Farmers in Southern Africa face numerous challenges, including water scarcity, climate change, fluctuating commodity prices, and pest management. To tackle these issues, the Southern African Agricultural Initiative (SAAI) has introduced AI Farmer, an innovative AI platform designed specifically for family farmers. This world-first, WhatsApp-based platform leverages vast amounts of data to provide timely answers to farmers’ most pressing questions. CEO Francois Rossouw said in an interview with Biznews that SAAI recognised the potential of AI technology after advancements in large language models, understanding its value in agriculture. With over 13,000 topics available, AI Farmer has already made a significant impact by addressing issues like a water contamination crisis in an Eastern Cape goat herd and helping vegetable farmers in Rustenburg access new fertilizers and seeds. The platform even identified a recent outbreak of foot-and-mouth disease through increased inquiries, alerting authorities before government agencies were notified. In collaboration with local partner Vectormind, AI Farmer is available in 83 languages to bridge the knowledge gap for small farmers. Rossouw cites the example of a demonstration in Malawi where the app provided a perfect answer in Chechewa. With too few experts available to reach small farmers, he emphasises that AI offers the potential to extend knowledge to approximately 570 million smallholders who cultivate about 24% of the world’s agricultural land.
Sign up for your early morning brew of the BizNews Insider to keep you up to speed with the content that matters. The newsletter will land in your inbox at 5:30am weekdays. Register here.
Watch here
Listen here
Highlights from the interview
In a recent interview, Francia Rossouw, CEO of the Southern African Agricultural Initiative (SAAI), discussed the launch of their AI platform, AI Farmer, which aims to support family farmers across Southern Africa. Rousseau explained that the platform utilises generative AI to provide farmers with immediate access to crucial agricultural information, addressing the need for efficient agricultural extension services. This technology is particularly beneficial for smallholders, who often lack institutional support.
Initially, SAAI experimented with a WhatsApp integration over ChatGPT, learning that generic information was inadequate for localized farming issues. They pivoted to train the AI using over 13,000 South African-specific agricultural data points, ensuring tailored advice. The platform emphasizes traceability and ethical AI practices, allowing users to understand and trust the information provided.
Rossouw shared success stories, including a rural farmer who identified and resolved livestock health issues using the app, and vegetable farmers who accessed new product information to improve crop yields. The platform, which supports 83 languages including South Africa’s 11 official languages, is now in use by over 4,500 farmers, with plans for additional features such as market prices and voice activation.
Rossouw highlighted the transformative potential of AI in overcoming barriers in agricultural extension services, particularly in remote areas where expert advice is often inaccessible. The initiative seeks to build trust among farmers by ensuring reliable, localised, and actionable information.
Extended transcript of the interview
Linda van Tilburg (00:04.828)
Joining us today is Francia Rousseau, the CEO of the Southern African Agricultural Initiative, also known as SAAI. This organisation ithat s leveraging the power of artificial intelligence or AI to support family farmers across Southern Africa. They’ve recently launched an AI platform called AI Farmer, which is designed to provide farmers with comprehensive answers within just minutes.
Linda van Tilburg (00:41.768)
Can tell us more about this AI platform that you have to advise farmers?
Francois Rossouw (00:48.078)
We started looking into this technology and using it right away after we had this large transformation in large language models called transformers, and we knew that this technology, especially generative AI, would have a lot of applications in agriculture. Our main focus was with regard to agricultural extension services. Agricultural extension programmes are all about bringing the latest scientific discoveries and farming technologies directly to farmers. The goal here is to boost efficiency and sustainability to help farms operate sustainably, which not only increases profit but also strengthens food security and farmers’ lives. For this, you need good agricultural information to keep up to date with the latest technology, training, and research
It’s not just in Africa; it’s throughout the world. Even in developed countries, they do not have the institutional capacity to drive agricultural extension services. If we look at the scale of what agricultural extension services can bring to farms, they can aid some 570 million smallholders who cultivate about 24% of the world’s agricultural land. So, this is an extremely important function that we need to bring back and scale to assist farmers. With this technology, we knew that we could help these farmers, especially with the vast amounts of information. We’re talking about the latest legislation, regulations issued by governments, product information like pesticides, access to tractor manuals, and recommendations and analytics. All of this is now available to farmers 24/7. It can be the best expert in its category.
So, that’s where we started. Initially, we had a WhatsApp layer over ChatGPT, and we realised there were many pitfalls and we learned a lot from there. Just the fact that information was too generic. We never knew where the documents or the educational content was coming from, and we soon realised that a farmer in Canada cannot use the same information as a farmer in South Africa, and the same for different provinces.
Francois Rossouw (03:34.504)
So, we started training the AI on various information sources such as high school agricultural papers, research documents, and so on, and then we built it out to about 13,000 subcategories of agricultural information, specifically South African agricultural information. We’ve been monitoring and tweaking this model.
I think there are a couple of things that make this agricultural extension services model different, as well as the technology it is based on. We have partnered with a local company called Vectormind. They’re South African-based, and all of the programmers are from South Africa as well. What makes this so different is that you’ve got your own document management platform, which means that you as the user or organisation control the flow of information. It doesn’t break out and go into ChatGPT or Claude or any of these other large language models. This means complete traceability, and with complete traceability, together with the whole AI ethical framework that we’ve built, we can see how the system answers, why it answers, what it answers, and then go and fix or replace the information. This document management platform is a huge win for us, and it also gives us great insights. It’s currently available on the App Store for both iPhone and on Google Play Store. We are giving it to family farmers in Southern Africa to try and test, and we have some really wonderful use cases that indicate we are on the right track.
Linda van Tilburg (05:21.532)
It sounds amazing. You don’t kind of think of technology and farming going together, especially for small farmers. They don’t use a lot of tech. So, what kind of success stories you share?
Francois Rossouw (05:35.276)
We told farmers, please download the app and talk to it. We want to see how you interact with it, and that’s also how we learned that you can’t just use generic information. You have to see how farmers interact. In the beginning, farmers didn’t ask questions; they made statements to it. I think also the big thing is that people needed to realise it’s a conversation; it’s not a search engine. Some of the app functionalities we are also building in include suggested questions so that farmers can easily engage in a conversation and explore specific use cases.
One rural farmer I am thinking about is from a goat project we had in Eastern Cape, where there were a lot of problems with a specific herd of goats, and they couldn’t figure out what the issue was. The problem with us as an organisation is that we’re very small. We simply don’t have the capacity to send an agricultural extension officer or a veterinarian 1,800 kilometres to check what is wrong, especially if we have six or seven development projects across South Africa. So, we asked them to download the app and see if they could have some success with the advice provided. One of the recommendations was that the water was dirty and that there was some kind of contamination leading to these deaths. They replaced the JoJo tank that was there and installed a water filtering system, and it immediately stopped the deaths.
Another instance involved a project we have in Rustenburg with vegetable farmers. They gained access to new fertilizers and seeds—products that they hadn’t used before but which were sponsored. Again, these farmers didn’t have someone on the ground with them to assist and to provide advice on dosage, usage, and timings. We asked them to provide all the product manuals of those items, we loaded them into the document management system, and they were able to easily access the information and talk to the product labels.
Francois Rossouw (07:55.036)
Just imagine this: you are a farmer anywhere in the world, and you have access to the largest agricultural library that exists. I’m not just talking about product information. We collaborate with large universities, research institutions, and private companies that fund research. Usually, that research is published only in English or Afrikaans or in whatever language is relevant to its location. However, large language models enable us to, in our case, test and converse with all of this research and product information in 83 different languages; this is at the core of what large language models can achieve.
Another recent story involved a farmer who conducted a soil analysis for his macadamia orchard in George. He sent the soil analysis away, and after two weeks, he grew impatient and input the results into our AI, asking, “This is my soil analysis; tell me what is lacking.” The AI gave him a recommendation, he implemented it, and a week later, when he received the soil analysis back, it was essentially similar. This is due to the AI being trained on high-quality educational data that enables it to provide accurate answers. Use cases like these give us real hope, especially regarding agricultural extension.
Linda van Tilburg (09:18.662)
So, what do farmers say to you after they’ve used the app, are they really keen on continuing this journey with AI?
Francois Rossouw (09:25.821)
Yes, I think there’s a bit of disbelief as well, because although we marketed it as an agricultural extension product in South Africa, our farmers are very reliant on what the legislation and regulations stipulate. From the start, yes, we’ve got this AI agricultural neural network built into the app, and there are 13,000 subdivisions making it a real agricultural expert, but I’ve also plugged in a substantial legal component with every piece of agricultural legislation, especially criminal law. I believe that farmers just need to trust what the technology can do, as they can observe the conversation flow and see their engagement with it. However, I think there’s still a bit of scepticism about what is on the other side.
Because it’s AI and is driven by the information it was fed and not by what it scavenged from the internet, it is actually more reliable than any other person you’d consult or a Google search because we genuinely try to segment these different sectors. So yes, I believe user reliability will improve as they feel comfortable with it. But I also think we have a significant role to play in user education, especially in how to prompt and communicate effectively with the AI to receive the answers they need.
Linda van Tilburg (11:07.268)
What’s the uptake been like in terms of numbers because you’ve been going, what was it, since February last year?
Francois Rossouw (11:15.223)
Yes, we recently relaunched. As I mentioned, we’ve gone through several iterations. We only launched the latest platform and made it truly accessible to any family farmer. We’ve just surpassed about four and a half thousand consistent users. The reason we haven’t formally launched before now was due to ensuring stickiness; it’s vital that when someone engages with the app, they return and continue using it.
We have a lot of functionality that we need to add to the system, but we are at a stage where we believe that the information within AI Farmer or the South African platform is robust enough to address any question. We also encourage farmers to send us other documents they think would be valuable to them, which then go through a thorough document verification process, as the ethics of AI are paramount to us. We believe that by the end of this year or perhaps next year, every person will be able to deploy an AI instance on their phone or laptop. The value comes from the information, and if that information is not verified,
Credibility is extremely important to us, which is why we’ve taken the time to develop AI Farmer to its current state. We have several wonderful functions that we will be adding, such as market prices for all the different types of agricultural subsectors in South Africa. The weather function is already live, and we’re introducing voice activation. Soon, you will be able to get photo and video analyses, by early next year. So, we may have been patient, but we need to ensure that we drive stickiness so farmers trust and feel safe using the product.
Linda van Tilburg (13:14.75)
You talked about that it is available in so many languages. Is it available in the 11 languages of South Africa?
Francois Rossouw (13:22.372)
Francois Rossouw (13:22.372) Yes, and more. I must tell you, in the languages we’ve tested, I conducted a demonstration in Malawi recently, and the lady who asked a question in Chichewa received a perfect answer. While we were demonstrating in Germany, a question was asked in Hebrew; although I cannot verify its accuracy, the individual who asked it stated that it appeared correct to him. This again comes back to trust, and to establish that trust, one needs to understand what large language models can do.
Another aspect we’re working to convey to farmers is that we have invested considerable time in eliminating issues like hallucinations within the large language model, and in preventing the drifting of the data; it is something we check for daily. The AI undergoes a built-in check of questions and answers each morning, as we cannot afford to have problems in our system.
Linda van Tilburg (14:22.526)
Is there anything else you would like to tell us about this system and platform?
Francois Rossouw (14:27.384)
I think it’s just important, from an agricultural extension perspective, to emphasise why this technology is so transformative. We currently face seven impediments that hamper agricultural extension services. The first is limited access to knowledge; many smallholder farmers don’t have access to up-to-date information, and they often encounter structural impediments, needing to travel far to get to a university or expert. AI can provide farmers access in real time.
But now with AI, they have an institution backing the information that is populated within the AI framework. It’s not necessary to attend a training session on specific topics where you know the instance was trained on that. I just don’t know how farmers can farm during the day and do research at night to answer their pressing questions or sift through various research documents to tackle specific issues they face. Again, this is what makes AI so powerful; you pose your problem to the library of information, and it processes that information to provide an answer. It doesn’t serve you some abstract result to decipher, as the farmer who is not an expert in that field.
This capacity to streamline access to knowledge makes AI exceptionally powerful. Additionally, as I previously mentioned, institutional capacity is a significant issue. Most governments, both in South Africa and globally, lack the resources to fund and maintain agricultural extension officers who can consistently visit remote rural areas to build relationships with farmers.
Francois Rossouw (16:47.379)
The shortage of field experts who can reach farmers is another issue. Those who train often do not possess the specific topic knowledge that farmers require. Furthermore, from a government perspective, the monitoring and supervision of extension officers are always problematic. Lastly, there are accessibility issues; some farmers are favoured over others. Larger farmers often gain preferential treatment, as extension agents tend to consider them more relevant. These are just some of the challenges we are aiming to address in the agricultural extension space, and we believe that AI can play a vital role here. It’s nice to have this technology, if you can’t apply it to a specific problem, it’s not going to be sustainable.
Read also:
