The National Agriculture and Food Research Organization (NARO) has developed a generative AI specialized in agricultural knowledge and will begin an experimental study in Mie Prefecture on October 21, 2024. This is the first generative AI designed specifically for the agricultural sector. A key feature is its Continual Pre-Training, which incorporates not only publicly available data from the internet but also data gathered from agricultural organizations across Japan, including cultivation techniques from various production sites and NARO's research findings. We plan to deploy this generative AI nationwide and continuously improve its performance through field feedback. By training new farmers early and providing advanced technologies to existing farmers, this initiative will contribute to the sustainable development of agriculture and help foster the next generation of farmers.
Overview
In June 2024, the Basic Act on Food, Agriculture, and Rural Areas was revised for the first time in 25 years. In an increasingly unstable global situation, there is a strong need to advance smart agriculture to ensure a stable food supply. Meanwhile, the average age of the farmers in Japan is 68.7 years old (as of 2023), and it is predicted that the number of farmers will decrease to 300,000 over the next 20 years. Therefore, it is crucial to rapidly train new farmers and provide the latest agricultural techniques to existing farmers.
Currently, the use of generative AI is being explored across various sectors, but general-purpose generative AI often gives incorrect answers to questions that require domain-specific knowledge. To address this issue, we incorporated not only widely available agricultural information on the Internet but also specialized, non-public information, such as research data accumulated by NARO, cultivation manuals, calendars, and farming instruction records owned by public research institutes of local governments and Japan Agricultural cooperatives. The AI was trained on a vast amount of detailed, Japan-specific data, including the distinct characteristics of each crop variety, different cultivation methods tailored to the soil and weather conditions in each region, and farmers' meticulous attention to consumer needs. As a result, it was shown that the accuracy rate of this generative AI was 40% higher for questions about specialized agricultural knowledge than general-purpose generative AI. This marks the first case of a generative AI specialized in agriculture in Japan.
An experimental study of this generative AI for strawberry cultivation will begin in Mie Prefecture on October 21, 2024. The generative AI will be combined with a chat application and provided to agricultural extension officers in the region. NARO also plans to develop generative AI for other crops, to reduce the survey time of extension officers in their offices by 30%, allowing them to provide more advanced guidance to farmers.
NARO will continue to play a central role in expanding collaboration with the national extension and guidance centers, public research institutes, agricultural corporations, startups, and other stakeholders, to gather agricultural data and deploy generative AI to transform the future of agriculture. Through these efforts, we will collect information from extension officers and farmers on the accuracy and usability of the AI responses, allowing us to further enhance its performance. This will enable the farmers, including new ones, to easily access technical guidance using the generative AI in any situation, and create an environment where advanced guidance can be received through extension officers. Ultimately, this initiative contributes to the sustainable development of agriculture, including fostering the next generation of farmers.
For Inquiries
Contact: https://www.naro.go.jp/english/inquiry/index.html
