Extension Notebook: AI in agriculture

Illinois horseradish growers have a weed problem: there are relatively few herbicides labeled for the crop. So researchers began teaching an AI system to tell horseradish plants apart from everything else growing around them, so equipment could target only the unwanted plants. That kind of project is exactly why artificial intelligence, or AI, has become one of the most talked-about new technologies in agriculture. But what does AI actually mean for farmers?

At its simplest, AI allows computers and machines to learn from data, recognize patterns, solve problems and make decisions. In agriculture, that can mean using information from sensors, drones, satellites, machinery, weather and soil data to help farmers make better decisions or allow equipment to perform tasks with less human intervention.

AI is not a single technology. Farmers are already encountering several forms of it.

Predictive AI has been used for years to forecast yields, disease outbreaks and other potential outcomes.

Machine-embedded AI is being built directly into equipment, such as sprayers, tractors and combines.

Generative AI tools can answer questions, analyze images and create reports.

Agentic AI, a newer concept, goes a step further by allowing AI to complete a series of tasks using multiple tools — for example, checking weather, soil conditions and equipment readiness together rather than one at a time.

Some of these applications are already moving from research into practical agriculture.

Take that horseradish project mentioned earlier. The system doesn’t necessarily need to identify every weed by name. It simply needs to recognize what is horseradish and what isn’t, allowing it to target unwanted plants — a narrower, more achievable task than full weed identification.

AI can also work with drones. One system uses drone imagery to map a farmstead and identify individual trees, buildings, water and other features. The AI can then help create an application map and flight route so a spray drone can autonomously treat each tree.

Generative AI offers another set of possibilities. For example, a farmer can upload a photograph of a soybean field and ask an AI system what might be wrong. The system can examine symptoms such as yellow leaves and dry soil and suggest possibilities including nutrient deficiency, drought stress or pest damage. The important word here is possibilities. AI can help organize information and point us toward potential answers, but it should not replace local expertise or field verification.

That is one of the most important lessons when using AI in agriculture: Don’t assume an AI answer is correct simply because it sounds convincing.

That is one of the most important lessons when using AI in agriculture: Don’t assume an AI answer is correct simply because it sounds convincing.

Generative AI can sometimes “hallucinate,” or produce information that isn’t accurate. AI systems can also lack current information, and algorithms developed under one set of conditions may not perform well somewhere else. An Illinois nitrogen-management experiment provided an early reminder of this. An algorithm developed for Oklahoma did not perform adequately under Illinois conditions. It reduced nitrogen use, but the resulting yield loss, 33 bushels per acre, far outweighed any savings on fertilizer.

So where is AI headed?

I expect some of the biggest changes will come from AI becoming increasingly integrated into farm equipment and farm-management systems. Instead of simply providing information, AI will increasingly help machines determine what to do in the field.

Generative AI will also become a more useful decision-support tool, helping farmers sort through the tremendous amount of agricultural information available today. And agentic AI may eventually allow farmers to give a system a goal — such as getting everything ready for planting — and have it check weather, soil conditions, inventories, equipment readiness and work assignments.

We are still in the early stages of this technology. Farmers don’t need to become computer scientists to benefit from AI. They do need to understand what these tools can — and cannot — do, identify real problems where AI might provide value, and verify important information before making costly management decisions.

The question isn’t whether AI will become part of agriculture. It is already happening. The more important question is how we can use it responsibly to make better decisions and solve real-world problems. Try it first on lower-risk work, sorting field notes or drafting a scouting report, before letting it weigh in on a fertility or spray decision.

N. Dennis Bowman is a University of Illinois Extension specialist — digital agriculture.