Workers’ perceptions of the economic impact of generative AI on sustainable economic development: evidence from ensemble tree models and explainable machine learning

Abstract Generative artificial intelligence (GenAI) is transforming society, including its economic foundations. This study examines the drivers and barriers shaping perceptions of GenAI’s influence on sustainable economic development (SED) among the working population. We adopt a conceptual framework in which these perceptions are shaped by four cognitive factors: three derived from the Technology Readiness and Acceptance Model, namely technological optimism, technological insecurity, and habitual use of GenAI, and a fourth related to the perceived need for regulation. The framework also incorporates three professional and sociodemographic characteristics: entrepreneurial status, gender, and membership in digital versus non-digital generations. Using a survey from Spain, we employ an analytical approach that combines ensemble tree–based machine learning models with Shapley Additive Explanations (SHAP) to enhance interpretability and explanatory insight. Among the evaluated models, XGBoost and Bayesian Additive Regression Trees achieve the strongest predictive performance. Random Forest exhibits comparatively weaker generalization performance. SHAP analyses consistently reveal that cognitive factors related to technology perceptions and habit in GenAI use play a central role in shaping evaluations of its contribution to sustainable economic development, whereas entrepreneurial status and generational cohort exert a secondary influence. Gender, however, shows greater explanatory relevance than perceptions regarding the need for regulation and reveals a pronounced gender gap in GenAI acceptance, with women exhibiting systematically less favorable perceptions than men. Although entrepreneurship exerts a relatively small effect, entrepreneurs tend to hold more favorable views regarding GenAI’s role in fostering sustainable economic development. Overall, the findings demonstrate that ensemble-based machine learning models, combined with SHAP, provide a powerful and transparent framework for explaining perceptions of GenAI’s contribution to SED.