Why general-purpose AI falls short in agriculture
TechRadar

Why general-purpose AI falls short in agriculture

AI can transform farming, but without field-specific data, its advice risks costly mistakes.

GenAI knows farming. It doesn't know your field.

In July, a farmer in Chuzhou, China, turned to an AI chatbot for help. He needed a plan to control weeds and pests in his fields.

The model produced what a Futurism journalist later called β€œa Long Island iced tea of agrochemicals.” The farmer sprayed this cocktail across 25 acres of crops. By the next day, every single sesame seedling was dead, costing him roughly 150,000 yuan, or about 18,000 euros.

When Wu asked the chatbot what went wrong, the AI admitted it made mistakes. That mix it proposed was suitable for broadleaf weeds. But sesame is widely classified as a broadleaf plant, so it was never the right option to begin with. Another error was suggesting spraying the entire field instead of using the targeted application.

As horrible as it is, that is just one example of AI missing the mark. Researchers from the U.S., the U.K., Kenya, Nigeria, and Colombia recently tested whether generative AI could replace human agricultural advisors. Apparently, it can’t. CPO & COO at OneSoil.

According to their paper published in Nature Food, the team first asked chatbots to recommend treatment for the fall armyworm, a pest that destroys corn crops worldwide. In response, OpenAI's GPT-3.5 and GPT-4 models provided highly ambiguous pesticide advice.

Next, the researchers simulated questions from cassava farmers in Nigeria. ChatGPT recommended specific herbicides but gave the wrong application timing. Following that advice would have killed the crops, threatening food security for millions of people, since Nigeria is Africa's most important cassava producer.

More frequent mistakes

What's more disturbing is that these mistakes might become more frequent. With rising labor, land, and equipment costs, many farmers can no longer afford professional advice.

On top of that, there is a severe shortage of trained agricultural extensionists. Extension workers rarely reach the world's most isolated fields, such as those in South America and Africa, while farmers in those regions cannot travel hundreds of kilometers to attend seminars to improve their knowledge. Instead, they often rely on extension agencies based in capital cities. But those lack both the funding and the staff to update their outdated guides or track new farming trends.

So farmers turn to digital assistants. According to a recent McKinsey survey, roughly 17 percent of farmers worldwide already use generative AI in day-to-day decisions. That is more than any other new technology in the sector, including robotics and electric-powered machinery, Bloomberg reported.

Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!

To be fair, AI itself is not the issue. The problem is that many farmers turn to general-purpose AI chatbots such as ChatGPT or Claude for advice. These tools know crop types, chemical classes, standard practices, and textbook planting calendars. But they are not trained specifically for agriculture purposes.

For example, they won't ask for necessary data from the user before providing advice. As a result, their answers can sound professional while being useless or even harmful.

Training AI for agronomic purposes

There is a way to fix it: train an AI model for agronomic purposes. That takes two steps.

  1. First, the model learns from real data about a particular field: satellite images, past harvests, current weather and the crop's stage of growth.
  2. Second, it learns from working agronomists, who teach it how to respond in different situations.

Those lessons also set limits on what the model will say. If it knows what crop is in the ground, it won't recommend a herbicide that is banned for that crop.

To better understand how this works, imagine a farmer who needs to spray a wheat field to get rid of weeds. A standard AI chatbot might draw on general textbooks and suggest a strong chemical. But because it cannot see the actual field, it does not know two critical things: that the wheat is at a fragile stage of growth where this chemical will ruin the crop, and that tomorrow's high winds will blow the spray onto neighboring land. Following the AI's advice could destroy the harvest.

By contrast, an AI connected to real data checks satellite photos and weather reports first. It automatically spots the high winds, recognizes that the wheat is too fragile for certain chemicals, and picks a safe alternative. It also asks questions when information is unclear or missing. Then it shows the farmer exactly where to spray, using coordinates or a task map that targets only the weed patches. In the end, this saves money and protects future yields.

Collecting information

Having said all that, expecting farmers to gather all this information and enter it into a generic AI tool isn't realistic. It takes time most growers don't have, as well as technical knowledge. Plus, farmers have no time to do inputs regularly.

This is why I believe a new market will emerge: of specialized agricultural AI assistants with built-in access to dashboards and satellite imagery, allowing seamless movement between data sources and making accurate, field-specific information far easier to access and act on.

This market is already taking shape. Some assistants can revisit satellite data layers, analyze them visually, and answer questions like β€œWhat has changed in the last 48 hours?” in plain language. Some products also let farmers send voice messages describing what they have applied. The agent files those details in the right parts of the system and keeps them as context. This makes it easier to keep field records up to date, which helps the model give more useful answers.

Such tools can help farmers make sense of their own field data, spot meaningful changes, and spend less time piecing together multiple sources. With some luck, a wave of these purpose-built tools will help eliminate costly mistakes in agriculture, so more farmers worldwide can thrive instead of losing their crops, money and spirit.

We've reviewed and ranked the best business cloud storage services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

CPO & COO at OneSoil.

You must confirm your public display name before commenting

Please logout and then login again, you will then be prompted to enter your display name.

Read on TechRadar ↗ ← Back to News

Comments

No comments yet. Start the discussion.