Artificial intelligence in African economics: a systematic review of performance and ethical risks

Abstract Objective This systematic review maps how artificial intelligence (AI)—specifically machine learning (ML) sub-fields—are employed within the economics and finance sectors of sub-Saharan Africa. It identifies dominant research trends, key application areas, performance benchmarks for models, and the significant socio-economic implications of these technologies in the region. Method A systematic literature search following PRISMA guidelines was carried out across Scopus, Web of Science, and Google Scholar for publications from 2000 to 2025. The analysis is based on a final dataset of 103 scholarly articles. We adopt a PRISMA-AI extension that applies three specific filters: the Model Reproducibility Score (MRS), Local Data Ratio (LDR), and Ethical Disclosure Index (EDI). Results Research activity has surged since 2017, with deep-learning models becoming pivotal. Key applications include poverty mapping using satellite imagery, which explains more than 70% of wealth variation (CI 95%: 64%–76%), and AI-driven credit scoring and fraud detection that achieve F1 scores above 0.80 (range: 0.72–0.94), thereby enhancing financial inclusion. The review highlights a shift from survey-based data to multimodal data fusion, especially after 2017. East Africa—specifically Kenya and Rwanda—is identified as a primary AI research hub. Originality This review provides the first meta-performance comparison and leaderboard for AI models across principal economic tasks in sub-Saharan Africa (SSA). It introduces a novel AI Readiness Index (ARI) for ranking country-level investment potential and proposes a structured ethical risk assessment framework, quantifying the prevalence of challenges such as algorithmic bias (68% of studies) and data sovereignty (42%).