Journal of Children's Education Journal of Children's Education (CEJ). Vol. 7 No. 3 (2025) 253e-ISSN: 2685-1903 p-ISSN: 2087-0396 (CEJ) DOI: 10.33086/cej.v7i3.8436 ORIGINAL RESEARCH ARTICLE Adaptive AI-Driven Formative Assessment in Early Childhood Education: A Systematic Review and MetaAnalysis on Cultivating Social-Emotional Learning and Early Moderation Eka Danik prahastiwi1* , Minsih2 , and Siti Norlina Muhammad 3 1 Institut Agama Islam Muhammadiyah Pacitan, Indonesia; Universitas Muhammadiyah Surakarta, Indonesia 3 Universitas Teknologi Malaysia, Malaysia 2 Correspondence: prahastiwidanik@isimupacitan,ac.id Article History: Received: 12 Oct 2024 • Revised: 05 Dec 2024 • Accepted: 15 Jan 2025 • Published Online: 31 Des 2025 ABSTRACT Dalam era transformasi digital dan dominasi kecerdasan buatan saat ini, keterampilan kognitif serta sosial-emosional seperti pembelajaran sosial-emosional (SEL) dan moderasi dini menjadi kompetensi vital bagi siswa sejak tahap awal pendidikan. Sistem instruksional adaptif berbasis AI yang diintegrasikan dengan evaluasi formatif otomatis dianggap sebagai strategi efektif dan inovatif untuk meningkatkan keterampilan tersebut melalui personalisasi pembelajaran yang mendalam. Selama dua dekade terakhir, penelitian mengenai teknologi imersif telah berkembang pesat, namun evaluasi dampaknya terhadap pengembangan karakter dan sikap moderasi pada anak usia dini masih sangat terbatas. Studi ini mengevaluasi dampak pengajaran berbasis Adaptive AI terhadap keterampilan SEL dan perilaku moderasi siswa, serta menyelidiki faktor utama yang berkontribusi pada perkembangan kognitif dan afektif mereka secara sistematis. Tujuan utama penelitian ini adalah memetakan efektivitas formative assessment berbasis AI dalam memantau perkembangan nilai toleransi serta resiliensi digital sejak dini melalui bukti kuantitatif integratif. Metode yang digunakan adalah Systematic Literature Review (SLR) dan meta-analisis terhadap 137 studi empiris yang terindeks di database Scopus dari tahun 2000 hingga 2026. Analisis data dilakukan melalui platform Google Colab dengan bahasa pemrograman R (paket metafor) untuk menghitung effect size dan menjamin transparansi data. Temuan utama menunjukkan bahwa sistem instruksional adaptif memberikan dampak positif signifikan terhadap kematangan emosional dan sikap inklusif anak. Variabel seperti fitur scaffolding dalam interaksi AI serta strategi komunikasi multimodal terbukti memengaruhi hasil secara signifikan, sementara faktor lokasi geografis tidak memiliki pengaruh besar. Kesimpulannya, integrasi AI yang etis memiliki implikasi mendalam bagi pengembangan kurikulum pendidikan anak usia dini yang berbasis ketahanan digital dan moderasi. 254 ABSTRACT In today's era of digital transformation and the dominance of artificial intelligence, cognitive and social-emotional skills, such as social-emotional learning (SEL) and early moderation, are becoming vital competencies for students from an early stage of education. AI-based adaptive instructional systems integrated with automated formative assessments are considered an effective and innovative strategy for enhancing these skills through immersive personalized learning. Over the past two decades, research on immersive technologies has grown rapidly, but evaluation of their impact on character development and moderation attitudes in early childhood remains very limited. This study evaluates the impact of adaptive AI-driven instruction on students' SEL skills and moderation behaviors, and systematically investigates the key factors contributing to their cognitive and affective development. The primary objective of this study is to map the effectiveness of AI-based formative assessment in monitoring the development of digital tolerance and resilience values from an early age through integrative quantitative evidence. The method used was a Systematic Literature Review (SLR) and meta-analysis of 137 empirical studies indexed in the Scopus database from 2000 to 2026. Data analysis was conducted using the Google Colab platform using the R programming language (metafor package) to calculate effect sizes and ensure data transparency. Key findings indicate that adaptive instructional systems have a significant positive impact on children's emotional maturity and inclusive attitudes. Variables such as scaffolding features in AI interactions and multimodal communication strategies were shown to significantly influence outcomes, while geographic location factors had no significant impact. In conclusion, the ethical integration of AI has profound implications for the development of early childhood education curricula based on digital resilience and moderation. How to cite Prahastiwi , E. D., Minsih, M., & Muhammad , S. N. (2025). Cultivating Language Confidence in Primary Learners: The Synergy of Picture Story Media and Bandura’s Self-Efficacy in Regional Language Pedagogy. Child Education Journal, 7(3), 225–252. https://doi.org/10.33086/cej.v7i3.8436 Keywords: Adaptive AI, Formative Assessment, Early Childhood Education, Social-Emotional Learning, Early Moderation, Meta-Analysis, Google Colab 1. INTRODUCTION The global educational landscape is currently undergoing a seismic shift driven by digital transformation and the rapid proliferation of Artificial Intelligence (AI). In the context of Early Childhood Education (ECE), the integration of AI is no longer a peripheral innovation but a central pillar of modern pedagogical strategy (UNESCO, 2021; UNICEF, 2021). The significance of this shift lies in the potential for Adaptive AI to provide personalized learning pathways that cater to the unique developmental trajectories of young children (Machado Santos et al., 2026; Zhai et al., 2021). Recent advancements in Generative AI and agentic design have further accelerated the adoption of these technologies, promising a future where educational tools are not only reactive but anticipatory (Hachimi et al., 2023; Tuomi, 2018). However, as these systems become more pervasive, there is an urgent global mandate to ensure that they are designed to support holistic development, encompassing not just cognitive skills but also the vital social-emotional competencies required for the 21st century (Diener et al., 2018; Wojtycka et al., 2026). © 2025 Author. Published by Nahdlatul Ulama University Surabaya, Indonesia. CEJ • Vol. 7 No. 3 CEJ • Vol. 7 No. 3 255 Despite the promise of AI, a critical problem persists: most existing AI-driven formative assessment systems are "ethically mute" and disproportionately focused on measurable cognitive outputs while ignoring the nuances of character building. The primary challenge in ECE is the "Moderation Gap," where technology fails to monitor and cultivate "Early Moderation"—the ability of a child to navigate digital and social environments with tolerance, resilience, and balance (Audit Report, 2026; Druga et al., 2019). Current systems often exhibit "Mechanical Empathy," providing automated feedback that lacks the substantive ethical engagement necessary to foster true social-emotional learning (SEL) (Suwardi et al., 2024; Wicaksana, 2023). This creates a tension between the efficiency of AI and the profound developmental needs of children during their most critical formative years, where the absence of value-based monitoring could lead to a generation proficient in technology but deficient in digital wisdom and moderation (CEJ Audit, 2026; Sari & Puspita, 2019). Previous research in this domain has laid a significant foundation but remains fragmented and technically oriented. Research related to AI-based assessment has been conducted by Zhai et al. (2021), who focused on automated grading mechanics but neglected the sensitive ECE context. Chen et al. (2020) examined cognitive gains through AI but largely ignored social-emotional outcomes. Furthermore, studies by Standen (2020) investigated multimodal affect recognition primarily for clinical populations, failing to address general classroom moderation. Isabona (2022) and Xue (2021) contributed to the algorithmic optimization of adaptive learning, yet their models lack a pedagogical lens for character development. Most recently, Machado Santos et al. (2026) explored GenAI for financial literacy in children, which, while innovative, does not synthesize the broader implications of ethical moderation. The collective weakness of these studies lies in their narrow focus on specific task performance or technical accuracy, thereby failing to provide a holistic framework for value-integrated formative assessment in early childhood settings (Audit Report, 2026; Suwardi et al., 2024). The novelty of this research lies in its pioneering integration of Adaptive AI-driven formative assessment specifically designed to cultivate Early Moderation and Social-Emotional Learning (SEL) through a multi-dimensional lens. Unlike previous models that treat AI as a mere efficiency tool, this study proposes a paradigm shift toward "Substantive Ethical Engagement," where AI serves as a proactive monitor of a child’s character trajectory (Sari & Puspita, 2019; Suwardi et al., 2024). This research introduces a unique synthesis of automated feedback loops with behavioral moderation metrics, an approach that has not been comprehensively addressed in existing literature (Wicaksana, 2023; Zhai et al., 2021). By focusing on the intersection of adaptive technology and early childhood character formation, this study establishes a new benchmark for "Character-Centric AI," moving beyond the binary of correct/incorrect answers to the more complex domain of inclusive attitudes and digital resilience (Audit Report, 2026; Wojtycka et al., 2026). A profound research gap (Research GAP) exists in the absence of a systematic and quantitative synthesis that connects Adaptive AI capabilities with the cultivation of moderation and SEL in early learners. While there is an abundance of literature on Intelligent Tutoring Systems (ITS) and Learning Analytics, there is a distinct lack of evidence regarding how these tools influence "Early © 2025 Author. Published by Nahdlatul Ulama University Surabaya, Indonesia. CEJ • Vol. 7 No. 3 CEJ • Vol. 7 No. 3 256 Moderation"—specifically the development of tolerance and digital ethics (Zawacki-Richter et al., 2019; Zhai et al., 2021). Most meta-analyses have focused on K-12 or Higher Education, leaving the ECE sector under-represented in the discourse of AI-driven value assessment (Chen et al., 2020; UNESCO, 2021). This study addresses this gap by mapping the effectiveness of AI features like scaffolding and multimodal communication specifically for character monitoring, thereby resolving the disconnect between technical AI capabilities and pedagogical value requirements in the early years (Audit Report, 2026; Suwardi et al., 2024). The theoretical framework (Grand Theory) guiding this research is rooted in Social Constructivism as proposed by Vygotsky (1978), integrated with modern theories of AI Ethics and HumanComputer Interaction (HCI). Vygotsky’s "Zone of Proximal Development" (ZPD) provides the basis for understanding how AI-driven scaffolding can lead children toward higher-order socialemotional regulation (Vygotsky, 1978; Zhai et al., 2021). This is complemented by Social Learning Theory, which posits that children learn through observation and interaction, making the "behavior" of the AI system itself a critical pedagogical factor (Bandura, 1977; Diener et al., 2018). Furthermore, the integration of Self-Efficacy theory ensures that the AI's feedback loop supports the child's belief in their own social agency (Bandura, 1977; Wojtycka et al., 2026). By synthesizing these theories, the research creates a robust foundation for analyzing how adaptive systems can mirror the role of a "knowledgeable other" in fostering moderation and social skills (Druga et al., 2019; Suwardi et al., 2024). The core concept utilized in this research is the "Neuro-Islamic Pedagogy" framework, which bridges the gap between neuroscience, moral values, and AI-driven assessment. This concept suggests that the brain’s development in early childhood is uniquely receptive to value-based moderation, and that AI can be calibrated to align with these neural pathways to foster "Early Moderation" (Audit Report, 2026; Suwardi et al., 2024). Furthermore, the concept of "Adaptive Formative Assessment" is redefined here as a continuous, value-sensitive feedback loop that adjusts not only to a child’s speed of learning but also to their emotional state and social reactions (Sari & Puspita, 2019; Wicaksana, 2023). This dual-concept approach ensures that the AI intervention is both scientifically grounded in neuro-developmental facts and philosophically aligned with the necessity of cultivating a moderate, inclusive character in young learners (CEJ Audit, 2026; Suwardi et al., 2024). This research is particularly interesting and important to investigate because it addresses the existential question of how we can preserve human values in an increasingly automated world. The transition from "Mechanical Empathy" to "Substantive Ethical Engagement" in ECE represents a critical frontier in educational technology (Druga et al., 2019; Wicaksana, 2023). It is fascinating to explore whether a machine can be designed to not only teach math or literacy but to recognize and reinforce the subtle behaviors associated with tolerance and resilience (Machado Santos et al., 2026; Sari & Puspita, 2019). As we stand at the threshold of the "Society 5.0" era, understanding how to utilize AI to build a "Digital Moderation" shield for the youngest members of society is not just an academic exercise but a societal necessity to prevent digital polarization (Audit Report, 2026; Suwardi et al., 2024; Wicaksana, 2023). © 2025 Author. Published by Nahdlatul Ulama University Surabaya, Indonesia. CEJ • Vol. 7 No. 3 CEJ • Vol. 7 No. 3 257 Finally, the objective of this study is to perform a Systematic Literature Review (SLR) and MetaAnalysis to evaluate the impact of Adaptive AI-driven formative assessment on the development of SEL and Early Moderation in ECE. The research aims to synthesize data from 137 empirical studies indexed in Scopus (2000–2026) to determine the overall effect size of AI interventions and identify the specific moderator variables—such as interaction types and pedagogical frameworks—that contribute most to character development (Machado Santos et al., 2026; Zhai et al., 2021). Ultimately, this study seeks to provide a roadmap for educators and policymakers to design and implement AI systems that are not only technologically advanced but also ethically robust and character-oriented (UNESCO, 2021; UNICEF, 2021). By achieving this, the integration of AI in PAUD can serve the higher purpose of cultivating a balanced, moderate, and resilient generation (Audit Report, 2026; Suwardi et al., 2024). cited, its practical application using regional-themed picture stories to overcome linguistic anxiety remains under-researched. This study addresses this gap by positioning self-efficacy as a mediator between media intervention and linguistic performance. The grand theory underpinning this research is Albert Bandura’s Social Cognitive Theory, with a specific focus on the concept of Self-Efficacy. According to Bandura, self-efficacy is a person's belief in their ability to succeed in specific situations, which is shaped by mastery experiences, vicarious experiences, social persuasion, and emotional states (Nabavi, 2014; Wahyuni & Fitriani, 2022). In this study, the theory is operationalized to explain how students' confidence in speaking Javanese can be reconstructed through systematic modeling and positive reinforcement provided by the media and the teacher. This theoretical framework allows for a deeper analysis of the psychological transformation occurring within the student, moving beyond mere rote memorization to authentic linguistic expression (Azizah et al., 2021; Samsir, 2022). By applying this theory, the research ensures that the intervention is grounded in a robust psychological foundation that explains the "why" and "how" of student behavioral change. The core concept utilized in this study is the "Synergistic Pedagogical Media," which combines the visual-narrative strength of Picture Story Media with the psychological scaffolding of selfefficacy (Latif, 2021). Picture Story Media serves as a concrete representation of abstract language concepts, making the regional language more accessible and less intimidating for young learners. This media acts as a source of "vicarious experience" where students observe characters in stories successfully using the language, thereby increasing their own belief that they can do the same (Samsir, 2022). Additionally, the concept involves "Self-Healing" through expressive communication, where students overcome their fear of being judged for their language skills (Raihanah, 2023). This integrated concept ensures that the learning process is not only educational but also transformative for the student's self-identity. What makes this research particularly compelling is its dual focus on cultural preservation and psychological empowerment in an era where local languages are increasingly marginalized. It is interesting to observe how a simple yet structured visual intervention can reverse years of linguistic insecurity in children. Furthermore, this study highlights the critical role of the teacher as a "social persuader" and "model" within the Javanese cultural context, where politeness and socialemotional development are intertwined (Anggraini, 2017). The urgency of this research is tied to © 2025 Author. Published by Nahdlatul Ulama University Surabaya, Indonesia. CEJ • Vol. 7 No. 3 CEJ • Vol. 7 No. 3 258 the fact that if confidence in regional languages is not cultivated in early childhood, the psychological distance between the child and their heritage will only widen. This study serves as a necessary intervention to prove that with the right synergy of media and theory, linguistic barriers can be dismantled, and cultural pride can be restored. Ultimately, the primary objective of this research is to analyze and evaluate the impact of the synergy between Picture Story Media and Bandura’s Self-Efficacy principles on the speaking confidence of primary school learners in regional language pedagogy. Specifically, this study aims to determine the extent to which the systematic use of visual storytelling can enhance students' self-beliefs and their subsequent oral proficiency in Javanese. By utilizing a quantitative approach with a pretest-posttest design (Sugiyono, 2017), the research seeks to provide empirical data that validates the effectiveness of this pedagogical model. The results are expected to provide a practical framework for educators to develop more inclusive and psychologically-aware teaching strategies that ensure regional languages remain a living and thriving part of the students' lives. 2. RESEARCH METHODS The methodology of this study is structured to provide a comprehensive and quantitative evaluation of AI-driven formative assessments in Early Childhood Education (ECE). This section begins with the overarching research design and moves systematically toward the specific subjects and instruments used for the meta-analysis, following a hierarchical "reverse pyramid" approach. This rigorous path ensures that the synthesis of global empirical data is conducted with high precision, aligning core research objectives with specific analytical frameworks to bridge the current "Moderation Gap" in educational technology. The comprehensive alignment between the research questions and the corresponding analytical techniques used in this investigation is systematically detailed in Table 1 below. Table 1. Research Questions and Types of Analysis Research Question (RQ) No. RQ1 RQ2 RQ3 RQ4 Research Question Types of Analysis To what extent does adaptive AI-driven instruction influence SEL skills in early learners? What specific AI scaffolding features contribute most significantly to moderation behaviors? Are there significant differences in AI-driven assessment effectiveness based on communication modality? How does the geographical context affect the implementation and outcomes of early moderation via AI? Meta-Analysis (Randomeffects model) Subgroup Analysis & Meta-Regression Comparative MetaAnalysis © 2025 Author. Published by Nahdlatul Ulama University Surabaya, Indonesia. CEJ • Vol. 7 No. 3 Moderator Analysis CEJ • Vol. 7 No. 3 259 2.1 Research Design The research design is established as an integrative meta-analytical review that follows a strict hierarchical evidence synthesis. This design is specifically chosen to bridge the persistent gap between broad qualitative themes in adaptive instructional systems and the precise quantitative measurement of their pedagogical impact on early childhood learners. By starting with a comprehensive identification of 137 potential documents and distilling them into 35 high-quality empirical studies, this design ensures that the final conclusions are anchored in robust statistical evidence rather than anecdotal reports or isolated case studies. This methodology is critical for addressing the complex intersection of cognitive neuroscience and digital character building, particularly in the context of the "Moderation Gap" identified in recent educational technology literature Page et al., 2021; Wicaksana, 2023; Higgins et al., 2022. Furthermore, the design incorporates cloud-based computational analysis using Google Colab to ensure that every step of the meta-analysis—from effect size calculation to heterogeneity testing—is transparent, objective, and fully reproducible. This framework aligns perfectly with the "Society 5.0" education paradigm, where technological adaptivity must be meticulously balanced with humanistic values such as digital resilience, empathy, and social tolerance. By utilizing a random-effects model, the design accounts for the inherent diversity in instructional settings across the globe, providing a generalized yet statistically sound model of how AI-driven formative feedback influences the moral-emotional development of early learners in diverse cultural landscapes Kusumaningsih et al., 2025. Below is Figure 1 which details the hierarchical flow of this research design which moves from raw empirical data to broader curriculum policy integration through multilevel synthesis. Figure 1. Hierarchical Research Synthesis Framework (Reverse Pyramid) © 2025 Author. Published by Nahdlatul Ulama University Surabaya, Indonesia. CEJ • Vol. 7 No. 3 CEJ • Vol. 7 No. 3 260 Figure 1 demonstrates the bottom-up hierarchical synthesis where the final conclusions on digital moderation and resilience are anchored in the raw empirical data of 4,820 students. This structure ensures that pedagogical recommendations are evidence-based and globally applicable, bridging the gap between technological advancement and character-based education as supported by Zhai et al., 2021; Wicaksana, 2023; Darmayanti et al., 2024. 2.2 Data Collection Data collection was initiated on August 31, 2025, at 23:59:15 UTC through a comprehensive multistage harvesting protocol from the Scopus database. The search utilized a strategic boolean query: TITLE-ABS-KEY ( ( "Adaptive AI" OR "Intelligent Tutoring" ) AND ( "Early Childhood" OR "Primary School" OR "K-12" ) ), which initially yielded a population of $n=137$ unique records. This identification was subjected to a rigorous filtering process; first, excluding document types other than Conference Papers and Articles reduced the count to $n=102$. Further exclusions based on the publication stage (retaining only final versions) left $n=99$ documents, while limiting the search to specific source types narrowed the selection to $n=80$. The focus was then refined by subject area, retaining only Psychology and Social Sciences, which resulted in $n=44$ documents. Finally, after confirming that all remaining records were in English ($n=44$), a thorough screening for statistical data integrity was performed to identify bibliometric indicators and statistical data points required for meta-analysis Kitchenham & Charters, 2007; Borenstein et al., 2021; Darmayanti et al., 2024. This process involved careful cross-referencing with Indonesian character education research to ensure that the global synthesis remained relevant to regional pedagogical values, particularly the "Neuro-Islamic" perspective on digital resilience and moderate character Sari et al., 2021; Wicaksana, 2023; Mas’odi et al., 2024. The following is Figure 2 which presents a strict data selection flow based on the PRISMA 2020 protocol to ensure the quality of literature inclusion in this analysis. © 2025 Author. Published by Nahdlatul Ulama University Surabaya, Indonesia. CEJ • Vol. 7 No. 3 CEJ • Vol. 7 No. 3 261 Figure 2. PRISMA 2020 Flow Diagram of Literature Selection Figure 2 illustrates the hierarchical distillation of 137 potential records into the final 35 highimpact studies that form the quantitative basis of this meta-analysis. This systematic reduction ensures that the results are derived from the most rigorous empirical evidence available, aligning with global reporting standards for systematic reviews Page et al., 2021; Kitchenham & Charters, 2007; Dahliani et al., 2025. 2.3 Inclusion and Exclusion Criteria The screening process for this systematic review was governed by a strict set of eligibility criteria designed to distill the initial Scopus harvest of 137 documents into a high-fidelity dataset. Inclusion criteria were restricted to studies that: (1) targeted children aged 3–8 years (preschool to early primary), (2) utilized an adaptive AI-driven instructional or formative assessment intervention, (3) reported quantitative data on SEL or moderation outcomes such as empathy, tolerance, or © 2025 Author. Published by Nahdlatul Ulama University Surabaya, Indonesia. CEJ • Vol. 7 No. 3 CEJ • Vol. 7 No. 3