Advancing biological taxonomy in the AI era: deep learning applications, challenges, and future directions

Biological taxonomy faces an inflection point. In this review, we trace its progress through three technology-driven eras-morphology, molecular, and today's emerging artificial intelligence (AI)-driven stage-and discuss how each successive toolkit has expanded rather than replaced the last. This review elucidates the transformative impact of deep learning across four domains: biological image-based classification, bioacoustics-based classification, genetic sequence-based classification, and the elucidation of species traits. Foundation models that treat genomes as a "language" have begun to link sequence variation with protein structure, phenotype, and ecological niche, hinting at a more fundamental, data-driven basis for delimiting species. We highlight the recent breakthroughs in deep learning and foundation models and argue that fully integrated, causality-aware models could deliver a step-change in biological taxonomy. However, key challenges persist, spanning data quality, algorithmic robustness, reference-library completeness, model transparency, and shared standards. Taxonomists' deep knowledge of trait evolution gives them a unique role in the ongoing convergence of AI and biological taxonomy, particularly in guiding foundation-model development. As AI moves toward reasoning over complex biological causality, even core taxonomic concepts may evolve; recognizing and steering that transformation is both the challenge and the opportunity of this AI-driven era.