Empirical Advances in Applied Artificial Intelligence Across Domains: A Cross-Sectoral Review of Methods, Evaluation Practices, and Emerging Trends

Abstract

Over the past three years, artificial intelligence (AI) and machine learning (ML) methods have diffused far beyond their original strongholds in computer science, reshaping empirical practice in medicine, finance, cybersecurity, marketing, law, logistics, and the social sciences. This review synthesizes 227 recent peer-reviewed and conference-presented studies spanning thirteen thematic domains to characterize how applied AI research is currently conducted, evaluated, and reported. Rather than proposing a single new algorithm, the review takes a horizontal, cross-sectoral view: it asks what methodological patterns recur across otherwise unrelated application areas, and what these patterns reveal about the maturation of empirical AI research as a discipline. Three convergent themes emerge. First, comparative and benchmarking methodology, rather than single-model demonstration, has become the dominant evaluation paradigm, with a large share of the reviewed literature explicitly framed as head-to-head comparisons of algorithms, feature engineering strategies, or prompting techniques. Second, privacy-preserving computation, particularly differential privacy and federated learning, has moved from a specialized subfield into a general-purpose design constraint applied across healthcare, finance, advertising, and government document processing. Third, explainability and fairness auditing are increasingly treated not as optional add-ons but as integral components of model evaluation, especially in credit scoring, medical diagnosis, and risk management. The review closes with a discussion of the limitations inherent to a title-level cross-sectoral synthesis and offers directions for future domain-specific and methodological research.