Predicting Sustainable Development Goals with Global AI Readiness: A Hybrid Bayesian Belief Network and Artificial Neural Network Approach

This is the authors' abstract. We don't add key points for this paper.

This study examines whether global artificial intelligence (AI) readiness can be understood as a structured national capability system and whether its dimensions can predict Sustainable Development Goal (SDG) performance. Drawing on the 2025 Government AI Readiness Index and the 2025 SDG Index, the analysis used AI readiness data for 195 countries and a matched AI-SDG sample of 167 countries. A hybrid modelling strategy was adopted. First, Bayesian Belief Networks (BBNs) were used to model the probabilistic architecture of AI readiness at pillar and dimension levels. Second, Least Absolute Shrinkage and Selection Operator (LASSO) and artificial neural network-multilayer perceptron (ANN-MLP) models were used to predict SDG scores from fourteen AI readiness dimensions. The BBN results show that AI readiness is not a loose aggregation of indicators, but an interdependent architecture shaped by policy commitment, e-Government delivery, government digital policy, data quality, and regulatory compliance. The dimension-level BBN outperformed the pillar-level model, supporting the use of granular readiness dimensions for SDG prediction. In the predictive stage, LASSO performed slightly better than MLP, whereas both the LASSO and ANN analysis revealed that societal transition and data quality were the most consistent predictors of SDG. The conclusion follows that AI readiness contributes to sustainable development only when technical capacity is embedded in social, institutional, and data systems. Policy-wise, the findings motivate an AI Readiness-to-Sustainable Development Goals Policy Framework (AI-SDG Framework), which recommends prioritising institutional steering, reliable data infrastructure, public-sector delivery capacity, and social embedding before scaling AI diffusion.