Today's Edition
ISIDORA CIRIC (ABU DHABI)
In many remote parts of the world, one agricultural adviser is responsible for assisting 5,000 to 10,000 farmers. As climate change brings unfamiliar diseases, shifting crop patterns and new pests, farmers are facing tougher questions - often without anyone to turn to for answers.
To tackle that, CGIAR is building a chatbot that knows how to talk about pests, crop diseases and drought - not in theory, but in ways that can help real farmers in the middle of a growing season. The tool, AgriLLM, is one of the first AI models fine-tuned for smallholder agriculture in low and middle-income countries.
"We already have the science," said Khuloud Odeh, Chief Digital Transformation Officer at CGIAR, in an interview with Aletihad.
"The challenge is how fast we can get it to the people who need it - and in a format that actually works for them."
AgriLLM, short for Agricultural Large Language Model, is among the projects selected for a $200 million climate-tech fund announced at COP28 in Dubai. The fund, launched by the UAE and the Gates Foundation, is designed to support AI and advanced technology solutions that strengthen food systems under pressure from climate disruption. A pilot project was launched at COP29 in Azerbaijan.
The chatbot is being trained to respond to practical on-the-ground questions, especially around crop diseases, pest outbreaks and other challenges linked to climate shifts. Farmers will be able to interact with the tool via text and voice, in local languages, with the aim of making it accessible in areas with limited connectivity and digital literacy.
"We want to build an interface that allows farmers to interact with it in a very seamless, user-friendly way. Hopefully, it will include voice integration, so they can ask questions in their local dialect while working in the field and get answers in a timely manner," Odeh explained.
AgriLLM is doing that by adapting Falcon, an open-source AI model developed by the UAE's Technology Innovation Institute and Odeh's team is fine-tuning Falcon using CGIAR's own datasets, alongside publicly available information from FAO, the World Bank, the Gates Foundation, and others.
"We're starting with an initial corpus of data and fine-tuning Falcon at different sizes. That's the beauty of the model. It gives us both lightweight and larger options, so we can adapt it to different needs," she said.
Unlike conventional AI applications, this one will also be publicly shareable. Odeh added that the team plans to publish model weights so others in the agriculture and AI communities can build on their work instead of duplicating it.
"And we realised that if assigning weight to agricultural information and evidence improves outcomes for farmers, then we need to share those weights. We don't want people to reinvent the wheel."
Validation is another major part of the process. The team intends to benchmark the tool's responses against certified advisers in the field, and its results will be measured against current tools being used in the field, some of which rely on generic AI. The chatbot will also be piloted with farmers directly, to test how well it performs in actual use, Odeh added.
While the technology is central, the aim remains grounded in people: improving access to trusted advice for farmers who would otherwise go without it.
"Our goal, really, is to reach 500 million farmers - or even more - because this is just an indicative number. We hope that we don't leave any farmer behind, especially in remote and vulnerable climate areas."
