Advances in Agronomy: Chapter One - The role of artificial intelligence in crop improvement

The growing global demands for agricultural goods will require accelerated crop improvement. High-throughput genomic, phenomic, enviromic and other multi-omic data collection methods have largely satisfied data acquisition bottlenecks that previously existed within crop breeding and management. Fully capitalizing on large, high-dimensional datasets has now evolved as a new challenge. Artificial intelligence (AI) is currently the foremost solution. Types of AI with the capacity to learn (machine learning), such as neural networks, can better facilitate the translation of data into useful predictions by bypassing the limitations of human expert-driven learning. The potential for applying AI to major crop improvement methods has already been demonstrated with preliminary successes shown using deep learning for genomic selection, feature selection for enviromics, ensembles and knowledge-based AI for crop growth modeling, computer vision and convolutional neural networks for phenomics, and unsupervised machine learning for multi-omics. Other types of neural networks including transformer, recurrent, encoding decoding, and generative networks as well as symbolic (non-learning) AI such as robotic process automation, expert systems, and inductive logic programming are also reviewed to contextualize the rapidly changing AI field. Overall, AI has shown strong potential to leverage data for a variety of crop improvement tasks.