssen et al. (2021), Kim et al. (2024), Tykhonenko et al. (2026), Taylor et al. (2023), Pinto and Belpaeme (2024), and Araguas et al. (2025) require reading detection/scoring/tracking/HRI/emotion for what they measure—with transfers—and for what they do not equate to a pedagogy of negotiation. Galbraith (2022) is a peripheral contrast: adult ≠ detector or GenAI of scripts. When there is support there are real materials in dispute, peer negotiation, co-presence that names/models, and children’s agency to repair (NAEYC, 2022; OECD, 2021, 2023). Conflict negotiation / turn-taking is distinguished from SEL, participation-voice, CLASS-interactions, joint attention, free play, symbolic play, and clinical emotion monitoring as axes. The guidelines require human oversight (UNESCO, 2021; Miao and Holmes, 2023).

Where the sources do not measure a kindergarten, this article does not claim it. Where they measure AI mappings, conflict detection, multimodal competence/SEC, spatial tracking, emotion monitoring, sharing/friendship robots, HRI turn-timing, or pedagogical peer engagement, it does not translate them into support via conflict-count, turn-compliance, or script by prompt. Accompanying three-to-six-year-old girls and boys in peer conflict negotiation / turn-taking is to exercise situated disputes over turns, materials, roles, and space; verbal and nonverbal negotiation; adult mediation that names emotions, models request/wait/offer, and sustains agency to repair; and co-presence that interprets conflict as social learning, not as an infraction of a «score». Everything else is a CV/multimodal detector, GenAI of scripts/lesson plans/worksheets, conflicts-per-hour dashboard, correctness or «say sorry» robot/app, and social developmental ranking. It is not support for peer conflict negotiation / turn-taking in early childhood education, and it must not be presented as what it is not.

Laboratorio Editorial de NEXTECH.IA / Ingeniero Mitre.

References

  1. Araguas, A., Blanchard, A., Derégnaucourt, S., Chopin, A., y Guellai, B. (2025). Body knowledge and emotion recognition in preschool children: A comparative study of human versus robot tutors. Behavioral Sciences, 16(1), Article 29. https://doi.org/10.3390/bs16010029
  2. 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
  3. Galbraith, J. (2022). “A prescription for play”: Developing early childhood preservice teachers’ pedagogies of play. Journal of Early Childhood Teacher Education, 43(3), 474–494. https://doi.org/10.1080/10901027.2022.2054035
  4. Horn, L., Karsai, M., y Markova, G. (2024). An automated, data-driven approach to children's social dynamics in space and time. Child Development Perspectives, 18(2). https://doi.org/10.1111/cdep.12495
  5. Kengesbay, D. (2025). Detecting social conflicts in kindergartens using deep learning and computer vision. Journal of Emerging Technologies and Computing. https://doi.org/10.47344/7x77b619
  6. Kim, Y., Hwang, J., Lim, S., Cho, M.-H., y Lee, S. (2024). Child–robot interaction: Designing robot mediation to facilitate friendship behaviors. Interactive Learning Environments, 32(8), 4169–4182. https://doi.org/10.1080/10494820.2023.2194936
  7. Li, J., y Li, X. (2026). A multimodal artificial intelligence system integrating facial expression voice emotion and behavioral movement analysis for assessing social-emotional competence in preschool children. Scientific Reports. https://doi.org/10.1038/s41598-026-61418-5
  8. Ljungcrantz, L. (2026). The interaction of AI and early childhood education. A state-of-the-art review 2020–2024. Early Childhood Education Journal, 54(5), 3565–3581. https://doi.org/10.1007/s10643-025-02079-3
  9. Markova, G., Arato, J., Ceral, R., Hofbauer, M., Quigley, C., y Horn, L. (2026). Decoding preschool social dynamics: Automated tracking of spatial and temporal patterns to investigate social interactions and relationships in peer groups. Developmental Science. https://doi.org/10.1111/desc.70241
  10. Miao, F., y Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  11. NAEYC. (2022). Developmentally appropriate practice in early childhood programs serving children from birth through age 8 (4.ª ed.). NAEYC. https://www.naeyc.org/resources/pubs/books/dap-fourth-edition
  12. Nijssen, S. R. R., Müller, B. C. N., Bosse, T., y Paulus, M. (2021). You, robot? The role of anthropomorphic emotion attributions in children’s sharing with a robot. International Journal of Child-Computer Interaction, 30, Article 100319. https://doi.org/10.1016/j.ijcci.2021.100319
  13. Nikolopoulou, K. (2025). Child-centered integration of generative AI in early learning: Balancing promises and challenges. AI, Brain and Child, 1(1). https://doi.org/10.1007/s44436-025-00023-1
  14. OECD. (2021). Starting Strong VI: Supporting meaningful interactions in early childhood education and care. OECD Publishing. https://doi.org/10.1787/f47a06ae-en
  15. OECD. (2023). Empowering young children in the digital age (Starting Strong). OECD Publishing. https://doi.org/10.1787/50967622-en
  16. Pinto, M. G., y Belpaeme, T. (2024). Predictive turn-taking: Leveraging language models to anticipate turn transitions in human-robot dialogue. En 2024 33rd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN). IEEE. https://doi.org/10.1109/ro-man60168.2024.10731379
  17. Taylor, M., Alamos, P., Turnbull, K. L. P., LoCasale-Crouch, J., y Howes, C. (2023). Examining individual children's peer engagement in pre-kindergarten classrooms: Relations with classroom-level teacher-child interaction quality. Early Childhood Research Quarterly, 64, 221–232. https://doi.org/10.1016/j.ecresq.2023.04.007
  18. Tykhonenko, V., Tolksdorf, N. F., y Rohlfing, K. J. (2026). The relation between children’s shyness and their contingent dialogical actions when reacting to a social robot’s instructions. Philosophical Transactions of the Royal Society B. https://doi.org/10.1098/rstb.2024.0375
  19. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137
  20. Xiao, X., y Li, S. (2026). An AI-driven multimodal fusion and graph-temporal modelling framework for developmental assessment of social competence and emotional expression in preschool children. International Journal of Reasoning-based Intelligent Systems. https://doi.org/10.1504/ijris.2026.154537
  21. Yu, G. (2021). Emotion monitoring for preschool children based on face recognition and emotion recognition algorithms. Complexity, 2021, Article 6654455. https://doi.org/10.1155/2021/6654455