A joint research team from Shanghai University has published a review systematically examining the core challenges and future opportunities for applying machine learning in metal additive manufacturing. The study reveals that the complex relationships among process parameters, microstructure, and mechanical properties in metal additive manufacturing provide a powerful framework for machine learning to address quality control issues, yet data scarcity remains the primary bottleneck limiting its application. The article discusses various data generation, sampling, and fusion techniques, including high-throughput experiments and simulations, active learning, and multi-fidelity fusion, with a focus on the interpretability challenge—the trade-offs among data volume, prediction accuracy, and physical consistency. By categorizing domain knowledge integration methods, it provides a systematic guide to enhancing model reliability. Finally, the article proposes a development roadmap emphasizing the collaborative evolution of knowledge graph-driven RAG agents, high-fidelity digital twins, and embodied intelligence, aiming to achieve autonomous manufacturing through self-optimizing closed-loop systems.