Revolutionizing Extension Models with Artificial Intelligence in Service of Smallholder Farmers
Today, smallholder farmers find themselves at the center of several complex global issues and extension services must adapt to meet their evolving needs. Traditional extension services focused exclusively on increasing agricultural productivity, whereas the extension services of today must additionally assist farmers in meeting multiple societal demands from rural development to food security to carbon sequestration.
Artificial intelligence (AI) has the potential to improve access to agricultural extension services and bring quality advice to the most geographically and socially marginalized farmers. Models such as ChatGPT, Bard and Claude have already mobilized to pilot some of the world’s first AI-assisted agricultural extension models, such as agri1 and Farmer.chat.
Acceso is now piloting its own AI-assisted extension tool called ExtensioBot to help extension agents in Latin America and the Caribbean better serve the farmers in their service area.
The decision to branch into AI came from user demand
Acceso was already communicating with over 30,000 farmers through its AgriTech platform Extensio, sending messages with alerts about pest outbreaks, weather predictions and lessons on good agricultural practices, among others. Farmers often replied to these messages asking follow-up questions, a service the software was not designed to provide.
At the same time, Acceso’s extension agents were struggling to meet the demand for their services.
Extension agents are struggling everywhere
The extension agent-to-farmer ratio is a key indicator to assess a nation's capacity to provide adequate extension services. While this ratio varies widely — from 1:6,804 in Nigeria to 1:800 in Mexico to 1:252 in Japan — most countries have a long way to meet the recommended ratio of 1:200 or 1:500. Unfortunately, the investment needed to close the gap by increasing the number of extension agents is cost prohibitive for most countries.
Equipping extension agents with the necessary tools
We can invest to hire and train hundreds of thousands more extension agents, or we can equip existing agents with powerful tools capable of expanding their reach and fortifying their knowledge.
Generative AI and language learning models (LLMs) have the unique ability to quickly and cost-effectively identify relevant information, distill key points and present them in accessible language, as well as assess a farmer’s specific challenges and generate data-driven, personalized recommendations.
Acceso’s ExtensioBot has the potential to directly address farmer questions and offload work from extension agents. This allows each agent to deliver their services more efficiently by strategically combining AI and in-person services, thereby expanding their reach to a greater number of farmers.
Acceso has no intention of fully replacing in-person extension agents. Extension agents are critical to achieving Acceso’s impact as the quality of advice and attention they provide farmers is paramount to a smallholder farmer’s ability to connect sustainably and successfully to formal markets. Acceso’s objective is to further refine and train its ExtensioBot to enable in-person agents to serve additional farmers more effectively and efficiently.
ExtensioBot is Acceso’s answer to the demand for innovation in extension service delivery models
Acceso is piloting an in-house AI tool, ExtensioBot, to answer farmers’ questions and provide personalized advice through text and audio using Azure AI Speech built on ChatGPT 4.0. In addition, ExtensioBot uses Azure AI Vision’s image recognition technology to accurately identify pests and plant diseases shared by farmers.
ExtensioBot is still learning and is limited to answering questions on four topics: agriculture, fish farming, weather and phases of the moon. Answers for the first two categories draw from information widely available on the internet. For weather-related questions, Acceso has partnered with a meteorological organization to provide farmers with parcel-specific insights. Phases of the moon is a topic that farmers and fishers regularly requested because moon phases are an important variable in their decision-making. For this category, ExtensioBot draws information from Acceso’s internal knowledge base, a combination of agronomic data and traditional ecological knowledge from the various regions that Acceso serves.
This is the first step in a steep collective learning curve
ExtensioBot has confronted several common issues: 1) responses can be too generic or not practical for smallholders with limited resources and 2) the lack of representation of Indigenous and minority languages on the internet poses a significant barrier to achieving its full potential. Collaborative efforts will be essential to address the technology’s current limitations and achieve meaningful impact for smallholder farmers.
Acceso will continue to prioritize engagement with ExtensioBot users, namely smallholder farmers and fishers, while also investing in further training of the model to provide location-specific information in languages other than Spanish, especially Nahuatl, Maya and Quique. In the long term, developing more decentralized and open-source AI models will be necessary to ensure equitable access to these technologies and the capability to train AI models to serve specific subgroups of the population.
The potential of this technology cannot be understated, and Acceso looks forward to playing a role in revolutionizing extension models in service of smallholder farmers around the world. The stakes are high, and farmers need all the support they can get to meet the challenges of tomorrow.
