Cotton Genomics and Genetics 2025, Vol.16, No.3, 148-162 http://cropscipublisher.com/index.php/cgg 160 confidence is low. These are all the work being promoted in the current application of AI breeding. In terms of promotion and application, it is also important to train breeders to master basic data analysis and model interpretation skills. Only when breeders understand the model can they better use it to guide practical work such as hybrid combination design and generation advancement. China has already taken some actions in this regard, such as organizing national cotton breeding backbones to participate in the "Digital Seed Industry and Intelligent Breeding" training course to share AI breeding cases and experiences. It can be foreseen that in the next few years, artificial intelligence breeding tools will be gradually implemented in front-line breeding units and continuously improved based on feedback. At that time, the model will no longer be a mysterious black box, but will become a daily auxiliary tool similar to soil testing and disease diagnosis, and will be skillfully used by breeders. 7 Concluding Remarks Artificial intelligence technology is gradually being integrated into cotton breeding research and has achieved initial results. Predictive breeding models such as genomic selection have made useful attempts to improve cotton yield, fiber quality and stress resistance traits: in terms of yield, high-yield genotypes are predicted through whole genome markers, which improves the selection efficiency of early breeding generations; in terms of fiber quality, the GS model achieves high-precision prediction of indicators such as length and strength, promoting the selection of high-quality new lines; in terms of stress resistance, the prediction model combined with machine learning successfully identified multi-environmentally stable disease-resistant QTLs, accelerating the screening of disease-resistant varieties. Internationally, Australia's CSIRO took the lead in integrating genomic prediction into the breeding process, significantly shortening the breeding cycle and cultivating high-quality and high-yield new varieties; the US public breeding department systematically evaluated the effects and limitations of GS, laying the foundation for further promotion and application. Domestic scientific research institutions have also actively deployed intelligent breeding research, developed a cotton intelligent breeding platform, and used AI technology to crack the genetic mechanism of some complex cotton traits. It can be said that artificial intelligence is helping breeders break through the bottleneck of traditional breeding and realize the transformation of breeding decisions from experience-driven to data-driven. Although the application of AI-assisted breeding in cotton is still in its infancy, the existing results have proved its great potential and bright prospects. Looking to the future, the deep integration of artificial intelligence and genome prediction will lead cotton breeding into a new era. On the one hand, with the advancement of sequencing and phenotyping technologies, breeding will obtain exponentially growing multidimensional data to provide fuel for AI models. Whole genome selection will be combined with new technologies such as gene editing and epigenetic regulation to form integrated innovation in breeding technology. Intelligent algorithms will be able to handle more complex breeding goals, such as improving yield, quality and multi-resistance at the same time, and realizing the optimal design of comprehensive traits. On the other hand, the emergence of a new generation of AI models (such as neural networks, generative AI, etc.) is expected to further improve the accuracy and breadth of breeding predictions. In the future, cotton breeding decisions may be generated by AI with countless breeding schemes, taking into account gene combinations, environmental adaptability and market demand, and selecting the best scheme for breeders' reference. The breeding cycle will also be greatly shortened due to assisted generation prediction. In theory, a breeder is expected to experience a complete iteration of multiple breeding cycles in his career, which was unimaginable in the past. Of course, we also need to realize that on the road to "breeding 5.0", there are still many scientific problems to be solved, such as how to accurately simulate the impact of gene interactions on traits, how to integrate evolution and niche theory in the model, etc. These require in-depth cross-disciplinary and cooperation between genetics and artificial intelligence. But what is certain is that artificial intelligence will serve as a powerful new engine to drive cotton breeding forward and contribute to the safety of textile raw materials and sustainable agriculture. In order to better integrate artificial intelligence technology into cotton breeding practice, we put forward the following suggestions: First, establish a standardized cotton breeding big data system. Including unified phenotypic measurement specifications, building a national joint breeding database, and improving the genotype information sharing platform, etc., to provide high-quality training data for AI models. Secondly, strengthen the
RkJQdWJsaXNoZXIy MjQ4ODYzNA==