AI-Powered Traditional Medicine Discovery: A Structured NarrativeReview of “AI”-Titled Studies

INTRODUCTION: Artificial intelligence is increasingly applied in Traditional Medicine research, yet whether studies explicitly self-labelled as "AI-based" substantively advance drug-discovery remains unclear. This review characterises how studies with "AI" in the article title apply artificial intelligence to Traditional Medicine drug-discovery and identifies methodological patterns, limitations, and emerging directions. METHODS: Five databases were searched for records containing artificial intelligence terminology in the title, retrieving over 2,000 records. After staged deduplication and structured screening against predefined criteria, twelve primary studies were included. Data were extracted across Traditional Medicine systems, computational techniques, discovery tasks, data sources, and validation strategies, and synthesised narratively due to heterogeneity. RESULTS: The included studies covered single herbs, classical multi-herb formulas, Traditional Medicine compound libraries, and safety-oriented modelling. Artificial intelligence methods ranged from supervised learning and deep learning to optimisation routines, molecular modelling, and network-based approaches. Most studies combined computational prediction with cell-based or animal-based validation, while only one linked algorithmic outputs to a clinical trial. Convergent biological signals were noted across metabolic, inflammatory, and endothelial pathways, and model performance frequently exceeded baseline statistical approaches. DISCUSSION: The included studies demonstrated stronger experimental grounding than most external AI-Traditional Medicine work, yet predictive sophistication consistently outpaced clinical validation. Inconsistent reporting standards and variable use of the term "AI" limit cross-study comparability and generalisability. CONCLUSION: AI-titled Traditional Medicine studies form a small but methodologically diverse evidence subset demonstrating repeated predictive success and moderate experimental support, yet rarely achieving clinical validation. Future progress requires clearer reporting, stronger external testing, and integrated pathways connecting computational outputs to reproducible biological and clinical evidence.