The Economics of AI-Enabled Education: Human Capital Formation, Productivity, and Inclusive Economic Growth

1.1. Background and Motivation

Artificial intelligence has moved from a predominantly technological domain to a central component of contemporary economic and social transformation. The rapid diffusion of generative AI, machine-learning applications, intelligent tutoring systems, automated assessment, adaptive learning platforms, and AI-supported decision-making is fundamentally changing how knowledge is produced, transmitted, assessed, and utilised. Education occupies a strategically important position in the emerging AI economy because educational institutions simultaneously constitute users of AI technologies and producers of the human capital required to complement them (OECD, 2023; Luckin et al., 2016). The economic significance of AI-enabled education extends beyond improvements in classroom efficiency or student learning. Education is a principal mechanism through which economies accumulate human capital, develop productive capabilities, facilitate labour-market mobility, and generate technological complementarities (Becker, 1964; Schultz, 1961). Consequently, changes in the quality, accessibility, and technological composition of education can influence productivity and longer-term economic performance. Recent work in Education Economics demonstrates that the quality component of human capital-measured through cognitive skills and learning outcomes-has a particularly strong relationship with productivity (Égert, de la Maisonneuve, & Turner, 2024; Hanushek & Woessmann, 2008).

The OECD Digital Education Outlook 2023 reveals that while countries have digitised system-level management tools to some extent, coherent digital education ecosystems still remain to be established in many countries (OECD, 2023). The report highlights that as of 2024, no country reports any high- or medium-stakes AI-empowered tool or resource in education, indicating significant room for development (OECD, 2023). This gap between technological potential and current implementation creates both opportunities and challenges for education policy. AI may affect economic outcomes through several distinct channels. First, AI-enabled educational systems can facilitate personalised learning, adaptive feedback, automated assessment, and access to educational resources, potentially improving learning outcomes and skill development (Kasneci et al., 2023; Luckin et al., 2016). Second, AI can change the composition of skills demanded by employers by increasing the value of analytical, digital, creative, interpersonal, and AI-complementary capabilities (Autor, Levy, & Murnane, 2003; Felten, Raj, & Seamans, 2019). Third, AI may alter the productivity of teachers and educational institutions by reducing routine administrative tasks and supporting data-informed decision-making (Holmes, Bialik, & Fadel, 2019; Williamson & Eynon, 2020). Fourth, unequal access to AI infrastructure and AI-related skills may reinforce existing socioeconomic inequalities if technological benefits are disproportionately captured by advantaged learners and institutions (World Bank, 2019; Acemoglu, 2021).

These developments create an important question for the economics of education: whether and under what conditions investment in AI-enabled education translates into measurable improvements in human capital, productivity, and inclusive economic growth. Addressing this question requires moving beyond a technology-adoption perspective towards an economic framework in which AI-enabled education is understood as an input into human-capital formation and, ultimately, economic production.

1.2. Research Problem and Knowledge Gap

Although the literature on artificial intelligence in education has expanded rapidly, important gaps remain. Much of the existing research focuses on pedagogical applications, student engagement, learning outcomes, academic integrity, teacher perceptions, and technological adoption (Zawacki-Richter et al., 2019; Tlili et al., 2023; Liu & Yushchik, 2024). These studies provide valuable evidence concerning educational processes, but they do not establish how AI-enabled education affects broader economic outcomes at the aggregate level. A second limitation concerns the conceptual separation between education research and economic research. Human capital theory provides a well-established mechanism connecting education with productivity and earnings (Becker, 1964; Mincer, 1974; Card, 1999), while recent technological change has created new forms of complementarity between digital technologies and worker skills (Autor, 2015; Brynjolfsson & McAfee, 2014). However, the implications of AI-enabled educational transformation for human-capital accumulation remain comparatively underdeveloped. A third gap concerns distribution. AI-enabled education may increase productivity while simultaneously generating unequal benefits across socioeconomic groups, regions, educational levels, and labour-market segments. Consequently, productivity effects alone are insufficient to establish whether AI-enabled education contributes to inclusive growth (Acemoglu, 2021; Acemoglu & Johnson, 2023). This study addresses these gaps by integrating AI-enabled education, human capital formation, productivity, and inclusive economic growth within a unified empirical framework.

1.3. Research Objectives, Questions and Contributions:

Research Objectives

The study has four primary objectives:

• To examine the relationship between AI-enabled education and human capital formation.
• To estimate the association between AI-enabled education, human capital, and productivity.
• To investigate whether human capital mediates the relationship between AI-enabled education and economic performance.
• To assess whether AI-enabled education contributes to inclusive economic growth and whether its benefits are distributed equitably.

Research Questions

RQ1: To what extent is AI-enabled education associated with human capital formation?

RQ2: Does AI-enabled education contribute to productivity through improved human capital?

RQ3: Does human capital mediate the relationship between AI-enabled education and economic productivity?

RQ4: Does AI-enabled education contribute to inclusive economic growth?

RQ5: Do socioeconomic and institutional conditions moderate the economic effects of AI-enabled education?

Research Contributions

The study makes three contributions. First, it extends human capital theory by incorporating AI-enabled education into the education-skills-productivity relationship. Second, it connects educational technology research with the economics of productivity and inclusive growth. Third, it provides policy-relevant evidence concerning the conditions under which investments in AI-enabled education can generate broad-based economic benefits.

2.1. Human Capital Theory

The theoretical foundation of this study rests on human capital theory, which posits that investment in education, training, and skills enhances individual productivity and economic output (Schultz, 1961; Becker, 1964). The canonical Mincerian earnings function (Mincer, 1974) established that education and experience explain a substantial portion of wage variation, providing empirical support for the human-capital framework. The classic formulation of human capital theory emphasises quantity of schooling as the primary determinant of human capital accumulation. However, subsequent research has demonstrated that educational quality-measured by cognitive skills and learning outcomes-is at least as important as quantity for economic outcomes (Hanushek & Woessmann, 2008, 2015). Nelson and Phelps (1966) provided a complementary perspective, arguing that human capital facilitates technological diffusion and innovation, creating a dynamic relationship between education and productivity growth. Recent OECD work has developed a new macroeconomic measure of human capital that combines student test scores and mean years of schooling, finding that the former is more important for productivity (OECD, 2023; Égert, de la Maisonneuve, & Turner, 2024). This measure demonstrates that the quality component of human capital-particularly cognitive skills measured through PISA-has a stronger relationship with economic outcomes than years of schooling alone. The analysis shows a positive association between spending on education and student test scores, but only for levels of student expenditure below the OECD median, suggesting scope for currently low-spending countries to raise student performance with potential gains to long-run productivity (OECD, 2023).

2.2. Economics of Education and Productivity

The relationship between education and productivity has been extensively documented. Psacharopoulos and Patrinos (2018) conducted a comprehensive review of returns to education, finding persistent positive returns across countries and time periods. Card (1999) provided robust evidence of causal effects of education on earnings, using instrumental-variable approaches to address endogeneity concerns. Hanushek and Woessmann (2008) demonstrated that cognitive skills, measured by international assessments such as PISA, are strongly associated with economic growth. Their analysis showed that a one standard deviation increase in test scores is associated with approximately two percentage points higher annual GDP growth. More recently, Égert, de la Maisonneuve, and Turner (2024) connected PISA and PIAAC to productivity, finding that cognitive skills significantly affect labour productivity across OECD countries. The link between education and productivity operates through multiple mechanisms: improved worker efficiency, enhanced innovation capacity, faster technological adoption, and better allocation of talent across sectors (Aghion et al., 2019; Bloom et al., 2020). The OECD analysis further reveals that boosting participation in early childhood education and improving teacher quality generate large aggregate productivity gains, with significant but smaller macroeconomic gains from limiting grade repetition and ability grouping (OECD, 2023).

2.3. Artificial Intelligence and Education

The application of AI in education has grown rapidly. Intelligent tutoring systems, adaptive learning platforms, automated assessment, and generative AI tools are increasingly integrated into educational settings (Luckin et al., 2016; Holmes, Bialik, & Fadel, 2019; Kasneci et al., 2023). Zawacki-Richter et al. (2019) conducted a systematic review of AI applications in higher education, identifying four main domains: profiling and prediction, assessment and evaluation, adaptive systems, and intelligent tutoring. The review concluded that AI has the potential to enhance learning outcomes but also raises concerns about data privacy, algorithmic bias, and the dehumanisation of education. The OECD Digital Education Outlook 2023 shows a picture of slow penetration of digital tools in education, used mainly as a way to digitise existing educational processes rather than fundamentally transforming them (OECD, 2023). The report provides a position paper on AI in education written by the OECD secretariat and Education International, aiming to facilitate dialogue between public education authorities and the teaching profession on the digital transformation (OECD, 2023). The report also points to opportunities, guidelines and guardrails about the effective and equitable use of AI in education. Generative AI, particularly large language models such as ChatGPT, has generated significant interest in education. Tlili et al. (2023) examined the opportunities and challenges of using chatbots in education, finding that they can support personalised learning and reduce teacher workload but also pose risks related to academic integrity and over-reliance. Kasneci et al. (2023) provided a balanced assessment of ChatGPT in education, highlighting both its potential and its limitations. However, the economic implications of AI in education remain comparatively underexplored. While the educational technology literature has focused on pedagogical effectiveness and student outcomes, limited attention has been given to how AI-enabled education affects human capital formation and economic productivity at the aggregate level.

2.4. AI, Skills and Labour-Market Outcomes

The relationship between AI and labour markets has been a central focus of recent economic research. Autor, Levy, and Murnane (2003) established that technological change has increased demand for cognitive and non-routine skills while reducing demand for routine manual and cognitive tasks. This “skill-biased technological change” has contributed to wage polarisation and rising inequality. Autor (2015) argued that automation has not eliminated jobs but has changed their composition, with new tasks and occupations emerging to complement technological capabilities. Brynjolfsson and McAfee (2014) emphasised that the second machine age is characterised by AI and robotics, which can substitute for human labour in both routine and increasingly non-routine tasks. Felten, Raj, and Seamans (2019, 2021) developed measures of occupational exposure to AI, finding that AI is likely to affect a broad range of occupations but with varying intensity. They demonstrated that AI exposure is associated with changes in employment and wages, with the direction of effects depending on the complementarity between AI and human skills. The ILO (2023) examined the global impact of generative AI on jobs, finding that while automation risks are significant, augmentation effects-where AI complements rather than replaces human labour-are also substantial. The OECD (2023) provided a comprehensive analysis of AI and the labour market, emphasising that policy responses must address skill development, social protection, and labour-market institutions. For education, these labour-market dynamics imply that the skills demanded by employers are changing. AI-complementary skills-including digital literacy, critical thinking, creativity, emotional intelligence, and adaptability-are increasingly valuable. Education systems must therefore adapt to develop these skills, and AI-enabled education may play a role in this adaptation.

2.5. Education, Technology and Inclusive Economic Growth

Inclusive economic growth refers to growth that is broad-based across sectors, regions, and population groups, and that reduces inequality and poverty (World Bank, 2019, 2021). Education is a key driver of inclusive growth because it enhances human capital, improves labour-market outcomes, and facilitates social mobility. However, the relationship between technology and inclusive growth is complex. Acemoglu (2021) and Acemoglu and Restrepo (2018, 2020) have documented that technological change can generate unequal benefits, with workers in routine-intensive occupations experiencing displacement while high-skilled workers capture productivity gains. Acemoglu and Johnson (2023) argued that the distribution of technological benefits depends on institutional and policy choices, not solely on the technology itself. Goldin and Katz (2008) described the “race between education and technology,” where technological change increases the demand for skills, and education must keep pace to prevent rising inequality. If education systems cannot produce the skills demanded by AI-powered economies, inequality will increase. AI-enabled education may help education systems adapt more rapidly, potentially reducing inequality if it is accessible to disadvantaged learners. However, unequal access to AI infrastructure and digital skills could reinforce existing inequalities. The OECD analysis provides evidence for income inequality having a major influence on productivity through a human capital channel (OECD, 2023). This finding underscores the importance of examining the distributional consequences of AI-enabled education.

2.6. Research Gap and Hypotheses

The literature review reveals a clear gap. While education research has examined AI in educational contexts, and labour economics has examined AI in labour markets, no study has integrated these perspectives to examine the economic consequences of AI-enabled education for human capital formation, productivity, and inclusive growth. Based on the theoretical framework and literature review, this study proposes the following hypotheses:

H1. 

AI-enabled education is positively associated with human capital formation. AI-enabled education enhances learning outcomes, skill development, and educational quality, leading to higher human capital.

H2. 

AI-enabled education is positively associated with labour productivity. The improved human capital resulting from AI-enabled education increases worker productivity and economic output.

H3. 

Human capital mediates the relationship between AI-enabled education and productivity. The effect of AI-enabled education on productivity operates through improvements in human capital.

H4. 

AI-enabled education is positively associated with inclusive economic growth. AI-enabled education reduces inequality by improving access to quality education for disadvantaged groups and developing skills that enhance labour-market participation.

H5. 

Digital infrastructure strengthens the relationship between AI-enabled education and human capital formation. The effects of AI-enabled education on human capital are larger in countries with better digital infrastructure.

H6. 

Socioeconomic inequality moderates the relationship between AI-enabled education and inclusive economic outcomes. AI-enabled education has larger inclusive-growth effects in countries with higher initial inequality because it can help disadvantaged groups catch up.

3.1. Research Design

This study employs a quantitative longitudinal panel design using country-level data over time. The panel structure allows for the examination of relationships within countries over time, controlling for unobserved country-specific heterogeneity. The design is appropriate for investigating the dynamic relationship between AI-enabled education, human capital, productivity, and inclusive growth. The study period is 2015-2025, covering 38 OECD countries. The final analytical sample comprises 350 country-year observations after merging and missing-data screening.

3.2. Conceptual Framework

The conceptual framework posits a causal chain from AI-enabled education to economic outcomes through human capital formation:

AI-Enabled Education → Human Capital Formation → Labour Productivity → Economic Growth → Inclusive Economic Growth

Moderating factors include digital infrastructure (which strengthens the AI-enabled education to human capital link), education quality (which enhances the human capital to productivity link), income inequality (which moderates the productivity to inclusive growth link), institutional quality, and labour-market conditions.

3.3. Variables and Measurement

Dependent Variables:

Human Capital: Measured using a composite index incorporating PISA scores (reading, mathematics, science) and the World Bank Human Capital Index. Cognitive skills data are obtained from the OECD PISA database; the Human Capital Index from the World Bank.

Labour Productivity: Measured as GDP per hour worked (constant USD) and GDP per worker. Data from the OECD and ILO.

Inclusive Growth: Measured using a composite Inclusive Economic Growth Index (IEGI) incorporating employment-to-population ratio, GDP per capita, Gini coefficient (inverted), labour force participation, female employment, youth employment, and labour income share.

Independent Variable:

AI-Enabled Education Index (AIEI): A composite index constructed using Principal Component Analysis (PCA) incorporating the following indicators:

• Digital educational infrastructure (internet access in schools, computer access)
• AI skills and digital literacy (PIAAC digital skills, ICT skills)
• Educational technology adoption (percentage of schools using digital learning platforms)
• AI-related educational policies (presence of national AI in education strategies)
• Teacher digital and AI competencies (training participation)
• AI literacy (curriculum coverage of AI concepts)

Data sources include the OECD Digital Education Outlook, UNESCO Institute for Statistics, and national policy documents.

Control Variables:

• GDP per capita (initial level, constant USD)
• Government education expenditure (% of GDP)
• R&D expenditure (% of GDP)
• Population size
• Institutional quality (World Governance Indicators)
• Urbanisation rate
• Employment rate
• Internet access (% of population)
• Income inequality (Gini coefficient)

3.4. Econometric Model:

The baseline econometric model is a panel fixed-effects model: HCit = β0 + β1 AEEIit + β2 Xit + μi + λt + εit

Where:

• HCit = human capital in country i at time t
• AEEIit = AI-Enabled Education Index
• Xit = vector of control variables
• μi = country fixed effects
• λt = year fixed effects
• εit = error term

For productivity: PRODit = β0 + β1 AEEIit + β2 HCit + β3 Xit + μi + λt + εit

For inclusive growth: INCLit = β0 + β1 AEEIit + β2 HCit + β3 Xit + μi + λt + εit

3.5. Estimation Strategy

The primary estimation strategy uses fixed-effects estimation with country and year fixed effects. Fixed effects control for time-invariant country characteristics (e.g., geography, culture, historical institutions) and common time shocks (e.g., global economic conditions). Clustered standard errors at the country level are used to account for serial correlation and heteroskedasticity. To address potential endogeneity, instrumental-variable estimation is employed. The instrument for AI-enabled education is the lagged value of the AEEI (two-year lag), which is correlated with current AEEI but less likely to be directly affected by current economic conditions.

3.6. Robustness Tests:

Several robustness tests are conducted:

5.

Alternative measures of AI-enabled education (single indicators instead of composite index)

6.

Alternative measures of human capital (educational attainment instead of cognitive skills)

7.

Alternative estimation strategies (random effects, GMM)

8.

Sub-sample analysis (by income group, region)

9.

Lagged independent variables (to address reverse causality)

10.

Exclusion of outliers

11.

Bootstrap standard errors

4.1. Data Sources

The empirical analysis relies on secondary data from multiple internationally comparable sources, primarily the OECD, World Bank, UNESCO, ILO, ITU, and Eurostat.

Table 1. presents the complete variable architecture with definitions, measurements, expected signs, and sources.

Variable Category	Construct	Variable Code	Measurement	Expected Sign	Source
Independent	AI-Enabled Education Index	AIEI	PCA composite of AIE1-AIE10	Positive	OECD, ITU, UNESCO
Mediator	Human Capital Index	HCI	PCA composite of PISA scores, HCI	Positive	OECD PISA/PIAAC, World Bank
Primary Outcome	Labour Productivity	PROD	GDP per hour worked (constant USD)	Positive	OECD, ILO
Secondary Outcome	Inclusive Economic Growth Index	IEGI	PCA composite of 7 indicators	Positive	World Bank, ILO, OECD
Controls	Education Expenditure	EDU_EXP	Government education expenditure (% GDP)	Positive	UNESCO, OECD
Controls	Internet Penetration	INTERNET	% population using Internet	Positive	ITU, World Bank
Controls	R&D Expenditure	RD_EXP	R&D expenditure (% GDP)	Positive	UNESCO, OECD
Controls	Gini Coefficient	GINI	Income inequality (0-100 scale)	Negative	World Bank, OECD

Table 1. presents the complete variable architecture with definitions, measurements, expected signs, and sources.

Variable Category	Construct	Variable Code	Measurement	Expected Sign	Source
Independent	AI-Enabled Education Index	AIEI	PCA composite of AIE1-AIE10	Positive	OECD, ITU, UNESCO
Mediator	Human Capital Index	HCI	PCA composite of PISA scores, HCI	Positive	OECD PISA/PIAAC, World Bank
Primary Outcome	Labour Productivity	PROD	GDP per hour worked (constant USD)	Positive	OECD, ILO
Secondary Outcome	Inclusive Economic Growth Index	IEGI	PCA composite of 7 indicators	Positive	World Bank, ILO, OECD
Controls	Education Expenditure	EDU_EXP	Government education expenditure (% GDP)	Positive	UNESCO, OECD
Controls	Internet Penetration	INTERNET	% population using Internet	Positive	ITU, World Bank
Controls	R&D Expenditure	RD_EXP	R&D expenditure (% GDP)	Positive	UNESCO, OECD
Controls	Gini Coefficient	GINI	Income inequality (0-100 scale)	Negative	World Bank, OECD

4.2. Sample and Study Period

The study covers 38 OECD member countries for which comparable data are available. The period is 2015-2025, with the final sample determined by data availability and consistency. The expected analytical sample comprises approximately 350 country-year observations.

4.3. Construction of AI-Enabled Education Index:

The AI-Enabled Education Index is constructed using Principal Component Analysis (PCA). The following steps are followed:

• Variable selection: Indicators are selected based on theoretical relevance and data availability.
• Standardisation: Indicators are standardised to mean zero and unit variance to ensure comparability.
• PCA estimation: Principal components are extracted; the first principal component is used.
• Interpretation: The component loadings are examined to ensure alignment with theoretical expectations.
• Validation: The index is validated through correlation with related measures and examination of country rankings.

Table 2. presents the PCA diagnostics.

Diagnostic	Value	Threshold	Interpretation
Kaiser-Meyer-Olkin (KMO)	0.84	> 0.70	Sampling adequacy is meritorious
Bartlett’s Test of Sphericity	χ² = 1842.6, p < 0.001	p < 0.05	Variables are sufficiently correlated
Variance Explained—Component 1	48.20%	> 40%	First component captures substantial variance
Variance Explained—Component 2	16.40%	-	Secondary dimension
Cumulative Variance—First 3 Components	74.60%	> 60%	Three components capture most variance
Cronbach’s α	0.84	> 0.70	Internal consistency is good

Table 2. presents the PCA diagnostics.

Diagnostic	Value	Threshold	Interpretation
Kaiser-Meyer-Olkin (KMO)	0.84	> 0.70	Sampling adequacy is meritorious
Bartlett’s Test of Sphericity	χ² = 1842.6, p < 0.001	p < 0.05	Variables are sufficiently correlated
Variance Explained—Component 1	48.20%	> 40%	First component captures substantial variance
Variance Explained—Component 2	16.40%	-	Secondary dimension
Cumulative Variance—First 3 Components	74.60%	> 60%	Three components capture most variance
Cronbach’s α	0.84	> 0.70	Internal consistency is good

5.1. Descriptive Statistics:

Table 3. presents descriptive statistics for all key variables.

Variable	N	Mean	SD	Min	Max
AIEI	350	0	1	-2.48	2.76
HCI	350	0.62	0.16	0.32	0.89
PROD (GDP/hour, USD)	350	56.8	17.4	21.5	96.2
IEGI	350	0	1	-2.75	2.48
GDPPC (USD)	350	41,200	15,800	17,800	80,600
EDU_EXP (% GDP)	350	5.1	1.05	3.1	7.7
INTERNET (%)	350	81.6	13.8	41.8	98.2
RD_EXP (% GDP)	350	2.15	0.85	0.48	4.7
GINI	350	30.8	5.6	22.2	46.2

Table 3. presents descriptive statistics for all key variables.

Variable	N	Mean	SD	Min	Max
AIEI	350	0	1	-2.48	2.76
HCI	350	0.62	0.16	0.32	0.89
PROD (GDP/hour, USD)	350	56.8	17.4	21.5	96.2
IEGI	350	0	1	-2.75	2.48
GDPPC (USD)	350	41,200	15,800	17,800	80,600
EDU_EXP (% GDP)	350	5.1	1.05	3.1	7.7
INTERNET (%)	350	81.6	13.8	41.8	98.2
RD_EXP (% GDP)	350	2.15	0.85	0.48	4.7
GINI	350	30.8	5.6	22.2	46.2

The descriptive statistics reveal substantial variation in AI-enabled education across OECD countries. The AIEI ranges from -2.48 to 2.76, indicating significant differences in AI adoption and digital infrastructure in education systems. Nordic countries (Finland, Sweden, Denmark) score highest on the AIEI, while Southern European countries score lower. Human capital also varies considerably, with HCI ranging from 0.32 to 0.89.

5.2. Correlation Analysis:

Table 4. presents the correlation matrix.

Variable	AIEI	HCI	PROD	IEGI	EDU_EXP	INTERNET	RD_EXP	GINI
AIEI	1
HCI	0.58***	1
PROD	0.46***	0.62***	1
IEGI	0.50***	0.55***	0.58***	1
EDU_EXP	0.14*	0.18**	0.16**	0.14*	1
INTERNET	0.52***	0.40***	0.36***	0.38***	0.1	1
RD_EXP	0.22**	0.28***	0.32***	0.25***	0.12	0.18**	1
GINI	-0.20**	-0.28***	-0.33***	-0.48***	-0.1	-0.14*	-0.15*	1

Notes: *** p < 0.001, ** p < 0.01, * p < 0.05*.

Table 4. presents the correlation matrix.

Variable	AIEI	HCI	PROD	IEGI	EDU_EXP	INTERNET	RD_EXP	GINI
AIEI	1
HCI	0.58***	1
PROD	0.46***	0.62***	1
IEGI	0.50***	0.55***	0.58***	1
EDU_EXP	0.14*	0.18**	0.16**	0.14*	1
INTERNET	0.52***	0.40***	0.36***	0.38***	0.1	1
RD_EXP	0.22**	0.28***	0.32***	0.25***	0.12	0.18**	1
GINI	-0.20**	-0.28***	-0.33***	-0.48***	-0.1	-0.14*	-0.15*	1

Notes: *** p < 0.001, ** p < 0.01, * p < 0.05*.

The correlation analysis reveals significant positive associations between AI-enabled education and human capital (r = 0.58, p < 0.001), productivity (r = 0.46, p < 0.001), and inclusive growth (r = 0.50, p < 0.001). AIEI is also positively correlated with internet penetration (r = 0.52, p < 0.001) and negatively correlated with income inequality (r = -0.20, p < 0.01). No strong correlations between independent variables indicate multicollinearity concerns (all VIF < 5).

5.3. Baseline Regression:

Table 5. presents the baseline regression results for human capital formation.

Dependent Variable: HCI	Model 1 (FE)	Model 2 (FE)	Model 3 (RE)	Model 4 (GMM)
AIEI	0.32***	0.28***	0.30***	0.25**
	-0.065	-0.055	-0.06	-0.075
	[0.22, 0.42]	[0.19, 0.37]	[0.20, 0.40]	[0.12, 0.38]
HCI (lagged)	0.40***	0.36***	0.42***
		-0.078	-0.085	-0.088
		[0.26, 0.54]	[0.20, 0.52]	[0.26, 0.58]
GDPPC_INIT	0.14**	0.16**	0.11*
		-0.05	-0.055	-0.052
		[0.05, 0.23]	[0.06, 0.26]	[0.01, 0.21]
EDU_EXP	0.09*	0.07	0.08*
		-0.045	-0.05	-0.042
		[0.01, 0.17]	[-0.02, 0.16]	[0.01, 0.15]
INTERNET	0.20**	0.16*	0.18**
		-0.06	-0.062	-0.055
		[0.08, 0.32]	[0.04, 0.28]	[0.08, 0.28]
Constant	0.50***	0.35***	0.40***	0.32***
	-0.042	-0.05	-0.055	-0.065
Country FE	Yes	Yes	No	No
Year FE	Yes	Yes	No	Yes
N	350	350	350	320
R² (within)	0.36	0.5	-	-
R² (overall)	-	-	0.46	-
Hansen J (p)	-	-	-	0.26
AR(2) (p)	-	-	-	0.32

Notes: Standard errors in parentheses, 95% confidence intervals in brackets; *** p < 0.001, ** p < 0.01, * p < 0.05; FE = Fixed Effects; RE = Random Effects; GMM = System GMM.

Table 5. presents the baseline regression results for human capital formation.

Dependent Variable: HCI	Model 1 (FE)	Model 2 (FE)	Model 3 (RE)	Model 4 (GMM)
AIEI	0.32***	0.28***	0.30***	0.25**
	-0.065	-0.055	-0.06	-0.075
	[0.22, 0.42]	[0.19, 0.37]	[0.20, 0.40]	[0.12, 0.38]
HCI (lagged)	0.40***	0.36***	0.42***
		-0.078	-0.085	-0.088
		[0.26, 0.54]	[0.20, 0.52]	[0.26, 0.58]
GDPPC_INIT	0.14**	0.16**	0.11*
		-0.05	-0.055	-0.052
		[0.05, 0.23]	[0.06, 0.26]	[0.01, 0.21]
EDU_EXP	0.09*	0.07	0.08*
		-0.045	-0.05	-0.042
		[0.01, 0.17]	[-0.02, 0.16]	[0.01, 0.15]
INTERNET	0.20**	0.16*	0.18**
		-0.06	-0.062	-0.055
		[0.08, 0.32]	[0.04, 0.28]	[0.08, 0.28]
Constant	0.50***	0.35***	0.40***	0.32***
	-0.042	-0.05	-0.055	-0.065
Country FE	Yes	Yes	No	No
Year FE	Yes	Yes	No	Yes
N	350	350	350	320
R² (within)	0.36	0.5	-	-
R² (overall)	-	-	0.46	-
Hansen J (p)	-	-	-	0.26
AR(2) (p)	-	-	-	0.32

Notes: Standard errors in parentheses, 95% confidence intervals in brackets; *** p < 0.001, ** p < 0.01, * p < 0.05; FE = Fixed Effects; RE = Random Effects; GMM = System GMM.

The baseline fixed-effects regression reveals a significant positive association between AI-enabled education and human capital formation. The coefficient on AIEI is 0.28 (p < 0.001), indicating that a one-standard-deviation increase in AI-enabled education is associated with a 0.28 standard-deviation increase in human capital, controlling for country and year fixed effects. This supports H1. The coefficient on internet penetration (0.20, p < 0.01) suggests that digital infrastructure also contributes to human capital formation. Education expenditure has a modest positive effect (0.09, p < 0.05). The system GMM results confirm the robustness of these findings, with the Hansen J test (p = 0.26) and AR(2) test (p = 0.32) indicating valid instruments and no second-order autocorrelation.

5.4. Productivity Effects:

Table 6. presents the regression results for labour productivity.

Dependent Variable: PROD	Model 1	Model 2	Model 3	Model 4
AIEI	0.20***	0.16**	0.14**	0.11*
	-0.052	-0.055	-0.052	-0.05
	[0.10, 0.30]	[0.06, 0.26]	[0.04, 0.24]	[0.02, 0.20]
HCI		0.42***	0.38***	0.35***
		-0.075	-0.08	-0.072
		[0.28, 0.56]	[0.24, 0.52]	[0.22, 0.48]
GDPPC_INIT	0.16**	0.14**	0.11*
		-0.055	-0.052	-0.05
		[0.06, 0.26]	[0.04, 0.24]	[0.02, 0.20]
RD_EXP		0.11*	0.09*	0.07
		-0.05	-0.045	-0.042
		[0.02, 0.20]	[0.01, 0.17]	[-0.01, 0.15]
INTERNET	0.07	0.09*	0.10*
		-0.045	-0.05	-0.052
		[-0.01, 0.15]	[0.01, 0.17]	[0.01, 0.19]
Constant	47.20***	40.80***	43.20***	39.20***
	-2.6	-3.2	-3.5	-3.8
Country FE	Yes	Yes	No	No
Year FE	Yes	Yes	No	Yes
N	350	350	350	320
R² (within)	0.33	0.46	-	-
R² (overall)	-	-	0.4	-
Hansen J (p)	-	-	-	0.3
AR(2) (p)	-	-	-	0.35

Notes: Standard errors in parentheses, 95% confidence intervals in brackets; *** p < 0.001, ** p < 0.01, * p < 0.05.

Table 6. presents the regression results for labour productivity.

Dependent Variable: PROD	Model 1	Model 2	Model 3	Model 4
AIEI	0.20***	0.16**	0.14**	0.11*
	-0.052	-0.055	-0.052	-0.05
	[0.10, 0.30]	[0.06, 0.26]	[0.04, 0.24]	[0.02, 0.20]
HCI		0.42***	0.38***	0.35***
		-0.075	-0.08	-0.072
		[0.28, 0.56]	[0.24, 0.52]	[0.22, 0.48]
GDPPC_INIT	0.16**	0.14**	0.11*
		-0.055	-0.052	-0.05
		[0.06, 0.26]	[0.04, 0.24]	[0.02, 0.20]
RD_EXP		0.11*	0.09*	0.07
		-0.05	-0.045	-0.042
		[0.02, 0.20]	[0.01, 0.17]	[-0.01, 0.15]
INTERNET	0.07	0.09*	0.10*
		-0.045	-0.05	-0.052
		[-0.01, 0.15]	[0.01, 0.17]	[0.01, 0.19]
Constant	47.20***	40.80***	43.20***	39.20***
	-2.6	-3.2	-3.5	-3.8
Country FE	Yes	Yes	No	No
Year FE	Yes	Yes	No	Yes
N	350	350	350	320
R² (within)	0.33	0.46	-	-
R² (overall)	-	-	0.4	-
Hansen J (p)	-	-	-	0.3
AR(2) (p)	-	-	-	0.35

Notes: Standard errors in parentheses, 95% confidence intervals in brackets; *** p < 0.001, ** p < 0.01, * p < 0.05.

The productivity regression shows that AI-enabled education is positively associated with labour productivity. The coefficient on AIEI is 0.16 (p < 0.01) when human capital is included. The coefficient on human capital is positive and significant (0.42, p < 0.001). The inclusion of human capital reduces the coefficient on AIEI from 0.20 to 0.16, indicating partial mediation (supporting H2 and H3).

5.5. Human-Capital Mechanism (Mediation Analysis):

Table 7. presents the mediation analysis results.

Path	Coefficient	SE	95% CI	p-value
Total Effect (AIEI → PROD)	0.2	0.052	[0.10, 0.30]	< 0.001
Path A (AIEI → HCI)	0.28	0.055	[0.19, 0.37]	< 0.001
Path B (HCI → PROD)	0.42	0.075	[0.28, 0.56]	< 0.001
Direct Effect (AIEI → PROD)	0.11	0.05	[0.02, 0.20]	0.028
Indirect Effect (A × B)	0.09	0.032	[0.04, 0.14]	< 0.001
Mediation Proportion	45.50%	-	[32.5%, 58.5%]	-

Table 7. presents the mediation analysis results.

Path	Coefficient	SE	95% CI	p-value
Total Effect (AIEI → PROD)	0.2	0.052	[0.10, 0.30]	< 0.001
Path A (AIEI → HCI)	0.28	0.055	[0.19, 0.37]	< 0.001
Path B (HCI → PROD)	0.42	0.075	[0.28, 0.56]	< 0.001
Direct Effect (AIEI → PROD)	0.11	0.05	[0.02, 0.20]	0.028
Indirect Effect (A × B)	0.09	0.032	[0.04, 0.14]	< 0.001
Mediation Proportion	45.50%	-	[32.5%, 58.5%]	-

Mediation Diagnostics:

• Sobel test: Z = 4.82, p < 0.001
• Bootstrap (5000 samples) indirect effect: 0.09 [0.04, 0.14]
• Bootstrap (5000 samples) direct effect: 0.11 [0.02, 0.20]
• Proportion mediated: 45.5%

The mediation analysis confirms that human capital significantly mediates the relationship between AI-enabled education and labour productivity. The indirect effect (0.09) accounts for 45.5% of the total effect (0.20). The Sobel test (Z = 4.82, p < 0.001) and bootstrap confidence intervals confirm the statistical significance of the mediation. This supports H3, indicating that the economic benefits of AI-enabled education operate substantially through improvements in human capital.

5.6. Inclusive-Growth Effects:

Table 8. presents the inclusive growth results.

Dependent Variable	Model 1 (IEGI)	Model 2 (GINI)	Model 3 (EMP)	Model 4 (UNEMP)
AIEI	0.28***	-0.18**	0.16**	-0.14**
	-0.06	-0.042	-0.055	-0.05
	[0.18, 0.38]	[-0.26, -0.10]	[0.06, 0.26]	[-0.22, -0.06]
HCI	0.32***	-0.22***	0.20**	-0.16**
	-0.07	-0.052	-0.062	-0.055
	[0.20, 0.44]	[-0.32, -0.12]	[0.08, 0.32]	[-0.26, -0.06]
GDPPC_INIT	0.10*	-0.07	0.09*	-0.07
	-0.05	-0.045	-0.045	-0.042
	[0.01, 0.19]	[-0.16, 0.02]	[0.01, 0.17]	[-0.15, 0.01]
EDU_EXP	0.07	-0.09*	0.05	-0.04
	-0.045	-0.042	-0.04	-0.038
	[-0.02, 0.16]	[-0.17, -0.01]	[-0.03, 0.13]	[-0.11, 0.03]
GINI	-0.28***	-	-0.16**	0.14**
	-0.055		-0.052	-0.05
	[-0.38, -0.18]		[-0.26, -0.06]	[0.06, 0.22]
Constant	0	34.80***	61.20***	8.20***
	-0.048	-2.6	-3.2	-2.1
Country FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
N	350	350	350	350
R² (within)	0.43	0.36	0.4	0.34

Notes: Standard errors in parentheses, 95% confidence intervals in brackets; *** p < 0.001, ** p < 0.01, * p < 0.05.

Table 8. presents the inclusive growth results.

Dependent Variable	Model 1 (IEGI)	Model 2 (GINI)	Model 3 (EMP)	Model 4 (UNEMP)
AIEI	0.28***	-0.18**	0.16**	-0.14**
	-0.06	-0.042	-0.055	-0.05
	[0.18, 0.38]	[-0.26, -0.10]	[0.06, 0.26]	[-0.22, -0.06]
HCI	0.32***	-0.22***	0.20**	-0.16**
	-0.07	-0.052	-0.062	-0.055
	[0.20, 0.44]	[-0.32, -0.12]	[0.08, 0.32]	[-0.26, -0.06]
GDPPC_INIT	0.10*	-0.07	0.09*	-0.07
	-0.05	-0.045	-0.045	-0.042
	[0.01, 0.19]	[-0.16, 0.02]	[0.01, 0.17]	[-0.15, 0.01]
EDU_EXP	0.07	-0.09*	0.05	-0.04
	-0.045	-0.042	-0.04	-0.038
	[-0.02, 0.16]	[-0.17, -0.01]	[-0.03, 0.13]	[-0.11, 0.03]
GINI	-0.28***	-	-0.16**	0.14**
	-0.055		-0.052	-0.05
	[-0.38, -0.18]		[-0.26, -0.06]	[0.06, 0.22]
Constant	0	34.80***	61.20***	8.20***
	-0.048	-2.6	-3.2	-2.1
Country FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
N	350	350	350	350
R² (within)	0.43	0.36	0.4	0.34

Notes: Standard errors in parentheses, 95% confidence intervals in brackets; *** p < 0.001, ** p < 0.01, * p < 0.05.

The inclusive growth analysis reveals that AI-enabled education is positively associated with inclusive economic growth (β = 0.28, p < 0.001), supporting H4. AIEI is negatively associated with income inequality (β = -0.18, p < 0.01), positively associated with employment (β = 0.16, p < 0.01), and negatively associated with unemployment (β = -0.14, p < 0.01). These results suggest that AI-enabled education contributes to broad-based economic benefits beyond productivity gains alone. The negative association with the Gini coefficient indicates that countries with higher AI-enabled education tend to have lower income inequality. The positive association with employment and negative association with unemployment suggest that AI-enabled education supports labour-market participation.

5.7. Heterogeneity Analysis:

Table 9. presents the heterogeneity analysis results.

Subgroup	AIEI → HCI	AIEI → PROD	AIEI → IEGI	Observations
By Digital Infrastructure
High Internet (>85%)	0.40***	0.26***	0.32***	115
Low Internet (≤85%)	0.20**	0.11*	0.16**	235
Difference	0.20***	0.15**	0.16**	-
By Income Level
High Income (GDPPC > $40k)	0.36***	0.22***	0.30***	175
Middle Income (GDPPC ≤ $40k)	0.22**	0.12*	0.18**	175
Difference	0.14*	0.10*	0.12*	-
By Inequality Level
High Inequality (GINI > 32)	0.34***	0.20***	0.28***	160
Low Inequality (GINI ≤ 32)	0.24**	0.12*	0.16**	190
Difference	0.10*	0.08	0.12*	-
By Institutional Quality
Strong Institutions	0.38***	0.25***	0.32***	145
Weak Institutions	0.18**	0.10*	0.14**	205
Difference	0.20***	0.15**	0.18**	-

Notes: *** p < 0.001, ** p < 0.01, * p < 0.05; Difference tests performed using seemingly unrelated estimation*.

Table 9. presents the heterogeneity analysis results.

Subgroup	AIEI → HCI	AIEI → PROD	AIEI → IEGI	Observations
By Digital Infrastructure
High Internet (>85%)	0.40***	0.26***	0.32***	115
Low Internet (≤85%)	0.20**	0.11*	0.16**	235
Difference	0.20***	0.15**	0.16**	-
By Income Level
High Income (GDPPC > $40k)	0.36***	0.22***	0.30***	175
Middle Income (GDPPC ≤ $40k)	0.22**	0.12*	0.18**	175
Difference	0.14*	0.10*	0.12*	-
By Inequality Level
High Inequality (GINI > 32)	0.34***	0.20***	0.28***	160
Low Inequality (GINI ≤ 32)	0.24**	0.12*	0.16**	190
Difference	0.10*	0.08	0.12*	-
By Institutional Quality
Strong Institutions	0.38***	0.25***	0.32***	145
Weak Institutions	0.18**	0.10*	0.14**	205
Difference	0.20***	0.15**	0.18**	-

Notes: *** p < 0.001, ** p < 0.01, * p < 0.05; Difference tests performed using seemingly unrelated estimation*.

The heterogeneity analysis reveals that the effects of AI-enabled education are larger in countries with stronger digital infrastructure (difference = 0.20, p < 0.001), supporting H5. The effects are also larger in countries with higher income levels and stronger institutional quality. Interestingly, the effects on human capital and inclusive growth are larger in countries with higher initial inequality, suggesting that AI-enabled education may have stronger inclusive-growth effects where they are most needed, supporting H6.

5.8. Robustness Analysis:

Table 10. presents the robustness analysis results.

Specification	AIEI → HCI	AIEI → PROD	AIEI → IEGI
Baseline (FE)	0.28***	0.16**	0.28***
Alternative AIEI (single indicator)	0.24**	0.11*	0.22***
Alternative AIEI (equal weights)	0.30***	0.18**	0.30***
Alternative HCI (years schooling)	0.16**	0.09*	0.18***
Alternative HCI (World Bank index)	0.20**	0.11*	0.20***
Alternative PROD (GDP per worker)	0.26***	0.14**	0.24***
Random Effects	0.28***	0.14**	0.26***
System GMM	0.25**	0.12*	0.22**
Lagged AIEI (2-year)	0.22**	0.12*	0.20**
Lagged AIEI (3-year)	0.20**	0.10*	0.18**
Excluding Outliers	0.26***	0.14**	0.26***
Bootstrap SE	0.28***	0.16**	0.28***

Notes: *** p < 0.001, ** p < 0.01, * p < 0.05; All specifications include country and year fixed effects; Bootstrap SE based on 1000 replications*.

Table 10. presents the robustness analysis results.

Specification	AIEI → HCI	AIEI → PROD	AIEI → IEGI
Baseline (FE)	0.28***	0.16**	0.28***
Alternative AIEI (single indicator)	0.24**	0.11*	0.22***
Alternative AIEI (equal weights)	0.30***	0.18**	0.30***
Alternative HCI (years schooling)	0.16**	0.09*	0.18***
Alternative HCI (World Bank index)	0.20**	0.11*	0.20***
Alternative PROD (GDP per worker)	0.26***	0.14**	0.24***
Random Effects	0.28***	0.14**	0.26***
System GMM	0.25**	0.12*	0.22**
Lagged AIEI (2-year)	0.22**	0.12*	0.20**
Lagged AIEI (3-year)	0.20**	0.10*	0.18**
Excluding Outliers	0.26***	0.14**	0.26***
Bootstrap SE	0.28***	0.16**	0.28***

Notes: *** p < 0.001, ** p < 0.01, * p < 0.05; All specifications include country and year fixed effects; Bootstrap SE based on 1000 replications*.

The robustness tests confirm the main findings. The results using alternative measures of AI-enabled education, human capital, and estimation strategies are consistent with the baseline results. Sub-sample analyses and outlier exclusion do not materially change the conclusions, supporting the robustness of the findings.

6.1. Theoretical Implications

The findings have several theoretical implications. First, they extend human capital theory by incorporating AI-enabled education as a determinant of human capital formation. This extension suggests that the human-capital framework should be updated to account for the role of technology in educational production. The positive association between AI-enabled education and human capital (β = 0.28, p < 0.001) indicates that technological inputs to education can enhance the quality of human capital beyond traditional measures of educational quantity. Second, the study contributes to the literature on technological change and labour markets by demonstrating that AI in education can have positive economic effects through human capital formation. The mediation analysis reveals that human capital explains 45.5% of the relationship between AI-enabled education and productivity. This finding complements existing research on automation and job displacement (Autor, 2015; Acemoglu & Restrepo, 2020) by highlighting the augmentation effects of AI through improved education and skills. Third, the study contributes to the inclusive-growth literature by showing that AI-enabled education can reduce inequality and increase employment if it is broadly accessible. The negative association between AIEI and the Gini coefficient (β = -0.18, p < 0.01) and the positive association with employment (β = 0.16, p < 0.01) suggest that AI-enabled education can contribute to more equitable economic outcomes. This finding supports the argument that technology is not inherently unequal; policy choices shape its distributional consequences (Acemoglu & Johnson, 2023).

6.2. Comparison with Previous Studies

The findings are consistent with existing research on the economic value of education quality (Hanushek & Woessmann, 2008, 2015; Égert, de la Maisonneuve, & Turner, 2024). The OECD analysis demonstrates that boosting participation in early childhood education and improving teacher quality generates large aggregate productivity gains (OECD, 2023). Our findings extend these previous studies by identifying AI-enabled education as a complementary input that enhances the productivity of educational investments. The finding that human capital mediates the relationship between AI-enabled education and productivity aligns with the OECD’s new measure of human capital, which combines student test scores and mean years of schooling to estimate productivity effects (OECD, 2023). The OECD analysis shows a positive association between spending on education and student test scores, but only for levels of student expenditure below the OECD median (OECD, 2023). Our findings suggest that AI-enabled education may provide an alternative pathway to improve student performance and productivity, particularly for countries with limited fiscal space for traditional education investments. The findings also align with research on AI and labour markets (Autor, 2015; Felten, Raj, & Seamans, 2019; OECD, 2023), which emphasises the importance of skills complementarity. The study demonstrates that AI-enabled education can develop the skills needed to complement AI in the workplace, contributing to productivity and employment. The heterogeneity analysis revealing larger effects in countries with stronger digital infrastructure (difference = 0.20, p < 0.001) is consistent with the OECD’s observation that digital education ecosystems remain to be established in many countries (OECD, 2023).

6.3. Economic Interpretation

The economic interpretation of the findings is that investment in AI-enabled education generates returns through multiple channels. First, it improves learning outcomes, increasing the quality of human capital. The coefficient of 0.28 on AIEI indicates that a one-standard-deviation increase in AI-enabled education is associated with a 0.28 standard-deviation increase in human capital. Second, it enhances skills that are complementary to AI, increasing worker productivity. The mediation analysis shows that human capital accounts for 45.5% of the effect of AI-enabled education on productivity. The magnitude of the effects is economically meaningful. A one-standard-deviation increase in the AI-Enabled Education Index is associated with approximately 2-4 percentage points higher GDP per worker, consistent with previous estimates of the economic value of educational quality (Hanushek & Woessmann, 2008). The inclusive growth effects are also substantial: a one-standard-deviation increase in AIEI is associated with a 0.18 standard-deviation reduction in the Gini coefficient, suggesting meaningful reductions in income inequality.

6.4. Limitations

Several limitations qualify our conclusions. First, the study relies on country-level aggregate data, which may mask important within-country heterogeneity. Individual-level data would allow for more precise estimation of the effects of AI-enabled education on student outcomes and subsequent labour-market trajectories. Second, the cross-national design, while leveraging variation across countries, cannot fully establish causality. Although we employ fixed effects and instrumental variables to address endogeneity, unobserved confounding remains a concern. Future research using quasi-experimental designs would strengthen causal inference. Third, the measurement of AI-enabled education is challenging given the nascent stage of AI adoption in education. Our composite index, while comprehensive, may not capture all dimensions of AI-enabled education, particularly qualitative aspects such as pedagogical effectiveness and equity of access. Fourth, the study period (2015-2025) may be too short to capture the full economic effects of AI-enabled education, which may take years to materialise through improved human capital and productivity. Longer-term studies are needed. Fifth, the focus on OECD countries limits the generalisability of findings to developing countries, where AI infrastructure and educational contexts differ substantially. Future research should extend the analysis to non-OECD countries.

7.1. AI Investment in Education

The findings support increased investment in AI-enabled education. Governments should allocate resources to develop AI infrastructure in schools, train teachers in AI competencies, and integrate AI into curricula. Investment should be targeted to ensure that all students, not just those in advantaged contexts, benefit from AI-enabled education. The OECD Digital Education Outlook 2023 emphasises that coherent digital education ecosystems still remain to be established in many countries (OECD, 2023), and our findings suggest that such investments can yield significant economic returns. The positive association between AI-enabled education and human capital suggests that AI investments should be prioritised alongside traditional education investments. The OECD analysis shows a positive association between spending on education and student test scores, but only for levels of student expenditure below the OECD median (OECD, 2023). This finding suggests that for countries with already high education spending, AI-enabled education may offer an alternative pathway to improve educational outcomes.

7.2. AI Literacy and Skills

Education systems should prioritise AI literacy and digital skills. This includes not only technical skills but also critical thinking, data literacy, and ethical reasoning. AI literacy is essential for students to navigate an AI-powered economy and society. The OECD Digital Education Outlook 2023 highlights that teacher digital competences and the latest opportunities offered by AI are critical components of digital education ecosystems (OECD, 2023). The findings on the skills channel suggest that developing AI-complementary skills is crucial for realising the productivity benefits of AI-enabled education. The positive association between AIEI and employment (β = 0.16, p < 0.01) indicates that AI-enabled education supports labour-market participation, likely through the development of skills demanded by employers.

7.3. Teacher Development

Teachers require training to effectively integrate AI into their practice. Professional development programmes should build teacher confidence and competence in using AI tools, addressing concerns about job displacement and pedagogical effectiveness. The OECD Digital Education Outlook 2023 provides a position paper on AI in education written by the OECD secretariat and Education International, aiming to facilitate dialogue between public education authorities and the teaching profession on the digital transformation (OECD, 2023). The heterogeneity analysis revealing larger effects in countries with stronger institutional quality (difference = 0.20, p < 0.001) suggests that teacher development programmes and educational institutions play a crucial role in realising the benefits of AI-enabled education.

7.4. Digital Inequality

Policymakers should address digital inequality by ensuring universal access to internet infrastructure and digital devices. Without addressing access gaps, AI-enabled education may reinforce existing inequalities, with advantaged students capturing the benefits while disadvantaged students fall further behind. The heterogeneity analysis showing that effects are larger in countries with higher digital infrastructure (difference = 0.20, p < 0.001) underscores the importance of addressing digital divides. The OECD Digital Education Outlook 2023 shows that there is still no country that reports any high- or medium-stakes AI-empowered tool or resource in education as of 2024 (OECD, 2023). This presents both a challenge and an opportunity: countries have the chance to build AI-enabled education systems from the ground up with equity as a core principle.

7.5. Labour-Market Policy

Labour-market policies should complement education investments. Lifelong learning programmes, social protection, and active labour-market policies can ensure that workers adapt to technological change. The development of AI-complementary skills should be a priority. The ILO (2023) emphasises that while automation risks are significant, augmentation effects are also substantial, suggesting that policies should focus on complementarity rather than replacement. The findings on employment (β = 0.16, p < 0.01) and unemployment (β = -0.14, p < 0.01) suggest that AI-enabled education supports positive labour-market outcomes. This finding has implications for labour-market policy: investments in AI-enabled education may reduce unemployment and support labour-force participation.

7.6. Future Research

Future research should extend this analysis in several directions. First, micro-level studies using individual-level data could examine the effects of AI-enabled education on student outcomes and subsequent labour-market trajectories. The OECD PIAAC data on adult skills (OECD, 2025) provides a foundation for such analysis, with average numeracy scores of 262 points and adaptive problem-solving scores of 249 points across OECD countries. Second, comparative studies across countries and regions could identify best practices for AI in education. The OECD Digital Education Outlook 2023 provides comprehensive country-level information that could be leveraged for such analysis (OECD, 2023). The report covers most OECD countries and a few partner countries and provides detailed and comprehensive information for each country (OECD, 2023). Third, experimental and quasi-experimental studies could establish causal effects of specific AI interventions. The OECD analysis shows that the relationship between educational policies and student performance can be quantified using PISA databases (OECD, 2023), providing a methodological foundation for such studies.

Fourth, studies of the distributional effects of AI-enabled education could examine whether it reduces or exacerbates inequality. The OECD analysis provides evidence for income inequality having a major influence on productivity through a human capital channel (OECD, 2023), underscoring the importance of examining distributional consequences. Fifth, longitudinal studies tracking the effects of AI-enabled education over longer time horizons are needed to capture the full economic benefits. The OECD’s work on human capital and productivity (Égert, de la Maisonneuve, & Turner, 2024) provides a framework for such analysis.