<p>Diatoms are an exceptionally diverse lineage of unicellular photosynthetic microalgae, primarily identified by valve morphology. Their silica frustules and high productivity make them central to nutrient cycling, climate regulation, and applications ranging from environmental monitoring to biotechnology. However, conventional morphological analysis is time-consuming, subjective, and often limited in taxonomic resolution. Over the past 2 decades, artificial intelligence has progressed from early feature engineering and basic machine learning to advanced deep learning approaches for diatom detection, classification, ecological prediction, and bioengineering optimization. Despite the rapid expansion, there are no comprehensive overviews spanning these domains. Existing reviews typically focus on isolated themes such as automated taxonomy, water quality assessment, or forensic applications, without integrating methodological evolution and technological progress across fields. This paper presents the first broad synthesis of AI applications in diatom research to our knowledge, tracing developments from early pattern recognition to contemporary multimodal, predictive, and systems-oriented frameworks. A structured table-based analysis documents datasets, algorithms, evaluation metrics, and classification performance over time. The results highlight a clear trajectory from engineered feature models to convolutional neural networks, transformers, and multimodal architectures, with reported accuracies frequently exceeding 90 to 98 percent. At the same time, persistent challenges remain, including class imbalance, limited representativeness of field training data, model explainability, uncertainty quantification, and regional transferability. Overall, this study shows that AI has evolved from a supportive tool to a foundational discipline within diatom science and outlines priorities for biologically informed model design, data standardization, and rigorous validation.</p>