Abstract
Artificial intelligence (AI) is revolutionizing polymer engineering by shifting materials development from traditional trial-and-error to integrated, data-driven, and autonomous workflows. This review summarizes recent progress in AI-driven polymer science, including structural representation, generative molecular design, property prediction, retrosynthetic planning, and closed-loop experimental processes. We discuss how polymer-specific encodings like BigSMILES and graph-based models enable machine-readable macromolecular structures, supporting transformer models, graph neural networks, and uncertainty-aware predictors that accurately estimate thermal, mechanical, dielectric, and functional properties. Beyond prediction, we examine generative architectures such as variational autoencoders, adversarial models, diffusion models, and reinforcement learning that enable inverse polymer design by exploring vast chemical spaces under performance constraints. Combining these computational approaches with techniques like Bayesian optimization, active learning, and autonomous labs signifies a shift toward closed-loop polymer engineering, where AI continuously suggests, synthesizes, analyzes, and refines materials with minimal human intervention. Key challenges, including data scarcity, polymer-specific representation learning, model interpretability, infrastructure integration, and sustainability goals, are thoroughly considered. Emerging solutions such as FAIR data ecosystems, physics-informed neural networks, explainable AI, and modular self-driving laboratories are evaluated for scalable implementation. Overall, these advancements point toward autonomous polymer discovery, with AI serving not only as a predictive tool but as the foundation for intelligent, adaptive, and sustainable polymer engineering.