Artificial intelligence (AI) is transforming lithium-ion battery (LIB) research by linking heterogeneous data, predictive modeling, mechanistic interpretation, and experimental validation within a closed-loop paradigm. AI does not uniquely enable nonlinear representation; rather, it complements electrochemical, statistical, and control models by learning high-dimensional relationships among composition, structure, processing, interfacial chemistry, operating history, and performance. This review critically examines three connected domains: AI-assisted materials discovery, battery state prediction, and trustworthy intelligent management. To distinguish this contribution from recent topic-specific surveys, we organize methods along four complementary axes—model architecture, learning strategy, physics integration, and deployment strategy—and compare landmark studies using quantitative evidence such as independent cell count, validation design, reported error, experimental budget, and closed-loop acceleration. Materials applications include cathodes, anodes, liquid and solid electrolytes, and electrode-electrolyte interphases; operational applications include state of charge, state of health, remaining useful life, degradation diagnosis, safety warning, and digital twin-enabled management. The critical synthesis identifies data leakage, inconsistent metadata, domain shift, uncalibrated uncertainty, black-box reasoning, computational constraints, and limited external validation as recurring barriers. Future progress will depend less on model complexity alone than on FAIR data, cell-independent evaluation, physics-informed learning, uncertainty-aware decisions, autonomous experimentation, human oversight, and deployment-aware design. These elements define an evidence chain for moving LIB AI from proof-of-concept prediction toward reproducible, closed-loop, and trustworthy battery innovation.