Use and application of artificial intelligence in public policy evaluation. A scoping review

[EN] Context: Despite growing interest in using AI to improve public policies and services, its application in evaluation lacks systematised evidence and scientific publications on the implications of these technologies in evaluation practice.

Objective: To map, through available literature, the current and emerging state of AI application in public policy evaluation. Methods: We conducted the study through an exploratory literature review, or scoping review, following the methodological framework of Levac et al. (2010). The synthesis was carried out with a thematic analysis of 27 studies and analytical-theoretical literature.

Results: AI is increasingly being applied in various phases of the evaluation cycle, primarily as a support tool ( human-in-the-loop ), especially in the operationalisation, report preparation and results dissemination phases. Use cases include the analysis of large volumes of administrative and textual data through Machine Learning (ML) and Natural Language Processing (NLP), the performance of simulations and counterfactual analyses, the potential for real-time monitoring, and the use of Large Language Models (LLMs) for synthesis or visualisation tasks, among others.

Conclusions and implications: Current evidence points toward the predominance of human-machine collaboration models (human-in-the-loop), indicating that realising the benefits of AI in this field does not involve total automation, but rather strategic, critically reflective and contextually adapted implementation.