AI poised to reshape farming from breeding to harvesting

Artificial intelligence is poised to reshape farming from breeding to harvesting, experts said on Wednesday, as a leading agricultural university seeks to train a new generation of AI-literate graduates.

Chen Wei, president of China Agricultural University, said that the university plans to set up a school of artificial intelligence. The move aims to meet the urgent demand for talent in the high-tech industries of the future.

"Agricultural science is a highly technology-intensive and knowledge-rich discipline," he said during the pre-conference meetings of the 2026 World Agrifood Innovation Conference, which is scheduled to be held in Beijing from Thursday to Saturday.

"Farming still relies on conventional machinery for sowing, fertilizing, and harvesting. That is set to change. Driverless machinery is on the way, drones are already a common sight in the fields, and fully automated farms could be next.

"Farming will ultimately become industrialized, and it will need large numbers of people who have not only specialist knowledge but also interdisciplinary expertise," he said.

Sun Qixin, an academician at the Chinese Academy of Engineering, said during the meetings that AI is already moving from pilot projects into the mainstream in parts of Chinese agriculture, but more data and patience are needed to build large models for agriculture.

"AI in agriculture is no longer just a dream. Some of it has become a reality," Sun said.

"If one day breeding shifts from relying on individual breeders to relying on intelligent agents that match or surpass the best human breeders, we will see that as the breakthrough that opens the door to practical application," he said.

Currently, breeders have to select one or two elite individuals from tens of thousands of candidates — a slow process. In the future, AI can handle much of the work, sharply cutting the time and investment needed to breed better varieties than those available today, he added.

"AI models can link crops' field phenotypes with their genotypes. If such models are well-trained, breeding work could become dozens or even hundreds of times more efficient," he said.

But available training data and knowledge bases are still insufficient. Efforts have been made. For example, many breeding stations now use aerial drones and ground-based multispectral sensors to track a plant's entire life cycle, from emergence to harvest, he added.

"Within three to five years, data on major crops, livestock and poultry could be sufficient to develop an intelligent breeding model," he said.

Zhang Xinyi contributed to this story.