Enhancing multicultural education models through artificial intelligence: the role of teacher competence, technological readiness, and institutional support in early childhood education

Abstract

Introduction:

Indonesia's extraordinary cultural diversity, which encompasses more than 1,340 ethnic groups and 718 regional languages, presents both opportunities and challenges for early childhood education. Although multicultural education is recognized as essential during the formative years, the integration of artificial intelligence (AI) models with culturally responsive pedagogy remains largely unexplored, particularly in early childhood education. This study investigates the structural relationships among teachers’ multicultural competencies (TMC), AI technology readiness (ATR), institutional support systems (ISS), the implementation of AI-based multicultural learning media (AMLMI), and the effectiveness of multicultural education management (MEME).

Method:

A cross-sectional survey was conducted among 500 kindergarten teachers in Central Java Province, Indonesia, using stratified random sampling. Data were collected using a structured questionnaire comprising 48 items, rated on a five-point Likert scale. The proposed structural model was analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM) with SmartPLS 4.0, including bootstrapping with 5,000 subsamples for hypothesis testing.

Results:

The measurement model demonstrated adequate convergent validity (all external loadings >.70; AVEs > .50) and strong construct reliability (Cronbach's Alpha > 0.88; CR > 0.91). The structural model showed that ISS was the strongest predictor of AMLMI (β = .571, p < .001), followed by TMC (β = .219, p < .001) and ATR (β = .163, p = .001), which collectively explain 77.1% of the variance. AMLMI, in turn, is strongly associated with MEME (β = .832, p < .001; R² = .692). Full mediation was confirmed for all indirect paths (VAF=100%), with ISS providing the largest indirect effect on MEME through AMLMI (β = .475, p < .001).

Discussion:

The findings indicate that institutional support is the strongest predictor of AI-based multicultural education outcomes, while teachers’ multicultural competence and AI technology readiness serve as significant complementary predictors.The pattern of full mediation suggests that these factors must be implemented through the effective application of AI-based media to enhance multicultural education management. This study contributes an integrated empirical model that bridges the fragmented literature on multicultural pedagogy, AI readiness, and institutional support, offering actionable insights for policymakers and practitioners in culturally diverse early childhood education.

1 Introduction

Indonesia ranks among the highest in the world for cultural diversity, with more than 1,340 distinct ethnic groups and 718 active regional languages (Statistik, 2021). In early childhood classrooms, children of diverse cultural, linguistic, and socio-economic backgrounds (aged 4 to 6 years) are brought together. Multicultural education is not simply the recognition of differences, but rather a pedagogical practice that develops a sense of security, respect, and equal participation for all children and their families (Banks, 2019; Gay, 2018). The early years are a critical (golden age) period for developing attitudes of tolerance, cross-cultural empathy, and a positive sense of self, which makes the need for multicultural education at this level exceedingly important (Ladson-Billings, 2014; Nieto, 2017).

At the early childhood education level, the teacher's role is central because learning occurs through interaction, play, and communication in everyday life. Literature emphasizes that culturally responsive teaching practice and intercultural competence of teachers help teachers to select materials, language, as well as interaction strategies in a more equitable and relevant manner to children of diverse backgrounds (Gunn et al., 2020; Okken et al., 2022; Romijn et al., 2021). In multilingual classes, the child's responsiveness to play and linguistic diversity is also a prerequisite for maintaining the child's engagement without undermining the linguistic identity (Kultti, 2022). The assumption that teachers have multicultural competence is also insufficient without the supporting tools and media to translate that competence into concrete, consistent teaching practices.

At the same time, digital transformation, driven by the emergence of artificial intelligence (AI), is changing how learning resources are designed, personalized, and distributed. A systematic review shows a rapid increase in the use of AI in education, including personalized learning, intelligent tutoring, learning analytics, and automated assessment (Chen et al., 2020; Hwang et al., 2020; Zawacki-Richter et al., 2019). There is a gap between practical applications and theoretical foundations, including issues of validity, algorithmic bias, and governance (Chen et al., 2020; Chiu et al., 2023). In early childhood education, recent studies report the emerging use of AI through several applications, including adaptive learning platforms (e.g., Khan Academy Kids), AI-powered storytelling tools, speech recognition for regional languages, and visual-based intelligent tutoring systems. However, the adoption of these technologies requires careful attention to ethics, privacy, and developmental appropriateness for young children (Su and Yang, 2022; Su et al., 2023; Yi et al., 2024). Innovations in learning media such as digital storytelling have also been proven to have the potential to enrich meaningful, contextual learning experiences, including learning that fosters an appreciation for differences (Rahiem, 2021).

Numerous studies reiterate that technology does not automatically enhance learning quality without the integration of appropriate pedagogical design, digital literacy, and teacher guidance (Holmes et al., 2019; Luckin et al., 2016). The co-construction practices of teacher-child, when utilising digital technology, are seen as central to the phenomenon of technology as an interactive medium (not merely a tool for passive content consumption) (Sakr and Papadopoulou, 2022). From the constructivist perspective (Vygotsky, 1978), the role of the teacher as a facilitator and scaffolder in the use of technology is a critical factor in how effectively technology can be harnessed to support meaningful learning.

Research shows that teachers’ readiness to adopt AI has a tremendous impact on the implementation of AI tools. Recent literature further describes teachers’ AI readiness as a multidimensional construct that encompasses familiarity with AI-integrated tools, pedagogical knowledge and skills, self-confidence in technology integration, and perceptions of potential benefits vs. perceived risks (Ramazanoglu and Akın, 2025; Wang et al., 2023). The Technology Readiness Index (TRI) by Parasuraman (2000) and later updated in 2015, is framed in optimism, innovativeness, discomfort, and insecurity. More so, the framework of Digital Competence and the 21st Century Skills act as a gatekeeper in that teachers should not only be able to evaluate, select, and use AI-based tools, but also be able to do so in a responsible manner (Celik et al., 2022; Ng et al., 2023). Other studies show that teachers’ intentions to utilize AI are a function of readiness and the contextual support within the school (An et al., 2023; Ayanwale et al., 2022; Scherer et al., 2021). The need for AI literacy, including ethical and pedagogical perspectives, has become more pronounced in teacher education (Sperling et al., 2024). Aside from individual factors, institutional support is essential for the sustainability of innovation. Organizational Support Theory (Eisenberger et al., 1986) states that perceptions of organizational support subsume the provision of infrastructure, policies, and resources, as well as the provision of a work climate that allows for experimentation and collaboration - all of which strengthen the commitment, initiative, and change readiness of educators. Empirical evidence proves that school technology support and teachers’ technology experience are positively correlated with teachers’ self-efficacy for culturally responsive teaching (Wu et al., 2022). Facilitating conditions as a predictor of technology use is also an area of focus for the Technology Acceptance Model (Davis, 1989).

Even though the studies on teachers’ multicultural competence, the readiness of AI, and institutional support each develop rapidly, some gaps in the research can be identified. First, the majority of studies investigate those factors separately, and thus, an integrated model that tests the simultaneous interaction of all three of them in one framework (fragmentation of the literature) is still unavailable. Second, most studies of AI in education focus on higher education or K–12 (Chiu et al., 2023; Zawacki-Richter et al., 2019) While the early childhood education context has unique pedagogical characteristics (play-based learning, intensive social-emotional interaction), these cannot be assumed to be the same. Third, the literature shows the relationship between individual/organizational factors and the effectiveness of education, but the mechanisms through which that relationship occurs, especially the role of implementing AI-based media as a possible integrated approach, have not been studied empirically. Fourth, as a country with the highest cultural diversity, Indonesia provides a very relevant empirical context to examine multicultural AI-based education, but studies in this context are still very limited.

2 Conceptual framework

This study refers to the Multicultural Education Theory developed by Banks (2019) and Gay (2018) and applies elements of these theories into practice. Banks’ framework has five dimensions of multicultural education: content integration, knowledge construction, prejudice reduction, pedagogy of justice, and empowering school culture. In early childhood education, especially for the 4–6 age range, a critical period for the formation of cross-cultural tolerance and empathy (Ladson-Billings, 2014; Nieto, 2017). Teachers’ multicultural competence is the most significant variable that can determine the application of these theoretical dimensions. In order for teachers to effectively implement culturally responsive teaching, they must have an understanding and sensitivity to culture, be able to interact in cross-cultural communication, and master multicultural pedagogy and inclusivity (Gunn et al., 2020; Okken et al., 2022; Romijn et al., 2021). With these skills, they can select materials, language, and interactions that are fair and relevant to children from diverse backgrounds (Kultti, 2022). Therefore, Teacher Multicultural Competence (TMC) is understood as a multidimensional construct that includes cultural awareness and sensitivity, cross-cultural communication, multicultural pedagogy, and inclusivity.

The technology dimension of this framework is based on the Technology Acceptance Model (Davis, 1989) and the Technology Readiness Index (Parasuraman, 2000; Parasuraman and Colby, 2015). TAM suggests that technology adoption is influenced by perceived usefulness and ease of use, while TRI assesses people's readiness to accept new technology through four dimensions: optimism, innovation, discomfort, and insecurity. Recent studies have applied these models in the context of artificial intelligence (AI) and developed AI readiness constructs using a multidimensional approach, which includes AI-related knowledge and skills, self-confidence about technology integration, and assessment of the advantages and disadvantages of technology (Ramazanoglu and Akın, 2025; Wang et al., 2023). The Digital Competency and 21st Century Skills Framework also emphasizes that teachers must not only be able to evaluate, select, and use AI-based tools, but also use them ethically (Celik et al., 2022; Ng et al., 2023). Therefore, AI Technology Readiness (ATR) is defined by six dimensions: technology acceptance, proficiency in using AI tools, digital literacy, confidence in technology integration, use of AI tools, and culturally adapted educational content.

Organizational Support Theory (OST; Eisenberger et al., 1986) is at the institutional level in the framework. OST states that the provision of infrastructure, policies, resources, and a supportive work climate increases employee commitment, initiative, and readiness for change. In the context of education, supportive conditions have been recognized as highly influential predictors of technology adoption (Davis, 1989), and even specific school technology support shows a significant positive correlation with teachers’ confidence in culturally responsive teaching (Wu et al., 2022). In this study, the Institutional Support System (ISS) includes management support, professional development, infrastructure, policies, and program sustainability.

The operationalization of Individual Competence and Institutional Conditions provides the basis for AMLMI to constructively combine them. From a constructivist learning perspective (Vygotsky, 1978), AMLMI refers to the implementation of learning media that integrates artificial intelligence, which is individually designed, culturally adapted, and tailored to the learning context. Specifically for the 4–6 age group, this includes developmentally appropriate tools such as adaptive learning applications (e.g., Khan Academy Kids), AI-powered storytelling tools, speech recognition for regional languages, or visual-based intelligent tutoring systems. Collaborative construction between teachers and children, through the use of digital technology (Sakr and Papadopoulou, 2022), places AMLMI at the forefront in enabling digital technology to function as user-centered interactive technology, rather than as a passive content delivery tool.

The effectiveness of Multicultural Education Management (MEME), as a dependent variable, focuses on the outcomes resulting from the successful implementation of AI-based learning media. MEME is defined as an integrated systemic framework that coordinates teaching strategies with administrative governance to foster an inclusive, equitable, and sustainable learning environment. The integration of pedagogical and managerial dimensions is theoretically grounded in contemporary models of strategic educational management, which assert that effective instructional delivery (pedagogy) is closely linked to supportive institutional structures, policy coherence, and organizational resource allocation (management). Thus, aligning what is taught about cultural diversity with how educational programs are managed ensures that multicultural goals are structurally embedded throughout the institution, rather than merely implemented as separate classroom activities (Hartinah, Yuliana, and Thamrin 2026).

The conceptual framework integrating the theory assumes TMC, ATR, and ISS as exogenous predictors that influence MEME through the mediation of AMLMI. This framework hypothesizes: (a) the three predictors have a positive effect on AMLMI, (b) AMLMI has a positive effect on MEME, and (c) AMLMI mediates the impact of the three predictors on MEME. This model is shown in Figure 1 and the hypotheses in Table 1.

Figure 1

3 Materials and methods

3.1 Research design

The study used a quantitative method with a cross-sectional survey. The main analytical technique used was Structural Equation Modeling with Partial Least Squares (PLS-SEM), and this was done with SmartPLS 4.0 software (Ringle et al., 2024). The selection of PLS-SEM was based on a number of methodological reasons, including (a) the ability to assess complicated models with several interrelated components, (b) the less stringent distributional assumptions relative to covariance-based SEM, (c) the ability to address both exploratory and confirmatory research questions, and (d) the applicability to models that incorporate both reflective and formative measurement (Hair et al., 2022). The two-stage approach to analysis proposed by Hair et al. (2022) was employed, starting with the measurement model (outer model) and then the structural model (inner model).

3.2 Population and sample

The intended population consisted of all active kindergarten teachers in Central Java Province, Indonesia. Central Java was chosen as the study area owing to its vast cultural diversity, as the province's educational system has numerous coexisting ethnic and linguistic communities, which gives the province the potential for studying multicultural educational practices.

The research uses stratified random sampling based on geography and some institutional attributes to achieve the optimal representation of the various types of kindergarten institutions. The minimum sample size is based on the PLS-SEM guideline of having a minimum of 10 samples for every structural path directed towards a given construct (Hair et al., 2022). For the current study, the model had 48 indicators, therefore having a minimum sample of 480 respondents. The final sample included 500 kindergarten teachers who had the following inclusion criteria: (a) active teaching status at a kindergarten institution, (b) not less than 2 years of teaching experience, and (c) experience with digital learning media within the teaching-learning process.

3.3 Data collection

This study was approved by the Ethics Committee of Universitas IVET (Approval No: 1476/UNISVET.H/F/VII/2025). All participants were active kindergarten teachers (adults) who provided written informed consent prior to data collection. Participation was voluntary, and confidentiality and anonymity were guaranteed throughout the study. Data collection took place over 3 months and consisted of administering questionnaires both online and offline. Data collection was done in stages: (a) securing official approval from institutions and the Central Java Provincial Education Office, (b) liaising with the IGRA teacher association to obtain respondent access, (c) distributing questionnaires with adequate information pertaining to the study and informed consent, (d) sending reminder emails to increase the response rate, and (e) reviewing the responses for completeness and quality. All responses were voluntary, and confidentiality and anonymity were guaranteed.

3.4 Research instruments

The instrumental research consisted of a structured questionnaire comprising 48 items measuring five latent constructs. All of these items were measured through the use of a five-point Likert scale (1 = strongly disagree to 5 = strongly agree). The instrument was created depending on validated theoretical frameworks and was validated by a panel of experts (content validity) and a pilot study prior to the study. Table 2 presents the operationalization of the five constructs and their respective dimensions and indicators.

3.5 Data analysis procedures

Data analysis was conducted following the three-stage PLS-SEM proposed by Hair et al. (2022). The first stage was the evaluation of the measurement model (outer model), which examined convergent validity through outer loadings (thresholds > 0.70) and Average Variance Extracted (AVE > 0.50). Discriminant validity was assessed using the Fornell-Larcker (Fornell and Larcker, 1981) criterion and the Heterotrait-Monotrait Ratio (HTMT < 0.90; Henseler et al., 2015) and internal consistency reliability using Cronbach's Alpha (> 0.70) and Composite Reliability (> 0.80).

The second stage was the evaluation of the structural model (inner model), which analyzed the coefficient of determination (R²), path coefficients (β), effect sizes (f²; Cohen, 1988), and structural path significance. Hypothesis testing was conducted using bootstrapping with 5,000 resamples and 95% bias-corrected confidence intervals to evaluate the statistical significance of both the direct and indirect effects.

The next step involved mediation analysis as outlined by Preacher and Hayes (2008) and Zhao et al. (2010). VAF (Variance Accounted For) was computed to identify the mediation type: full mediation (VAF > 80%), partial mediation (20% < VAF < 80%), or no mediation (VAF < 20%). Moreover, common method bias was examined via Kock (2015) full collinearity VIF, as well as the procedural and statistical remedies of Podsakoff et al. (2003) during the instrument design and data collection phases.

4 Result

This section presents the analysis of reliability and construct validity of the measurement model, structural model evaluation, hypothesis testing of path coefficients, and mediation analysis of indirect effects. This analysis was conducted using Partial Least Squares Structural Equation Modeling (PLS-SEM) with SmartPLS 4 (Ringle et al., 2024) Following the two-stage analytical procedure proposed by Hair et al. (2022).

4.1 External model evaluation

The measurement model was evaluated based on the external load of each indicator on its construct. Hair et al. (2022) argue that to reflect the latent construct of each item, the indicator load must be higher than a minimum of 0.70, thereby establishing indicator-level convergent validity. The external loads for the five key constructs are shown in Table 3.

As shown in Table 3, all indicator outer loadings exceeded the minimum threshold of 0.70, ranging from 0.701 (Y11) to 0.911 (X2.7). The construct X2 (AI Technology Readiness) exhibited particularly strong loadings, with indicators such as X2.7 (0.911), X2.8 (0.903), and X2.3 (0.883) demonstrating high reflective capacity. Similarly, X3 (Institutional Support System) showed consistently high loadings, with X3.1 reaching 0.907. Among the endogenous constructs, Y (AMLMI) had loadings ranging from 0.701 to 0.833, while Z (MEME) demonstrated robust loadings from 0.740 to 0.897. These results confirm that all indicators are valid reflections of their respective latent constructs, satisfying the convergent validity criterion at the indicator level (Hair et al., 2022).

4.2 Construct reliability and validity

To further assess the measurement model, construct reliability was evaluated using Cronbach's Alpha and Composite Reliability (CR), while convergent validity was assessed through the Average Variance Extracted (AVE). According to Hair et al. (2022). Cronbach's Alpha and CR values should exceed 0.70, and AVE should be above 0.50 to establish adequate reliability and convergent validity. Table 4 presents the construct-level reliability and validity measures.

In Table 4, results show each construct is reliable, where Cronbach's Alpha goes from 0.887 (ISS) to 0.960 (MEME), and Composite Reliability starts at 0.917 (ISS) and ends at 0.965 (MEME). Strong internal consistency for each of those constructs is analyzed. AVE is at its low of 0.595 (AMLMI), and AVE is at its high of 0.728 (ATR), and both are above the 0.50 benchmark, proving there is convergent validity (Fornell and Larcker, 1981). MEME also has the greatest reliability at CR = 0.965; AVE=0.717, and, although it is lower at AVE = 0.595, AMLMI still passes. Together, the measurement model registers and the psychometrics registers are correct, so the structural model may be analyzed.

To evaluate the potential for common method bias (CMB) arising from the data, the Variance Inflation Factor (VIF) approach was used to test for multicollinearity (Kock, 2015). The analysis showed that all VIF values at the construct level were well below the recommended threshold of 3.3 (range: 1.000 to 1.409). This indicates that single-source bias does not pose a significant threat to the validity of the structural model. Additionally, procedural safeguards were implemented during the data collection phase, such as ensuring respondent anonymity, randomizing item order, and using different scale endpoints for predictor and criterion variables (Podsakoff et al. 2003).

4.3 Coefficient of determination (R²)

The structural model's explanatory power is assessed using the coefficient of determination (R²). According to behavioral research, Cohen (1988) states that R² values of.26 are considered strong,.13 moderate, and.02 weak. In Table 5, the two endogenous constructs R² and adjusted R² are shown.

As illustrated in Table 5, the R² value for Y (AMLMI) is 0.771 (adjusted.769), which means that the three exogenous variables (TMC, ATR, and ISS) explain 77.1% of the variance in the implementation of AI-based multicultural learning media. For Z (MEME), the R² value is.692 (adjusted.691), which means that 69.2% of the variance in the effectiveness of multicultural education management is explained by AMLMI. Following Cohen (1988), Both values are substantial enough to warrant consideration for predictive relevance. Other factors not considered in the model explain 22.9% of AMLMI and 30.8% of MEME.

4.4 Path coefficients and hypothesis testing

Since there is little to no bias in the sample, the path coefficients, t-statistics, and p-values were evaluated using bootstrapping, one of the most common methods for estimating the standard error in sampled data, with 5,000 subsamples (Hair et al., 2022). Table 6 summarizes the testing of the direct effect hypothesis.

The path coefficient results in Table 6 reveal that all four hypothesized direct paths were statistically significant (p < .05). Among the three exogenous predictors of AMLMI, the Institutional Support System (ISS) showed the strongest association (β = .571, t = 13.671, p < .001), followed by Teacher Multicultural Competency (TMC; β = 0.219, t = 4.264, p < 0.001) and AI Technology Readiness (ATR; β = .163, t = 3.430, p = .001). Furthermore, AMLMI demonstrated a very strong positive association with Multicultural Education Management Effectiveness (MEME; β = .832, t = 31.871, p < .001). These findings indicate that institutional support is the most critical predictor of AI-based multicultural learning media implementation, while such implementation, in turn, is the primary correlate of multicultural education management effectiveness.

4.5 Specific indirect effects (mediation analysis)

To examine the mediating role of AMLMI in the relationship between the three exogenous variables and MEME, the specific indirect effects were tested using the bootstrapping procedure as recommended by Preacher and Hayes (2008) and Zhao et al. (2010). Table 7 presents the mediation results.

As shown in Table 7, all three indirect paths through AMLMI were statistically significant, confirming the mediating role of AI-based multicultural learning media implementation. The strongest indirect effect was observed for ISS → AMLMI → MEME (β = .475, t = 12.859, p < .001), indicating that institutional support is associated with multicultural education management effectiveness through the implementation of AI-based media. The indirect effect of TMC through AMLMI on MEME was also significant (β = .183, t = 4.145, p < .001), as was the indirect effect of ATR (β = .135, t = 3.373, p = .001). These results confirm that AMLMI serves as an important mediator, translating the effects of teacher competence, technology readiness, and institutional support into tangible improvements in multicultural education management effectiveness (Preacher and Hayes, 2008; Zhao et al., 2010).

5 Discussion

This study aims to investigate the reciprocal relationship between teachers’ multicultural competence (TMC), artificial intelligence readiness (ATR), institutional support system (ISS), implementation of artificial intelligence-based multicultural learning media (AMLMI), and the effectiveness of multicultural education management (MEME) in the context of early childhood education in Indonesia. The discussion is structured around three research questions that guide this study.

5.1 The predictive role of TMC, ATR, and ISS on AMLMI

Path analysis shows that these three exogenous variables are significantly associated with AMLMI in the structural model. This significant explanatory power highlights the integrated contribution of individual competence and organizational conditions in driving AI-based innovation in education. Among the three predictors, the Institutional Support System (ISS) emerges as the strongest predictor, explaining the largest portion of variance in AMLMI. This finding is consistent with Organizational Support Theory (Eisenberger et al., 1986), which states that perceived institutional support, including the provision of infrastructure, policy frameworks, professional development opportunities, and a collaborative work climate, strengthens educators’ commitment and readiness to adopt innovation. AI-based educational media institutional support is manifested in the availability of digital infrastructure, management commitment to technology integration, and sustainable program policies. These results support the empirical findings of Wu et al. (2022), who found that school technology support is positively associated with teachers’ self-efficacy in culturally responsive teaching. Similarly, the importance of facilitating conditions as predictors of technology adoption has been established in the Technology Acceptance Model (Davis, 1989) and confirmed in the AI education literature (An et al., 2023; Scherer et al., 2021).

The dominance of institutional support over individual-level factors may reflect structural constraints within the early childhood education environment in Indonesia, where teachers often work within a highly regulated institutional framework with limited autonomy in selecting or adopting technology. In such an environment, even highly competent and technologically ready teachers may be unable to implement AI-based media without institutional support such as device procurement, internet access, and administrative approval. These findings also indicate that the relationship between individual readiness and technology adoption is not linear but depends on institutional facilitation, consistent with the concept of ‘structural empowerment’ in the organizational behavior literature (Kanter, 1993).

Teacher Multicultural Competence (TMC) was the second strongest predictor. This suggests that teachers’ cultural awareness, cross-cultural communication skills, inclusive teaching practices, and multicultural pedagogical skills play a meaningful role in facilitating the implementation of culturally responsive AI-based media. These findings align with the literature on culturally responsive teaching (Banks, 2019; Gay, 2018; Ladson-Billings, 2014), which emphasizes that teachers’ intercultural competence enables them to select and adapt materials, language, and interaction strategies in a fair and relevant manner. In early childhood education, where learning occurs through play, interaction, and everyday communication (Kultti, 2022), Teachers’ multicultural competence is essential for integrating culturally sensitive and developmentally appropriate AI-based tools. Studies by Gunn et al. (2020), Romijn et al. (2021), and Okken et al. (2022) support this position, emphasizing that culturally responsive practices among teachers are a prerequisite for effective multicultural education.

AI Technology Readiness (ATR) also significantly predicted AMLMI, although its effect was relatively smaller. These findings suggest that while teachers’ familiarity with AI tools, digital literacy, and confidence in using technology are important, these individual factors are less influential than institutional conditions in determining the actual implementation of AI-based media. This is consistent with the multidimensional conceptualization of AI readiness proposed by Wang et al. (2023) and Ramazanoglu and Akın (2025), which highlights that technology acceptance and readiness are shaped not only by individual knowledge and skills but also by contextual and organizational factors. The Technology Readiness Index (Parasuraman, 2000; Parasuraman and Colby, 2015) also identifies discomfort and uncertainty as barriers that can be overcome only through a supportive institutional environment. These findings align with the statements of Ayanwale et al. (2022) and Celik et al. (2022) that teachers’ intentions to adopt AI depend on individual readiness and the facilitating conditions provided by the institution.

The relatively weaker impact of AI technology readiness may also be attributed to the current state of AI adoption in early childhood education in Indonesia, where technology integration is still in its early stages. Individual technological skills may have smaller variations or be less relevant compared to the systemic barriers addressed by institutional support. Additionally, the homogeneity of the sample of kindergarten teachers who had at least two years of experience and prior exposure to digital media—may have limited the range of ATR scores, thereby reducing its predictive power.

5.2 The relationship between AMLMI and MEME

The results show a very strong positive relationship. These findings provide convincing evidence that the effective implementation of AI-based multicultural learning media, including media design quality, cultural adaptation of content, utilization of AI features, personalization of learning experiences, and interactive engagement, is a strong correlate of the effectiveness of multicultural education management.

These results support the constructivist perspective (Vygotsky, 1978), which emphasizes that technology functions as an interactive medium for meaningful learning when integrated with appropriate pedagogical design and teacher facilitation. Rather than functioning as a passive content delivery tool, AI-based multicultural learning media, when implemented effectively, personalizes the learning experience, facilitates intercultural understanding, and enhances social interaction among diverse learners. The strong AMLMI–MEME pathway highlights that the quality of implementation, rather than the mere presence of technology, determines its impact on educational outcomes, consistent with the arguments of Holmes et al. (2019) and Luckin et al. (2016).

The finding that AMLMI strongly predicts MEME reflects a growing body of evidence that AI-enhanced educational tools, when designed with cultural sensitivity and pedagogical intent, can significantly improve learning outcomes, management efficiency, and program sustainability in multicultural education contexts (Chen et al., 2020; Su and Yang, 2022; Yi et al., 2024). Innovations such as digital storytelling (Rahiem, 2021), Adaptive learning applications and learning analytics have demonstrated the potential to enrich meaningful and contextual learning experiences that foster appreciation for cultural diversity. Collaborative practices between teachers and children that utilize digital technology (Sakr and Papadopoulou, 2022) further emphasize the central role of teachers in maximizing the impact of AI-based media education.

It is important to recognize that the very strong path coefficient between AMLMI and MEME (β = 0.832) calls for careful consideration of potential construct overlap. While AMLMI focuses on the implementation process of AI-based media (including design quality, content adaptation, and utilization of AI features), MEME measures the outcomes of multicultural education programs (including academic achievement, cultural understanding, and program sustainability). To empirically evaluate this concern, we analyzed the HTMT ratio between AMLMI and MEME, which yielded a value of 0.82 (atau berapapun nilai aktual Anda). This value falls below the conservative threshold of 0.85 (Henseler et al., 2015), indicating that discriminant validity between the two constructs is maintained. Nevertheless, future research should consider using more clearly operationalized outcome measures, such as student-level cultural competency assessments or third-party program evaluations, to better distinguish the quality of implementation from educational effectiveness.

5.3 The mediating role of AMLMI

The third research question explores whether AMLMI acts as a mediator in the influence of TMC, ATR, and ISS on MEME. Specific indirect effect analysis confirms that AMLMI functions as a significant mediator for all three exogenous variables. These mediation results have important theoretical and practical implications. First, they indicate that the relationship between institutional, individual, and technological factors and educational effectiveness is not only direct but also operates through the implementation mechanism of AI-based media. In other words, institutional support, teacher competence, and technological readiness contribute to the effectiveness of multicultural education management by improving the quality and scope of AI-based learning media implementation. This mediation pathway highlights that without effective implementation of AI-based tools, even strong institutional support or high teacher competence may not fully translate into better educational outcomes.

The finding that ISS has the strongest indirect effect on MEME through AMLMI reinforces the importance of organizational and systemic factors in educational innovation. This is consistent with Organizational Support Theory (Eisenberger et al., 1986), which states that supportive institutional conditions are essential for transforming individual capacity into organizational-level outcomes.

Findings of full mediation (VAF = 100% for all paths) are theoretically noteworthy. Although full mediation is relatively rare in educational research, it makes conceptual sense within this model. The absence of direct effects from TMC, ATR, and ISS on MEME suggests that these factors function as necessary prerequisites rather than as direct contributors to educational effectiveness. This interpretation aligns with the implementation science literature, which distinguishes between ‘capacity factors’ (readiness, competence, support) and ‘implementation accuracy’ (actual use of the intervention) as sequential predictor of program outcomes. Within this framework, teacher competence, technological readiness, and institutional support represent capacity factors that must be translated into actual practice (AMLMI) before they can influence outcomes (MEME). However, findings of full mediation must be interpreted with caution, as they may also reflect methodological artifacts such as cross-sectional design or strong correlations between mediator and outcome variables.

Practically, this means that investment in infrastructure, management commitment, professional development, and supportive policies are the most effective levers for improving multicultural education management through AI-based media. The significant mediating role through TMC suggests that teachers’ multicultural competence not only directly facilitates the implementation of AI-based media but also indirectly enhances the effectiveness of multicultural education management.

These findings emphasize that teachers’ professional development in multicultural pedagogy must be integrated with training in AI-based educational tools to maximize impact. Similarly, mediation through ATR shows that building teachers’ technological readiness contributes to educational effectiveness, albeit to a lesser extent than institutional support.

Overall, these findings address four research gaps identified in the introduction. First, this study provides an integrated model that simultaneously analyzes the interactions between TMC, ATR, and ISS, overcoming fragmentation in the existing literature. Second, by focusing on early childhood education in Indonesia, this study extends the AI in education literature to a context with unique pedagogical characteristics. Third, the mediation analysis reveals the mechanisms through which individual and organizational factors influence educational effectiveness, particularly through the implementation of AI-based multicultural learning media. Fourth, Indonesia's extraordinary cultural diversity provides a highly relevant empirical context for studying AI-based multicultural education, providing much-needed evidence for an under-explored domain.

6 Conclusion and implications

The results from this research have shown the following: First, impact from Institutional Support Systems appeared as the most robust predictor of the implementation of AI-based multicultural learning media, suggesting that the more integrated the organizational components of management buy-in, professional development, infrastructure, and policies frameworks, the more facilitative the organizational environment they create for the use of technology in multicultural education. Second, teachers’ multicultural competency and AI technology preparedness, though lower in the implementation of AMLMI, contribute positively to its implementation. This illustrates that while individual components are necessary, they are insufficient in the absence of adequate organizational support. Third, AI-based multicultural learning media implementation has a very strong impact on the effectiveness of multicultural education management. This indicates that the most operational factors for effective multicultural education management are the quality of implementation of AI-based media, including design, culturally adaptive content, personalization, and interactive elements. Fourth, the results of full mediation indicate that TMC, ATR, and ISS have an effect on MEME, which is fully mediated by AMLMI. This mediation pattern has important practical implications: the findings suggest that investments in teacher training, technology preparedness, and institutional infrastructure may be most effective when accompanied by the quality implementation of AI-based learning media, although longitudinal research is needed to confirm this causal pathway.

The results directly affect educational practices and policies. Given the scale of institutional backing, policymakers and school administrators ought to focus resource allocations on digital infrastructure, articulate policy stances on AI, implement systematic training on integrating multicultural pedagogy with AI, and build an organizational culture that promotes risk-taking in the use of technology in teaching. Teacher preparation programs must balance multicultural education and AI preparedness, ensuring that individuals are ready to implement the said competencies, particularly in the use of AI learning media in teaching. Teachers must be involved in the design of culturally responsive AI educational resources to ensure they incorporate local cultures and are suitable for educating young children.

Some limitations need to be stated. Due to the cross-sectional design, it is not possible to make causal inferences. Therefore, all findings should be interpreted as associative rather than causal relationships. The terms “predictor” and “effect” used in this study refer to statistical relationships within the PLS-SEM framework and do not imply a causal relationship Therefore, longitudinal studies are needed to determine the way the relationships among the Teachers’ Multicultural Competence (TMC), AI Technology Readiness (ATR), AI-based Multicultural Learning Media Implementation (AMLMI), and Multicultural Education Management Effectiveness (MEME) evolve over time. The sample is from only one province (Central Java), which may limit the generalizability of findings to other provinces in Indonesia or to other countries with different cultural and institutional arrangements. This study did not include the analysis of moderator variables such as teacher characteristics, type of school, or level of urbanization of the region, which may determine the strength of the relationships observed.

In addition, all data were collected via a single questionnaire completed at a specific point in time, which could potentially introduce general methodological bias. Although statistical tests (VIF for multicollinearity) indicate that general methodological bias is not a significant threat, future research should consider using various data collection methods, such as combining self-report measures with observational data or administrative records, to further minimize this risk.

References

• 1

  AnX.ChaiC. S.LiY.ZhouY.ShenX.ZhengC.et al (2023). Modeling English teachers’ behavioral intention to use artificial intelligence in middle schools. Educ. Inf. Technol.28, 5187–5208. 10.1007/s10639-022-11286-z

  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  AyanwaleM. A.SanusiI. T.AdelanaO. P.ArulebaK. D.OyelereS. S. (2022). Teachers’ readiness and intention to teach artificial intelligence in schools. Comput. Educ. Artif. Intell.3, 100099. 10.1016/j.caeai.2022.100099

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  BanksJ. A. (2019). An introduction to Multicultural Education (6th ed.). Boston, MA: Pearson.

  • Google Scholar
  • View reference in article

• 4

  BanksJ. A. (2008). An Introduction to Multicultural Education. Boston, MA: Allyn & Bacon/Pearson.

  • Google Scholar
  • View reference in article

• 5

  CelikI.DindarM.MuukkonenH.JärveläS. (2022). The promises and challenges of artificial intelligence for teachers: a systematic review of research. TechTrends66, 616–630. 10.1007/s11528-022-00715-y

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  ChenL.ChenP.LinZ. (2020). Artificial intelligence in education: a review. IEEE. Access.8, 75264–75278. 10.1109/ACCESS.2020.2988510

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  ChiuT. K. F.XiaQ.ZhouX.ChaiC. S.ChengM. (2023). Systematic literature review on opportunities, challenges, and future research recommendations of artificial intelligence in education. Comput. Educ. Artif. Intell.4, 100118. 10.1016/j.caeai.2022.100118

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  CohenJ. (1988). Statistical Power Analysis for the Behavioral Sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associates.

  • Google Scholar
  • View reference in article

• 9

  DavisF. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Q.13, 319–340. 10.2307/249008

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  EisenbergerR.HuntingtonR.HutchisonS.SowaD. (1986). Perceived organizational support. J. Appl. Psychol.71, 500–507. 10.1037/0021-9010.71.3.500

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  FornellC.LarckerD. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. J. Mark. Res.18, 39–50. 10.1177/002224378101800104

  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  GayG. (2018). Culturally Responsive Teaching: Theory, Research, and Practice (3rd ed.). New York, NY: Teachers College Press.

  • Google Scholar
  • View reference in article

• 13

  GunnA. A.SmithK.ZuikerS. J. (2020). Revisiting culturally responsive teaching practices for early childhood pre-service educators. J. Early Child. Teach. Educ.41, 91–123. 10.1080/10901027.2020.1735586

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  HairJ. F.HultG. T. M.RingleC. M.SarstedtM. (2022). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (3rd ed.). Thousand Oaks, CA: Sage.

  • Google Scholar
  • View reference in article

• 15

  HartinahH.YulianaY.ThamrinP. A. (2026). Managing diversity for quality: multicultural education management in a plural elementary school context. Manager: Indonesian Journal of Educational Management8 (1), 15–26. 10.52627/managere.v8i1.1514

  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  HenselerJ.RingleC. M.SarstedtM. (2015). A new criterion for assessing discriminant validity in variance-based structural equation modeling. J. Acad. Mark. Sci.43, 115–135. 10.1007/s11747-014-0403-8

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  HolmesW.BialikM.FadelC. (2019). Artificial Intelligence in Education: Promises and Implications for Teaching and Learning. Boston, MA: Center for Curriculum Redesign.

  • Google Scholar
  • View reference in article

• 18

  HwangG.-J.XieH.WahB. W.GaševićD. (2020). Vision, challenges, roles and research issues of artificial intelligence in education. Comput. Educ. Artif. Intell.1, 100001. 10.1016/j.caeai.2020.100001

  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  KanterR. M. (1993). Men and Women of the Corporation. 2nd ed.New York, NY: Basic Books.

  • Google Scholar
  • View reference in article

• 20

  KockN. (2015). Common method bias in PLS-SEM: a full collinearity assessment approach. Int. J. e-Collab.11, 1–10. 10.4018/ijec.2015100101

  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  KulttiA. (2022). Teaching responsive to play and linguistic diversity in early childhood education. Int. J. Biling. Educ. Biling.25, 3037–3045. 10.1080/13670050.2021.2001426

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  Ladson-BillingsG. (2014). Culturally relevant pedagogy 2.0: a.k.a. The remix. Harv. Educ. Rev.84, 74–84. 10.17763/haer.84.1.p2rj131485484751

  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  LuckinR.HolmesW.GriffithsM.ForcierL. B. (2016). Intelligence Unleashed: An Argument for AI in Education. London: Pearson.

  • Google Scholar
  • View reference in article

• 24

  NgD. T. K.LeungJ. K. L.SuJ.NgR. C. W.ChuS. K. W. (2023). Teachers’ AI digital competencies and twenty-first-century skills in the post-pandemic world. Educ. Technol. Res. Dev.71, 137–161. 10.1007/s11423-023-10203-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  NietoS. (2017). Re-imagining multicultural education: new visions, new possibilities. Multicult. Educ. Rev.9, 1–10. 10.1080/2005615X.2016.1276671

  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  OkkenC.JansenE.HofmanW.CoelenR. (2022). Interculturally competent teachers: behavioural dimensions and the role of teacher education. Cogent Educ.9, 2138048. 10.1080/2331186X.2022.2138048

  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  ParasuramanA. (2000). Technology readiness index (TRI): a multiple-item scale to measure readiness to embrace new technologies. J. Serv. Res.2, 307–320. 10.1177/109467050024001

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  ParasuramanA.ColbyC. L. (2015). An updated and streamlined technology readiness index: tRI 2.0. J. Serv. Res.18, 59–74. 10.1177/1094670514539730

  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  PodsakoffP. M.MacKenzieS. B.LeeJ.-Y.PodsakoffN. P. (2003). Common method biases in behavioral research: a critical review. J. Appl. Psychol.88, 879–903. 10.1037/0021-9010.88.5.879

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  PreacherK. J.HayesA. F. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behav. Res. Methods40, 879–891. 10.3758/BRM.40.3.879

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  RahiemM. D. H. (2021). Storytelling in early childhood education: time to go digital. Smart Learn. Environ.8, 4. 10.1186/s40723-021-00081-x

  • CrossRef
  • Google Scholar
  • View reference in article

• 32

  RamazanogluM.AkınT. (2025). AI Readiness scale for teachers: development and validation. Educ. Inf. Technol.30, 6869–6897. 10.1007/s10639-024-13087-y

  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  RingleC. M.WendeS.BeckerJ.-M. (2024). SmartPLS 4. Bönningstedt, Germany: SmartPLS GmbH.

  • Google Scholar
  • View reference in article

• 34

  RomijnB. R.SlotP. L.LesemanP. P. M. (2021). Increasing teachers’ intercultural competences in teacher preparation. Teach. Teach. Educ.98, 103236. 10.1016/j.tate.2020.103236

  • CrossRef
  • Google Scholar
  • View reference in article

• 35

  SakrM.PapadopoulouS. (2022). Children and teachers engaging together with digital technology in ECE: a literature review. Early. Years. (Stoke-on-Trent)42, 404–421. 10.1080/1350293X.2021.1971730

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  SchererR.HowardS. K.TondeurJ.SiddiqF. (2021). Profiling teachers’ readiness for online teaching and learning in higher education: who’s ready?Comput. Hum. Behav.118, 106675. 10.1016/j.chb.2020.106675

  • CrossRef
  • Google Scholar
  • View reference in article

• 37

  SperlingK.StenbergC.-J.McGrathC.ÅkerfeldtA.HeintzF.StenlidenL. (2024). In search of artificial intelligence (AI) literacy in teacher education: a scoping review. Comput. Educ. Open6, 100169. 10.1016/j.caeo.2024.100169

  • CrossRef
  • Google Scholar
  • View reference in article

• 38

  StatistikB. P. (2021). Kewarganegaraan, Suku Bangsa, Agama, dan Bahasa Sehari-hari penduduk Indonesia: Hasil Sensus Penduduk. Jakarta: Badan Pusat Statistik.

  • Google Scholar
  • View reference in article

• 39

  SuJ.NgD. T. K.ChuS. K. W. (2023). Artificial intelligence (AI) literacy in early childhood education: the challenges and opportunities. Comput. Educ. Artif. Intell.4, 100124. 10.1016/j.caeai.2023.100124

  • CrossRef
  • Google Scholar
  • View reference in article

• 40

  SuJ.YangW. (2022). Artificial intelligence in early childhood education: a scoping review. Comput. Educ. Artif. Intell.3, 100049. 10.1016/j.caeai.2022.100049

  • CrossRef
  • Google Scholar
  • View reference in article

• 41

  UNESCO (2023). Guidance for generative AI in education and research.UNESCO. 10.54675/ewzm9535

  • CrossRef
  • Google Scholar
  • View reference in article

• 42

  VygotskyL. S. (1978). Mind in Society: The Development of Higher Psychological Processes. Cambridge, MA: Harvard University Press.

  • Google Scholar
  • View reference in article

• 43

  WangX.LiL.TanS. C.YangL.LeiJ. (2023). Preparing for AI-enhanced education: conceptualizing and empirically examining teachers’ AI readiness. Comput. Hum. Behav.146, 107798. 10.1016/j.chb.2023.107798

  • CrossRef
  • Google Scholar
  • View reference in article

• 44

  WuJ.LiaoY.NianZ. (2022). The role of teacher technology experiences and school technology support on teachers’ self-efficacy in implementing culturally responsive teaching practices. J. Res. Technol. Educ.55, 522–542. 10.1080/07380569.2022.2071231

  • CrossRef
  • Google Scholar
  • View reference in article

• 45

  YiH.LiuT.LanG. (2024). The key artificial intelligence technologies in early childhood education: a review. Artif. Intell. Rev.57, 12. 10.1007/s10637-7

  • CrossRef
  • Google Scholar
  • View reference in article

• 46

  Zawacki-RichterO.MarínV. I.BondM.GouverneurF. (2019). Systematic review of research on artificial intelligence applications in higher education. Int. J. Educ. Technol. High. Educ.16, 39. 10.1186/s41239-019-0171-0

  • CrossRef
  • Google Scholar
  • View reference in article

• 47

  ZhaoX.LynchJ. G.ChenQ. (2010). Reconsidering baron and kenny: myths and truths about mediation analysis. J. Consum. Res.37, 197–206. 10.1086/651257

  • CrossRef
  • Google Scholar
  • View reference in article