Learning after COVID-19 and the ICT career aspirations: Are students entering the AI era with weaker skills?

Technical Summary: Learning after COVID-19 and the ICT Career Aspirations

Problem Statement
The paper addresses a critical gap in understanding whether students are entering the generative AI era with sufficient educational foundations to support future digital career aspirations. While existing literature extensively documents post-pandemic learning losses, digital inequalities, and the rising relevance of ICT skills, these dimensions are often analyzed in isolation. There is a lack of integrated, cross-national evidence linking the structural characteristics of learning environments (specifically student autonomy, digital skills, and teacher support) to longitudinal changes in students' ICT-related career expectations between 2018 and 2022. The study seeks to determine if education systems with stronger structural characteristics are better positioned to foster ICT career aspirations in the context of digital and AI-driven labor markets.

Methodology
The study employs a mixed-method approach utilizing harmonized country-level data from the OECD Programme for International Student Assessment (PISA) 2018 and 2022. The analytical pipeline is a fully reproducible, end-to-end quantitative framework implemented in Python, comprising the following stages:

1. Data Harmonization and Encoding: Raw PISA tables were parsed to extract country-level indicators for student autonomy, digital learning skills, and perceived teacher support (operationalized as effects on academic performance in Mathematics, Reading, and Science). Longitudinal changes in career aspirations (ICT, Health, Science & Engineering, Science & Technical) were computed as percentage point differences between 2018 and 2022.
2. Multivariate Profiling: Standardized indicators were subjected to unsupervised clustering (KMeans) to identify typologies of education systems based on distinct configurations of autonomy, digital preparedness, and teacher support.
3. Latent Representation Learning: A Variational Autoencoder (VAE) was trained on the multidimensional input vectors to project education systems into a low-dimensional (2D) latent space. This latent space serves as a continuous representation of "structural readiness" for the digital and AI era. A composite AI Readiness Index was constructed as a linear aggregation of these latent coordinates.
4. Cross-Domain Consistency: Pearson correlations and paired t-tests were used to evaluate the stability of learning environment effects (autonomy, digital skills, teacher support) across Mathematics, Reading, and Science.
5. Predictive and Inferential Modeling:
   • Regression Analysis: Multivariate linear regression was used to quantify the marginal associations between learning environment indicators and changes in ICT career aspirations.
   • Classification: Linear Discriminant Analysis (LDA) was applied to distinguish between education systems with high versus low growth in ICT aspirations.
   • Counterfactual Simulation: The regression model was used to simulate the impact of increasing student autonomy by one standard deviation while holding other factors constant.
   • Probabilistic Modeling: A Bayesian network was learned from discretized aspiration change variables to explore conditional dependencies between ICT aspirations and other occupational domains.

Key Results

• Aspiration Dynamics: There is a global but uneven increase in ICT career aspirations between 2018 and 2022. While most countries experienced moderate growth, significant outliers exist with both accelerated growth (e.g., Turkey, Kosovo) and declines (e.g., Poland, Netherlands). ICT aspirations appear to be decoupling from traditional STEM trajectories, evolving as a distinct domain.
• Predictors of Growth: Digital skills emerge as the strongest and most consistent positive predictor of ICT career aspiration growth. In contrast, the regression analysis revealed negative associations for teacher support and autonomy, though the authors interpret these as context-dependent rather than causal negatives.
• Role of Autonomy and Support: The relationship between autonomy and digital skills is conditional. In systems with low teacher support, autonomy acts as a compensatory mechanism; however, in high-support systems, the marginal effect of autonomy is reduced. The study finds that autonomy alone is insufficient to drive ICT aspirations without complementary instructional support.
• Latent Readiness: The VAE analysis reveals that educational readiness is a continuous, multi-dimensional construct rather than a categorical property. Systems with favorable configurations (high autonomy, strong digital skills, robust support) tend to show higher ICT aspiration growth, though significant heterogeneity remains within similar structural profiles.
• Independence of Domains: The Bayesian network analysis found no significant conditional dependencies between changes in ICT aspirations and changes in other occupational domains (Health, Science & Engineering), suggesting ICT career dynamics evolve relatively independently.

Key Contributions

• Integrated Framework: The study shifts the focus from isolated analyses of learning loss or digital skills to an integrated framework linking structural learning conditions with career aspiration dynamics across a longitudinal timeline.
• Methodological Novelty: It introduces the application of Variational Autoencoders (VAE) and Bayesian networks to PISA data to capture non-linear latent structures and probabilistic dependencies in educational readiness, moving beyond traditional linear rankings.
• Structural Perspective: The research reframes the transition to the generative AI era not merely as a technological shift but as a structural educational challenge shaped by institutional and pedagogical conditions.

Significance and Claims
The paper claims that educational readiness for the AI era is multidimensional and that ICT aspirations evolve relatively independently from other career domains. It concludes that while digital skills are a primary driver of ICT career interest, structural factors like teacher support play a complementary, enabling role. The study modestly asserts that educational factors alone do not fully determine ICT career dynamics; rather, they act as enabling conditions influenced by broader contextual factors such as labor market conditions and national digital strategies. Consequently, the authors argue that policy interventions focusing solely on increasing student autonomy are unlikely to yield substantial shifts in ICT career aspirations without parallel improvements in digital competencies and pedagogical support.