Artificial Intelligence in Regulating Production Volumes for Sustainable Development: Qualitative and Quantitative Aspects

This study explores the intersection of artificial intelligence and economic modeling by extending the classical Cobb–Douglas production function into a custom neural network architecture implemented in TensorFlow. Motivated by the growing emphasis on sustainable development and its often ambiguous role in economic performance, the research addresses a gap in existing literature: the lack of integrated models that quantify the effect of Sustainable Development Goals (SDGs) within production functions. While previous studies have assessed SDGs and productivity separately, few have embedded sustainability metrics directly into core economic frameworks alongside traditional inputs like capital and labor. To fill this gap, the proposed model features trainable subcomponents for total factor productivity (TFP), physical capital, human capital, and SDG-related factors. Key coefficients—including capital elasticity (α), labor elasticity (β), and an SDG penalty term (γ)—are optimized using gradient descent. Experimental results reveal that while SDG constraints can initially appear to limit economic output, the model identifies conditions under which specific SDG factors contribute positively to productivity. To manage this duality, a three-level AI-based regulatory mechanism is introduced: (1) post-training SDG weighting based on their marginal output contribution, (2) filtering of influential SDG indicators via the Pareto principle, and (3) architectural separation of SDG pathways with controlled activation. These innovations enhance the interpretability and efficiency of sustainability-aware economic forecasting. The findings not only challenge the assumption of a trade-off between growth and sustainability but also suggest that targeted regulation of sustainability inputs can optimize outcomes. Future work may expand this framework to sector-specific models or broader macroeconomic simulations.