Estonian farms move AI from spreadsheets toward robots and drones

Artificial intelligence is mainly helping farmers handle data in day-to-day work, though autonomous agricultural robots and drones are no longer treated as science fiction.

Airika Verrev, crop and livestock production manager and consultant at Pajo Ranch, says the tools have improved markedly in the short time she has used them. This spring the food producer tried to compile data in Excel with AI and found the technology was not ready. By grain harvest that had changed. "These days, I no longer create a single Excel spreadsheet myself. I tell AI what I want and what it should look like, and that's it," she said. She also uses it to analyse data and locate relevant legal provisions, while stressing that its answers must be checked constantly.

Marten Madissoo, assistant professor of agricultural engineering at the Estonian University of Life Sciences, said the technology does not remove farmers' duty to judge those answers. "We have to bear in mind that AI is simply a useful tool that allows us to make decisions faster, but we still need to be experts ourselves," he said.

He sees substantial scope for machine learning and AI in logistics, in organising storage and processing of farm products, and in analysing fields. There is no single solution, he said, because useful applications depend on the problem. One route is to make more of the data farmers already gather for field records, registers and reports. "We have the advantage of having data spanning several years. By bringing this data together and developing a model to analyze it, we can achieve those gains," Madissoo said. He added that agricultural and technical expertise have to be combined: people who know the fields must decide which indicators matter and what they want to learn from the data.

Chatbots, he said, are only the tip of the iceberg. Madissoo coordinates the AI and Robotics Center AIRE at the Estonian University of Life Sciences, which looks at how AI and robotics can improve the competitiveness of industrial companies. As an industrial example he cited an AIRE demonstration at meat producer Nõo Lihatööstus, which tested AI quality control of packaged goods on a conveyor. The system was meant to spot poorly attached labels, wrong dates or defects in appearance.

Researchers at the University of Tartu have used AI to identify field islands in aerial photographs and to sort grain. The Estonian University of Life Sciences has also experimented with autonomous agricultural machinery.

Further drone uses may follow soon. The Riigikogu is considering an amendment to the Plant Protection Act that would allow plant protection products to be sprayed from unmanned aircraft under certain conditions. Madissoo said several autonomous systems already exist, but legal limits and safety requirements slow deployment. The university's field robot, for example, remains firmly in testing and still has to be taught field boundaries and where it may operate. "It's not exactly science fiction, but it will still take some time before we can put all these technologies and capabilities to use," he said. He considers repetitive work with clearly defined procedures especially suitable for automation, because machine learning can make it more efficient.

Verrev expects combining data with AI to make farming more precise. Aerial and drone photographs could help decide how much fertiliser or plant protection product to order for the next season and avoid unnecessary spending. "The savings could actually be quite substantial. Right now, though, we're still estimating how much to order," she said. Many producers, she added, still use AI mainly as a more advanced search engine, and a lack of time makes it hard to follow new developments.

Madissoo said companies adopt new technology when the benefit is clear and the investment is justified. "How much do I need to invest and how quickly will that investment pay for itself?" he said, describing the question producers face. A first trial need not be costly: start with a chatbot and a specific question, analyse existing data, and test the advice on a smaller field. For more complex ideas he suggests support from universities or AIRE.

He recalled a conference comparison: horses still exist, but they no longer matter in transport, and the same logic applies to AI. "You can manage without it, but you'll simply fall behind the mainstream. So I encourage people to use it: Trust, but verify," Madissoo said.