Alifosina Mtseteka’s family has been farming for generations in central Malawi, but she recently ran into a pest infestation that had her stumped.
The okra she grows along with beans and corn on her tiny farm was wilting due to a lack of water. It was then invaded by a colony of aphids, threatening the crop her family depends on for food.
Mtseteka watered and watched the plants, tried the standard fixes passed down through generations, and asked neighbors for tips, but nothing worked. When she heard her daughters talking about a place with all the answers, she asked to go.
They brought her to a man who recited her problem into his phone in the local language, Chichewa.
“‘Ask it and it will answer what to do,’” she remembers her daughters telling her.
They were right.
The voice on the phone suggested an insecticide. She got it, and it worked.
“I bought the chemicals and sprayed them, and the okra became healthy again,” Mtseteka, who is older than 60 but doesn’t know her exact age, told Rest of World.
Mtseteka lives with her ailing husband and two grandchildren in Chisemphere village, about an hour’s drive from the capital Lilongwe. It is a cluster of small, mud-brick houses surrounded by fields.
She has never owned a smartphone or a computer, but her crop was saved by an artificial intelligence bot designed to help subsistence farmers like herself. It’s called Ulangizi AI.
“Ulangizi” means advice in Chichewa. It was developed by Opportunity International, a U.K. nonprofit, to answer farmers’ questions through text and voice, as well as diagnose problems found in farmers’ photos of crops and livestock. It has been trained on government guidelines and traditional know-how, and can be delivered through WhatsApp.
“We work directly with farmers to understand what they like and don’t like, making the tool as user-friendly and welcoming as possible,” Paul Essene, head of Opportunity International’s digital innovation team, told Rest of World.
-
Opportunity International’s Richard Chongo oversees the Ulangizi AI program in Malawi. -
A field support agent queries Ulangizi AI on how best to grow mustard greens. Thoko Chikondi for Rest of World
The chatbot’s verbal interaction with farmers shows how far the technology has reached and how quickly it is seeping into every aspect of work. It can be adapted to tackle long-standing issues in emerging markets due to its speed and flexibility. When it works, it can be particularly powerful in places where budgets are small and problems are complex.
In many developing markets, traditional tools often move at the pace of government paperwork and donor cycles. AI, by contrast, can be spun up in days on a laptop or a cloud server, delivering real-time insights and tools without the need for extensive new infrastructure.
A small team armed with open-source models and local data can now build solutions once reserved for well-funded labs in the West — automating crop forecasts, diagnosing diseases from phone cameras, or translating public-health messages into dozens of regional languages overnight.
AI can be tweaked for new regions, new dialects, or novel uses at low marginal cost. This flexibility could make it a tool kit for solving some problems in less affluent countries.
Hugging the shores of Lake Malawi in southeastern Africa, Malawi is a small, landlocked nation. It has a population of about 20 million, most of them farmers. Its gross domestic product per capita is less than $1,000, and more than 70% of the population lives below the poverty line by some measures.
Traditionally, farmers in Malawi relied on informal, peer-to-peer learning, primarily through word-of-mouth advice passed down through generations and shared among fellow farmers.
The government has attempted to promote and disseminate best practices through radio shows and a national network of extension officers who travel around rural regions advising farmers on how to manage crops, improve yields and lower costs.
Farmers have had to wait for weeks, sometimes months, to receive advice from extension officers: There is approximately one extension officer available per 3,000 farmers, according to the Ministry of Agriculture.
Extension officer Rachel Nyirenda goes from village to village educating the 1,743 farmers she oversees. She is lucky if she gets to see any given group of farmers more than once a month, she told Rest of World.
When she arrives in a village, they bombard her with questions. Sometimes she has had to crack open a big agricultural manual to look things up, she said.
“They could really see me open the book to find the information they asked for,” said Nyirenda.
Opportunity International developed Ulangizi AI as an instant oracle for Malawi farmers who are struggling with the impact of climate change, which has increased the occurrence of droughts and floods.
The hope is that easy access to the best farming tips will help lift more farmers above the poverty line, enabling them to do more than just subsistence farming.
The tool combines ChatGPT and the most relevant large language models with government agricultural guidelines and traditional knowledge, delivering the information conversationally in a local language.
Training the AI was difficult. Chichewa is a so-called low-resource language, meaning there is less of it online than other languages, making it more challenging for AI to master.
Early versions of Ulangizi sounded strange, farmers said. It sometimes seemed to speak only in a woman’s voice and at one point had an Indian accent. The developers persevered and figured it out, and Ulangizi is now used by thousands of farmers, according to Opportunity International.
The company didn’t just drop its new app on an app store. It trained the most tech-savvy farmers, enabling them to act as middlemen. The use of these informal support agents makes it easier to introduce the technology to communities that are not comfortable with smartphones or even keyboards, said Essene.
“The most effective driver of usage is on-the-ground training, using a ‘human in the loop’ model,” he said.
Kingsley Jasi grows corn and beans on his 8,000-square-meter (2 acres) farm in Chiradzulu. He recently saw evidence that worms were eating his corn. His go-to insecticide ended up damaging his crops, he told Rest of World.
Jasi snapped a photo and sent it to Ulangizi via WhatsApp, and it came back with an alternative insecticide right away.
“After following its recommendation, the worms were completely eliminated,” he said. “Since then, I have relied solely on the chatbot’s guidance.”
Jasi was able to buy a phone preloaded with the app through GiveDirectly, a charity that provides farmers with basic incomes and helps them find the tools they need to escape poverty. It has helped more than 200,000 people in Malawi, most recently assisting 2,800 farmers in obtaining phones with the AI app.
“The phone becomes a gateway: to money, to information, to connection,” Martin Kalima, a senior manager for external relations at GiveDirectly in Lilongwe, told Rest of World. “This [Ulangizi AI] pilot is about learning what that access makes possible.”
Like many AI tools, Ulangizi can sometimes generate false answers and get confused by new questions. It has been trained to handle the basics reliably, but sometimes makes suggestions that the Ministry of Agriculture does not approve, government officials said.
Once, farmers tried to ask the AI about a flowering plant called witch hazel. It refused to answer, saying it would not discuss witchcraft. The AI had to be taught that it was an agricultural term.
It also needs to be updated regularly with details of what is working on the ground.
“The Ulangizi is as good as the information that is contained in the chatbot,” Emmanuel Kaunda, vice chancellor of Lilongwe University of Agriculture and Natural Resources, told Rest of World. “Information has to be updated now and again for it to be meaningful. The diseases we had last year will not be the same as this year because of climate change issues. It means scientists must quickly adapt for the technology to be useful.”
Still, the farmers who use the app have found it so impressive and all-knowing that they even ask it to try to see into the future.
“If it could warn us ahead of time, whether the rains will come late, be excessive, or if dry spells are expected, we would be able to plan better,” Misha Katema, one of the farmers Opportunity International has trained as a field support agent, told Rest of World.