Jurnal Elemen, 11.
, 912-929.
October 2025 https:/doi.
org/10.
29408/jel.
Modeling the determinants of AI integration in primary mathematics education: A structural equation modeling Dwi Yulianto 1 *.
Egi Adha Juniawan 1.
Yusup Junaedi 1.
Astari 2.
Rahmat Nurcahyo 3 Department of Mathematics Education.
Latansa Mashiro University.
Indonesia Faculty of Computer Science.
Latansa Mashiro Islamic Religious High School.
Indonesia Faculty of Information Technology.
Latansa Mashiro University.
Indonesia Correspondence: dwiyulianto554@gmail.
A The Author.
2025 Abstract This study addresses a critical gap in educational technology research by simultaneously examining the internal and external determinants of Artificial Intelligence (AI) integration in primary mathematics instruction.
Using a second-order Structural Equation Modeling (SEM) framework, the study investigates how teachersAo attitudes and TPACK competencies .
nternal factor.
, alongside policy support, infrastructure, and community engagement .
xternal factor.
, influence AI utilization among 516 primary school mathematics teachers in Jakarta.
Indonesia.
The results reveal that internal factors have a strong direct effect on AI utilization ( = 0.
p < 0.
, while external factors exert a significant indirect influence via internal mediators ( = 0.
p < 0.
, despite an insignificant direct effect ( = 0.
p = 0.
The model explains 78.
1% of the variance in AI utilization (RA = 0.
and shows high predictive relevance (QA > 0.
These findings underscore the pivotal role of teacher readiness in AI integration, with systemic support enhancing its effectiveness through internal capacity-building.
The study contributes an empirically validated instrument and a comprehensive ecological model, offering actionable insights for policymakers and educators in developing nations pursuing ethical, equitable, and sustainable AI integration in primary Keywords: AI integration.
educational environment.
elementary mathematics.
TPACK
How to cite: Yulianto.
Juniawan.
Junaedi.
Astari, & Nurcahyo.
Modeling the determinants of AI integration in primary mathematics education: A structural Jurnal Elemen, 11.
, https://doi.
org/10.
29408/jel.
Received: 28 May 2025 | Revised: 18 June 2025 Accepted: 2 July 2025 | Published: 7 November 2025 Jurnal Elemen is licensed under a Creative Commons Attribution-ShareAlike 4.
0 International License.
Dwi Yulianto.
Egi Adha Juniawan.
Yusup Junaedi.
Astari.
Rahmat Nurcahyo Introduction As global education undergoes digital transformation.
Artificial Intelligence (AI) is increasingly recognized for reshaping teaching strategies and how students learn (SanabriaNavarro et al.
, 2023.
Walter, 2.
In Indonesia, this aligns with the 2022Ae2026 Digital Transformation Strategic Plan, which highlights AI's potential in enhancing primary mathematics education through adaptive tools, real-time insights, and personalized learning support (Olmo-Muyoz et al.
, 2023.
Pineda-Martynez et al.
, 2.
Tools like intelligent tutoring systems and AI-driven assessments have been linked to improved student engagement and conceptual mastery (Gadanidis, 2017.
Hwang & Tu, 2.
, while datainformed instruction has been shown to boost both motivation and learning outcomes (Annu & Kmeu, 2024.
Wei et al.
, 2.
Despite this, research remains concentrated on secondary education in developed nations, with limited focus on AI integration at the primary level in developing countries.
This study aims to bridge that gap by exploring how AI can support mathematics learning through the interaction of teacher preparedness and systemic support in IndonesiaAos educational landscape.
In this study, internal competences refer to teachersAo intrinsic capacities encompassing two core constructs: attitudes toward AI and TPACK proficiency (Technological Pedagogical Content Knowledg.
The attitudinal component represents teachersAo beliefs about AIAos utility, ease of use, and ethical implications (El Hajj & Harb, 2.
, rooted in the Technology Acceptance Model (TAM).
Positive attitudes have been associated with proactive engagement in AI-enabled platforms, such as intelligent tutoring systems and diagnostic learning analytics (S.
Ng et al.
, 2.
Meanwhile.
TPACK competence, derived from the model by Jia et al.
, reflects teachersAo ability to integrate content knowledge, pedagogy, and digital technologies effectively.
Teachers with high TPACK fluency are more capable of designing meaningful, data-driven instruction using AI-enhanced tools (Rahimi & Kim, 2021.
Ye et al.
In this study, internal competence functions not only as a cognitive-affective driver but also as a pedagogical filter that determines the quality and ethics of AI integration.
External competences, in contrast, represent the institutional and ecological capacities that scaffold and sustain AI integration.
These include policy-level support, availability of digital infrastructure, and parental or community involvement, as operationalized through the E-TPACK model and BronfenbrennerAos ecological systems theory (Tong & An, 2.
External factors shape the broader ecosystem in which teachers operate.
for instance, effective policy mandates, equitable access to digital devices, and culturally informed community engagement can significantly amplify teacher readiness (Azhar et al.
, 2022.
Flores-Vivar & Garcya-Peyalvo, 2.
Conversely, institutional apathy, the digital divide, or AI-related misconceptions within communities can constrain innovation uptake despite high individual capacity (Scherer & Siddiq, 2.
This study postulates that internal competences mediate the relationship between external supports and actual AI utilization, highlighting a reciprocal dependency wherein systemic support enhances teacher capacity, which in turn enables meaningful AI adoption.
Modeling the determinants of AI integration in primary mathematics education: A structural .
Effective integration of Artificial Intelligence (AI) in primary mathematics education demands a conceptual framework that bridges pedagogical, psychological, and systemic This study proposes a synthesized model combining TPACK.
TAM, and ETPACK.
TPACK (Mishra et al.
, 2.
underscores the balance of content, pedagogy, and technology as essential teacher competencies, while TAM (Davis & GraniN, 2.
highlights perceived usefulness and ease of use as key drivers of technology adoption (Li et al.
, 2024.
Ye et al.
, 2.
E-TPACK extends this by incorporating BronfenbrennerAos ecological model, emphasizing the role of contextual factors, individual, institutional, and structural, in shaping teacher readiness.
Together, these models form a comprehensive framework in which teacher attitudes.
TPACK mastery, and systemic support are critical to AI implementation.
Teachers with positive perceptions of AI, supported professionally and institutionally, are more likely to apply it effectively in adaptive learning settings (Khong et al.
, 2023.
Ma et al.
, 2.
Thus.
AI integration success hinges not only on individual readiness but also on coherent systemic support within the educational ecosystem.
The rapid integration of Artificial Intelligence (AI) into education underscores an urgent need to examine not only technological preparedness but also the alignment between stakeholder capacities and policy frameworks, particularly in developing countries.
While scholarly interest in pedagogical competence and systemic support for AI-based education has grown (El Hajj & Harb, 2023.
Jia et al.
, 2.
, there remains a notable scarcity of studies that concurrently investigate both internal .
eacher-leve.
and external .
nstitutional-leve.
factors, especially within the context of primary education in low- and middle-income countries such as Indonesia.
The novelty of this study lies in its ecological-contextual approach, which integrates three theoretical models.
TPACK.
TAM, and E-TPACK, into a comprehensive framework for examining how teacher readiness and systemic support interactively influence AI adoption in primary mathematics education.
This integrative model has rarely been applied with empirical rigor, particularly through second-order reflectiveAeformative modeling using PLS-SEM, as implemented in this research.
Existing studies have predominantly concentrated on secondary or higher education in technologically advanced contexts (Li et al.
, 2024.
Tang et al.
, 2.
, leaving a critical gap in understanding AI readiness among primary school teachers in emerging urban environments.
Furthermore, this study introduces an empirically validated diagnostic instrument, the Scale of Mathematics TeachersAo Technology Integration (SMTTI), which measures not only teachersAo TPACK and attitudes toward AI but also incorporates systemic support components such as educational policy, infrastructure, and community engagement, consistent with BronfenbrennerAos ecological systems theory.
This represents a methodological advancement over prior frameworks, which often analyse these variables in isolation.
Another key contribution is the studyAos focus on Jakarta as a representative urban ecosystem within a developing country.
This context offers transferable insights for comparable socio-educational settings across Southeast Asia and other similar regions.
In contrast to earlier research that tends to underemphasize the role of parents and communities (Khosravi et al.
, 2023.
Scherer & Siddiq, 2.
, this study empirically demonstrates the indirect but pivotal influence of these external factors, mediated through internal teacher readiness.
Overall, this research Dwi Yulianto.
Egi Adha Juniawan.
Yusup Junaedi.
Astari.
Rahmat Nurcahyo addresses both methodological and contextual gaps in the current AI-in-education literature.
offers a practical, evidence-based framework for policymakers and educators, providing actionable insights for designing inclusive and context-sensitive AI integration strategies in primary mathematics instruction.
This dual theoretical and applied contribution distinguishes the present study from previous work in the field.
Methods Research design This study adopts a second-order SEM within a reflective formative framework to examine how internal and external factors shape AI integration in primary mathematics education.
The model captures the complexity of constructs like TPACK and systemic support, while enabling analysis of indirect mediation effects (Hair & Alamer, 2.
Grounded in global literature and tailored to IndonesiaAos context, where many teachers lack technological pedagogical skills and digital infrastructure remains limited, the study combines TPACK.
TAM, and E-TPACK to explain the interplay between teacher readiness and environmental support in fostering effective AI adoption.
Participants As IndonesiaAos capital and a prominent urban education hub in Southeast Asia.
Jakarta offers a strategic context for investigating readiness and structural barriers to AI integration in primary mathematics instruction.
This study surveyed 516 primary mathematics teachers, proportionally sampled across the cityAos five administrative districts, representing public, faith-based private, and inclusive schools.
The sampleAos diversity ensured representation across socioeconomic and institutional contexts.
Predominantly female .
6%), in line with national and international trends (Reuter et al.
, 2.
, participants mostly taught grades 3 and 5, with 42.
6% identifying as senior teachers, educators with strong pedagogical backgrounds but potentially lower openness to technological innovation.
This demographic and institutional variation reflects JakartaAos complex educational landscape and strengthens the studyAos analytical relevance and applicability to other multicultural urban settings in the Global South.
Table 1.
Demographic characteristics of research participants Characteristic Gender Teaching Experience Grade Level Taught Category Female Male 0Ae5 years 6Ae10 years 11Ae15 years >15 years Grade 1 Grade 2 Grade 3 Grade 4 Grade 5 Frequency .
Percentage (%) Modeling the determinants of AI integration in primary mathematics education: A structural .
Characteristic School Type Administrative Region Category Grade 6 Public Private Central Jakarta North Jakarta West Jakarta South Jakarta East Jakarta Frequency .
Percentage (%) Instrument This study utilized the Scale of Mathematics TeachersAo Technology Integration (SMTTI), adapted from Li et al.
, to assess primary mathematics teachersAo readiness to incorporate Artificial Intelligence (AI) into instruction.
Cross-cultural validation involved forwardAebackward translation, expert evaluation, and a pilot with 35 teachers to ensure linguistic precision and contextual alignment.
The instrument is theoretically rooted in TPACK .
echno pedagogical knowledg.
TAM .
ttitudinal and behavioural intentio.
, and ecological theory .
ystemic contextual influence.
SMTTI measures two key domains:
internal factors (TPACK.
TCK/TPK, and AI attitude.
and external factors .
olicy, infrastructure, parental engagement, sociocultural norm.
Data were modelled using a second-order hierarchical structure combining reflective and formative indicators, following Hair and Alamer .
Comprising 31 items on a five-point Likert scale, the instrument captures beliefs about AI-enhanced learning tools and contextual supports such as digital literacy and infrastructure availability.
Psychometric validation yielded satisfactory results, with factor loadings > 0.
AVE > 0.
CR > 0.
80, and CronbachAos alpha between 0.
82Ae Discriminant validity was supported via FornellAeLarcker and HTMT (< 0.
The online survey format facilitated broad participation while ensuring data integrity through authentication and system safeguards.
Overall.
SMTTI is a theoretically sound and empirically validated tool for diagnosing teacher preparedness and guiding AI integration strategies in mathematics education.
Model fit and scale validity Model fit was assessed using second-order reflectiveAereflective SEM to examine attitudes toward AI.
TPACK competence, and external contextual factors in elementary mathematics The model demonstrated excellent fit (NA/df = 2.
RMSEA = 0.
SRMR =
CFI = 0.
TLI = 0.
, indicating strong structural validity.
All constructs showed high internal consistency (CR > 0.
CronbachAos > 0.
, with convergent validity supported by AVE values > 0.
Discriminant validity was confirmed via the FornellAe Larcker criterion.
These results affirm the modelAos empirical robustness and theoretical The web-based SMTTI instrument not only facilitated efficient data distribution but also enabled valid multi-construct measurement in a complex urban context such as Jakarta.
Nevertheless, it is important to expand the population scope and account for socioeconomic heterogeneity to further enhance external validity.
Taken together, the findings Dwi Yulianto.
Egi Adha Juniawan.
Yusup Junaedi.
Astari.
Rahmat Nurcahyo affirm that SMTTI is a psychometrically sound and contextually valid tool for assessing teachersAo readiness to adopt AI in a systemic and grounded manner.
Table 2.
Summary of model fit indices for the SMTTI
Fit
Index
NA/df RMSEA
SRMR
GFI
AGFI
NFI
CFI
TLI
Obtained Value Good Fit Criteria O 0.
O 0.
Ou 0.
Ou 0.
Ou 0.
Ou 0.
Ou 0.
Acceptable Fit Criteria 3Ae5 05 Ae 0.
05 Ae 0.
90 Ae 0.
85 Ae 0.
90 Ae 0.
90 Ae 0.
90 Ae 0.
Model Fit Assessment Good fit Close fit Good fit Acceptable fit Acceptable fit Acceptable fit Excellent fit Excellent fit Table 3 confirms the strong psychometric properties of the model, with CronbachAos alpha and Composite Reliability values exceeding 0.
85, indicating high reliability.
Elevated AVE scores, notably for TPCK .
and TCK_TPK .
, reflect substantial explanatory power of the latent constructs.
Discriminant validity, assessed via the FornellAeLarcker criterion, is evident as the square roots of AVE surpass inter-construct correlations .
OoAVE of TPCK = 0.
980 > correlation with TCK_TPK = 0.
, affirming construct These results validate the modelAos alignment with TPACK and E-TPACK frameworks and support its methodological suitability for application in urban primary education settings, including those with similar contextual features.
Table 3.
Reliability and validity Construct
ATT
AIU
CTX
TCKP
TPCK
EDC
PCI
AVE
Cronbach
ATT
AIU
CTX
TCKP
TPCK
EDC
PCI
This study adopts a standardized set of abbreviations to represent the key constructs in the research model.
ATT (Attitud.
captures educatorsAo perceptions of AI integration in teaching, while AIU (AI Utilizatio.
reflects the extent of AI implementation in instructional CTX (Contextual Factor.
encompasses external influences such as institutional support and school environments.
TCKP merges Technological Content Knowledge (TCK) and Technological Pedagogical Knowledge (TPK), indicating teachersAo techno-pedagogical TPCK denotes Technological Pedagogical Content Knowledge, emphasizing the integrated application of content, pedagogy, and technology.
EDC (Educational Challenge.
identifies structural and systemic barriers, and PCI (Parental and Community Involvemen.
measures family and community engagement in supporting AI use in education.
These constructs are consistently referenced throughout the empirical analysis, structural model interpretation, and theoretical discussion.
Modeling the determinants of AI integration in primary mathematics education: A structural .
Data collection Data were collected over three months through an online survey administered to elementary school mathematics teachers across the five administrative regions of Jakarta.
This region was selected as the study site because it serves as a national pilot area for the AuDigitalisasi SekolahAy (School Digitalizatio.
program and demonstrates high levels of ICT device ownership and infrastructure availability.
Utilizing a census-based purposive sampling approach, the survey targeted the entire population of active mathematics teachers, distributed officially through the Provincial Education Office.
A total of 516 valid responses were obtained, meeting the minimum sample requirement for estimating a second-order SEM model (Hair & Alamer, 2.
The instrument.
SMTTI, encompassed constructs such as TPACK, perceived usefulness, perceived ease of use, and attitudes toward AI.
It was culturally adapted through a forward-backward translation process and validated by a panel of experts (CVI > 0.
CronbachAos > 0.
The survey implementation incorporated tokenbased access.
IP restrictions, and screening questions to ensure data integrity and prevent Participation was voluntary, anonymous, and conducted following established research ethics protocols.
Preliminary data checks revealed no outliers, missing values, or duplicate responses, confirming the dataset's adequacy for analysis using PLS-SEM with SmartPLS 4.
Data analysis Data analysis was conducted using second-order Structural Equation Modelling (SEM) with a reflectiveAereflective structure and the repeated indicators method, following Hair and Alamer .
Due to model complexity and non-normal data distribution.
SmartPLS 4.
0 was employed as the primary tool, replacing AMOS, in line with Hair and Alamer .
Diagnostic tests confirmed violations of normality and homoscedasticity, supporting the appropriateness of the PLS-SEM approach.
The analysis comprised two stages: measurement and structural model assessment.
All indicators showed strong factor loadings (> 0.
internal consistency (CronbachAos and CR > 0.
, and convergent validity (AVE > 0.
with no multicollinearity issues (VIF < .
Discriminant validity was confirmed via the FornellAeLarcker criterion.
The structural model demonstrated high explanatory power (RA = 0.
and predictive relevance (QA = 0.
Internal factors fully mediated the relationship between external factors and AI utilization ( = 0.
217, p < 0.
VAF = 96.
4%), while the direct effect was non-significant.
Effect size analysis indicated moderate .
A = 0.
to strong .
A = 0.
These results highlight the pivotal role of internal readiness, supported by systemic external conditions, in successful AI integration.
They also reinforce the theoretical and practical validity of the E-TPACK framework in advancing digital mathematics education in primary schools.
Dwi Yulianto.
Egi Adha Juniawan.
Yusup Junaedi.
Astari.
Rahmat Nurcahyo Results This studyAos conceptual framework investigates the direct and mediated effects of external factors on AI integration in elementary mathematics education, with internal factors serving as a mediator.
Using second-order SEM via SmartPLS 4.
0, the model captures the interplay between teacher readiness and systemic support.
Internal Factors (IF) comprise Attitude.
TPACK, and the combined dimensions of TCK and TPK, while External Factors (EF) include Contextual Factors (CTX).
Educational Challenges (EDC), and Parental and Community Involvement (PCI), reflecting the broader ecological and institutional influences on AI adoption in classrooms.
Fig 1.
Second-order of SEM The measurement model was evaluated to ensure construct validity and reliability for Internal Factors (IF).
External Factors (EF), and AI Utilization, including analyses of factor loadings, internal consistency, convergent validity, and multicollinearity (Table .
Table 4.
Measurement model assessment Construct
Internal Factors AI Utilization External Factors Indicators TPACK
ATTITUDE
TCK_TPK
Q14
Q15
Q16
Q17
Q18
PCI
Indicators VIF CronbachAos CR AVE The measurement model shows strong internal consistency for AI Utilization and Internal Factors, with indicator loadings above 0.
One External Factor indicator (CF) Modeling the determinants of AI integration in primary mathematics education: A structural .
yielded a lower, yet acceptable loading .
, suggesting the need for semantic refinement.
VIF values under 4 confirm no multicollinearity.
Reliability is supported by CronbachAos alpha and CR > 0.
88, and AVE > 0.
70 confirms convergent validity.
Discriminant validity, assessed via the FornellAeLarcker criterion, is met, as AVE square roots exceed inter-construct correlations .
ee Table .
Tabel 5.
Discriminant validity Konstruk AI Utilization External Factors Internal Factors AI Utilization External Factors Internal Factors The measurement model met the required standards for reliability and validity, establishing a sound basis for structural analysis (Table .
Path analysis revealed a strong direct effect of internal factors, comprising teachersAo attitudes and TPACK competence, on AI utilization ( = 0.
T = 16.
p < 0.
fA = 0.
Although external factors did not exert a significant direct influence ( = 0.
p = 0.
, their indirect effect through internal mediation was statistically significant ( = 0.
T = 3.
p < 0.
, confirming a full mediating role.
The robust link between external and internal factors ( = 0.
T = 9.
highlights the critical role of systemic support in enhancing teacher capacity.
Table 6.
Summary of structural path effects and effect sizes Path External Ie AI Utilization External Ie Internal Factors (IF) Internal Factors Ie AI Utilization Direct Effect Indirect Effect Total Effect Tvalue p-value 000*** 0.
Ae Ae Note Full mediation via IF Significant direct 000*** 9.
Strongest direct 000*** 0.
Path analysis revealed that internal factors, specifically teachersAo attitudes and TPACK competence, exert a strong and statistically significant direct effect on AI utilization in primary mathematics classrooms ( = 0.
T = 16.
p < 0.
fA = 0.
In contrast, external factors showed no significant direct influence ( = 0.
p = 0.
However, their indirect effect through internal factors was significant ( = 0.
T = 3.
p < 0.
confirming a full mediation effect.
The notable link between external and internal factors ( = T = 9.
underscores the role of systemic support in enhancing teacher capacity.
These findings suggest that successful AI integration should prioritize strengthening teacher readiness, supported by an enabling educational environment.
Additional RA and QA analyses further validated the modelAos explanatory and predictive strength .
ee Table .
Table 7.
RA and QA values for main constructs Construct AI Utilization Internal Factors RA RA Interpretation QA Strong Weak QA Interpretation Highly Relevant (Larg.
Moderately Relevant (Mediu.
Dwi Yulianto.
Egi Adha Juniawan.
Yusup Junaedi.
Astari.
Rahmat Nurcahyo Table 7 shows that AI utilization has strong explanatory power (RA = 0.
and high predictive relevance (QA = 0.
, indicating that internal and external factors jointly play a significant role in teachers' AI adoption.
In contrast, internal readiness alone has a lower explanatory value (RA = 0.
, though its predictive relevance remains moderate (QA = This suggests that external factors alone cannot fully account for internal readiness, highlighting the need for additional mediating variables.
The findings affirm the central mediating role of internal factors in linking systemic support to AI implementation: external support is insufficient without teacher readiness, while individual capacity is most impactful within a supportive ecosystem.
Table 8 confirms the model's assumption of partial mediation.
Table 8.
Summary of hypothesis testing results Hypothesis Path Findings Interpretation External Factors Ie Supported ( = 0.
275, p < Educational Internal Factors infrastructure, and community support significantly enhance teachersAo readiness.
Internal Factors Ie Supported ( = 0.
791, p < Pedagogical readiness.
TPACK AI Utilization mastery, and positive attitudes predict effective AI adoption.
External Factors Ie Indirect effect supported via External factors influence AI use Utilization Internal Factors ( = 0.
217, p only when mediated by internal .
< 0.
direct effect not teacher capacity.
This model reflects a fully mediated structure, where the impact of external factors on AI adoption occurs entirely through teachersAo internal readiness.
Aligned with the E-TPACK and Technology Acceptance Model (TAM), this finding underscores that technology adoption is shaped not only by perceived utility but by the interaction between individual preparedness and systemic support.
Model robustness was confirmed through sensitivity analysis (A10% variatio.
and alternative model testing, with negligible effects on path coefficients.
RA, and QA, indicating strong stability.
The results reaffirm TAMAos emphasis on attitudes and perceptions and highlight the importance of integrating technological, pedagogical, and content knowledge as outlined in TPACK.
Within the E-TPACK framework, the synergy between internal readiness and external support is pivotal for effective AI integration.
Practically, the findings call for education policies that go beyond infrastructure provision, focusing instead on strengthening pedagogical competencies, enhancing digital literacy, and supporting community-based implementation.
Prioritizing targeted AI training and curriculum adaptation can foster a more holistic, adaptive, and sustainable digital transformation in primary education.
Discussion Table 8 summarizes the hypothesis testing results derived from the second-order structural model analyzed using SmartPLS 4.
The model integrates the TPACK framework (Mishra et , 2.
, the Technology Acceptance Model (Davis & GraniN, 2.
, and the ecological ETPACK approach.
Internal factors are modeled as a second-order reflective construct Modeling the determinants of AI integration in primary mathematics education: A structural .
comprising attitudes toward AI and TPACK competence, while external factors include educational policy, digital infrastructure, and community support.
Construct validity is supported by satisfactory AVE, rho_A, and HTMT values.
The model demonstrates good fit, with an SRMR of 0.
036, well below the 0.
08 threshold.
HarmanAos single-factor test also indicates no significant common method bias, as no single factor explained more than 50% of the variance, confirming the modelAos robustness.
The systemic impact of external factors on teachersAo internal readiness for AI integration in primary mathematics education The findings of the analysis demonstrate that external factors exert a statistically significant and positive influence on internal factors ( = 0.
T = 9.
p < 0.
fA = 0.
RA =
, suggesting a moderate effect on teachersAo internal preparedness, characterized by favorable attitudes toward artificial intelligence and proficiency in TPACK.
This internal readiness is not developed in isolation.
rather, it is shaped by the presence of enabling policies, reliable digital infrastructure, and the active engagement of parents and the wider These results substantiate BronfenbrennerAos ecological systems theory (Davis & GraniN, 2.
, which asserts that human development is deeply embedded within and influenced by multilayered socio-ecological systems.
Within the context of AI integration in education, external elements function both as structural enablers and as catalysts for psychological preparedness.
This aligns with insights from Flores-Vivar and Garcya-Peyalvo .
, who underscore the importance of institutional scaffolding in alleviating teacher resistance to AI, and Yue et al.
, who emphasize that effective leadership and community collaboration are vital to strengthening teachersAo readiness.
Unlike theoretical models that narrowly emphasize individual competencies (Mishra et al.
, 2023.
Sun & Chen, 2.
, the present study advances a more holistic perspective, positing that successful AI adoption in primary education is fundamentally shaped by the synergistic interaction between educational policy, technological capacity, and sociocultural conditions.
The influence of internal factors on AI utilization in primary mathematics The structural model assessment revealed that internal determinants specifically teachersAo attitudes toward artificial intelligence and their TPACK competencies exerted a robust and statistically significant influence on the integration of AI within primary mathematics instruction ( = 0.
T = 16.
523, p < 0.
95% CI .
692, 0.
fA = 0.
Internal readiness was found to explain 68% of the variance in AI utilization (RA = 0.
, signifying a substantial level of predictive accuracy following the criteria outlined by Hair and Alamer .
These findings empirically validate the core propositions of the Technology Acceptance Model (Davis & GraniN, 2.
while simultaneously highlighting the pivotal function of TPACK in guiding the pedagogical and technological integration of AI tools (Celik, 2023.
Mishra et al.
, 2.
Thus, effective and sustainable implementation of AI in educational settings is contingent not merely upon teachersAo positive dispositions but also Dwi Yulianto.
Egi Adha Juniawan.
Yusup Junaedi.
Astari.
Rahmat Nurcahyo upon their capacity to ethically and pedagogically embed AI into instruction that fosters meaningful student learning.
These results align with studies by Yue et al.
and Khong et al.
, which identified teachersAo attitudes and TPACK competencies as key predictors of post-pandemic However, unlike previous research that focused broadly on educational technologies, this study specifically highlights the unique complexities of AI, including automation, personalization, and ethical implications (Chen, 2020.
Flores-Vivar & GarcyaPeyalvo, 2.
As such, strengthening AI literacy and developing teachersAo TPACK should be prioritized in digital education transformation.
Teacher training must evolve beyond technical instruction to encompass holistic and reflective approaches, including AI-based simulations, adaptive learning scenarios, and digital ethics development.
This aligns with the Intelligent-TPACK framework (Celik, 2.
, which emphasizes the integration of digital competence with ethical considerations in classroom-based AI applications.
Although internal factors were found to be dominant, the findings also highlight the essential role of external Without enabling policies, adequate infrastructure, and cross-sector collaboration, teachersAo potential to harness AI effectively may be limited (Bronfenbrenner, 1986.
Zhao.
Therefore, sustainable AI integration in education requires a strong synergy between teachersAo internal capacities and a robust system of external support.
Direct and indirect effects of external factors on AI utilization The analysis revealed that the direct effect of external factors on teachersAo utilization of AI was statistically insignificant ( = 0.
T = 0.
115, p = 0.
However, there was a significant indirect effect mediated by internal factors ( = 0.
T = 3.
872, p < 0.
CI .
112, 0.
), emphasizing the crucial role of teachersAo internal readiness in bridging the influence of external environments.
These findings reinforce (Bronfenbrenner, 1.
Aos ecological theory, which posits that environmental influences on individuals are mediated by internal characteristics.
In this context.
TPACK competence.
AI literacy, and self-efficacy emerge as the primary mediators (Celik, 2023.
Mishra et al.
, 2.
This implies that external interventions, such as generic training, provision of devices, or policy implementation, will likely be ineffective without a parallel reinforcement of teachersAo internal capacities.
Antonenko and Abramowitz .
further observed that misconceptions about AI can foster resistance, even when infrastructural and policy support is present.
From a practical standpoint, capacity-building strategies must include AI-based TPACK development (Gagne et al.
, 2021.
Mishra et al.
, 2.
, critical digital literacy (K.
Ng et al.
, 2021.
Walter, 2.
, and enhancing self-efficacy and motivation within digital learning environments (Yue et al.
, 2.
Context-sensitive and personalized approaches are considered more effective than uniform institutional interventions (Ahmad et al.
, 2021.
Pineda-Martynez et al.
, 2.
Given the significant mediation path, teachers should not be viewed merely as policy recipients but as reflective pedagogical agents who ethically and contextually adapt AI in the classroom (El Hajj & Harb, 2023.
Gadanidis, 2.
Consequently.
AI integration policies must be designed holistically, addressing the cognitive, affective, and conative dimensions of Modeling the determinants of AI integration in primary mathematics education: A structural .
teacher development (Chen, 2020.
Hwang & Tu, 2.
The divergence from prior studies that emphasized external interventions (Guo & Wan, 2022.
Khong et al.
, 2.
underscores the need for a paradigm shift.
Interventions should focus on empowering teachers' adaptive capacities to navigate rapidly evolving digital ecosystems (Annu & Kmeu, 2024.
Sutrisman et al.
, 2.
, including understanding the ethical and social implications of AI use (Bibri & Allam, 2022.
Flores-Vivar & Garcya-Peyalvo, 2.
In this regard, internal factors are not merely supplementary but rather serve as the leverage point of AI adoption in education.
Meaningful digital transformation in the classroom can only be achieved by prioritizing internal teacher capacity-building over external provisioning or regulation.
Implications of the study The findings highlight that integrating AI into elementary mathematics education requires a strategic synergy between external support and teachersAo internal capacities.
In alignment with the E-TPACK framework, external factors, such as supportive policies, equitable digital infrastructure, and community involvement, serve as critical prerequisites.
However, the successful implementation of AI ultimately hinges on strengthening teachersAo internal competencies, particularly AI-enhanced TPACK and ethical technology literacy.
Professional development programs should be modularly designed to include: .
adaptation of the mathematics curriculum using locally relevant AI tools such as GeoGebra AI or Scribe AI.
ethical-pedagogical exploration through AI-assisted flipped microteaching.
collaborative reflection via digital lesson study.
Furthermore, the development of peer coaching systems within schools plays a vital role in fostering pedagogical autonomy and teacher confidence.
Nevertheless.
AI integration also poses critical challenges.
Without reflective practice.
AI risks reducing teachers to mere technology operators.
Dependence on proprietary algorithms, the dominance of foreign vendors, and potential systemic biases threaten to erode local values embedded within national curricula (Bibri & Allam, 2.
Therefore, technology adoption should prioritize open-source accessibility, interoperability, and cultural and linguistic relevance.
On a global scale, these findings are particularly relevant to developing countries in ASEAN and Africa, which face parallel challenges such as infrastructure inequality, limited access to meaningful training, and the pressures of technological globalization.
A promising recommendation lies in the design of micro-AIintegrated MOOCs, which offer just-in-time, teacher-centered learning experiences, paving the way for adaptive, ethical, and equitable AI integration in education.
Limitations and future research This study has limited generalizability as it focuses solely on elementary mathematics teachers in urban Jakarta, an area that typically benefits from better access to digital infrastructure, technology training, and professional learning communities.
As such, the findings do not capture the substantial disparities faced by rural schools, including limited connectivity, low AI literacy, and institutional differences among public schools.
Islamic Dwi Yulianto.
Egi Adha Juniawan.
Yusup Junaedi.
Astari.
Rahmat Nurcahyo madrasahs, and inclusive private institutions.
Moreover, the cross-sectional design presents epistemological limitations, as it provides only a snapshot in time and fails to track the progression of teacher competencies post-training or their responses to ongoing curricular and technological shifts in the post-pandemic landscape.
Another limitation lies in the studyAos monodisciplinary focus on mathematics education.
The integration of AI in other disciplines, such as literacy, science, and inclusive education, poses unique pedagogical and ethical challenges, particularly in terms of content adaptability, the validity of AI-driven assessments, and studentsAo affective responses.
Future research should adopt experimental and longitudinal approaches, including the development of an AI literacy framework tailored for elementary educators, structured trials of TPACK AI-based training programs, and critical analyses of algorithmic bias and its impact on marginalized student populations.
Normative discussions on the ethical use of AI in primary education are also essential, particularly concerning child data privacy, student digital agency, and the urgent need for protective state regulation.
Additionally, the functional integration of MOOCs and AI warrants investigation, especially regarding their potential to enhance personalized learning, scaffolding, and adaptive feedback for teachers with limited digital literacy.
summary, future research agendas must not only broaden empirical scope but also drive systemic transformation toward an AI-integrated educational ecosystem that is inclusive, ethical, and contextually grounded.
AI-based educational research must move beyond passive adaptation to technology and instead serve as a vehicle for educational justice in the digital Conclusion This study concludes that the successful adoption of Artificial Intelligence (AI) in elementary mathematics education depends on the synergistic interplay between teachersAo internal readiness, particularly their attitudes and TPACK proficiency, and external systemic support.
The E-TPACK model, combining TPACK.
TAM, and ecological systems theory, demonstrated strong empirical validity and predictive relevance, explaining 78.
1% of the variance in AI utilization.
Internal factors were the most decisive, while external factors played a supporting role by enhancing internal capacities.
Theoretically, this research refines and contextualizes the TPACK framework through an ecological lens, offering a more holistic approach to educational technology integration.
Practically, it emphasizes the need for targeted teacher training and inclusive digital policy frameworks.
However, the study is limited by its urban focus and the absence of cross-institutional analysis.
Future research should explore rural and underserved contexts while addressing digital equity, algorithmic bias, and data privacy to ensure inclusive and ethical AI integration in primary education.
Modeling the determinants of AI integration in primary mathematics education: A structural .
Acknowledgments The authors gratefully acknowledge the Provincial Education Office of Jakarta for its essential support in data access and survey distribution, which ensured the research's smooth Sincere thanks are also extended to the elementary mathematics teachers whose time and insights enriched the studyAos findings.
Appreciation is further given to academic colleagues and research collaborators for their valuable feedback and contributions that strengthened the studyAos quality and rigor.
Conflicts of Interest The authors declare no conflict of interest regarding the publication of this manuscript.
All ethical considerations related to the research and publication process, such as plagiarism, research misconduct, data fabrication or falsification, duplicate publication or submission, and redundancy, have been thoroughly addressed and complied with.
Funding Statement This work received no specific grant from any public, commercial, or not-for-profit funding Author Contributions Dwi Yulianto: Conceptualization, methodology, supervision, writing Ae original draft, visualization, project administration.
Rahmat Nurcahyo: Formal analysis, software, data curation, writing Ae review & editing, validation.
Yusup Junaedi: Resources, investigation, writing Ae review & editing.
Astari: Instrument development, software testing, and technical Egi Adha Juniawan: Funding acquisition, ethical validation, and administrative References