Saudi Arabia is positioning mining as a strategic non-oil pillar under Vision 2030, yet mineral beneficiation plants remain exposed to ore variability, high comminution energy, water constraints, reagent inefficiency, and inconsistent recovery. This review paper examines how artificial intelligence can optimize beneficiation processes in Saudi phosphate, gold, bauxite, copper, zinc, and emerging critical-mineral operations. The study synthesizes recent literature, sector reports, and industrial digitalization evidence from 2020 to 2025 to develop a process-oriented framework for applying machine learning, computer vision, digital twins, model predictive control, and explainable analytics across crushing, grinding, classification, flotation, leaching, dewatering, and plant-wide decision support. The review argues that AI should not be treated as a stand-alone automation layer; it must be governed as a socio-technical capability connected to metallurgical expertise, data quality, cybersecurity, operator trust, and sustainability targets. The paper contributes a Saudi-specific implementation roadmap and identifies research gaps in geometallurgical data integration, low-water flotation control, Arabic-enabled operator interfaces, MLOps governance, and validated economic impact measurement.