FAIR Data Standards for AI in Plant Biology: Current Practice and Case Studies

Abstract Artificial Intelligence (AI) has become a central analytical approach in plant science, supporting prediction, pattern discovery, and integration across molecular, phenotypic, and environmental data. While methodological advances have been rapid, progress is increasingly constrained less by algorithms than by the structure, semantics, and interoperability of the underlying data. In this review, we examine how the Findable, Accessible, Interoperable, and Reusable (FAIR) principles are currently realized in AI-driven plant research and how different implementations shape which AI analyses are feasible in practice. We introduce a readiness level perspective that treats FAIR not as a binary property, but as a spectrum of operational implementations ranging from unstructured data to fully machine-actionable FAIR Digital Objects. Applying this framework to molecular, phenotypic, and integrative use cases, we show that FAIR readiness is strongly method and context-dependent. Task-focused AI applications can succeed at lower readiness levels, whereas integrative, transferable, and reproducible analyses consistently require richer metadata, explicit semantics, and traceable provenance. We also note that many advances depend on shared infrastructure and community-maintained practices that allow curation and reuse to accumulate across projects. We argue for a shift from FAIR compliance toward FAIR capability to enable scalable and sustainable AI-driven discovery in plant science.