Using artificial intelligence in agriculture went from a quick way to obtain farming advice to a costly experience for a Chinese farmer after an incorrect recommendation from an AI application destroyed his sesame crop, which covered nearly 25 acres, equivalent to about 101 dunams.
The incident highlights the risks of relying on AI models for sensitive field decisions, where an answer that appears convincing on a smartphone can result in significant financial losses when applied on a large scale without proper verification or testing.
Growing Confidence in AI Recommendations
According to media reports, the 67-year-old farmer, identified as Wu, had been using an AI application to obtain advice on farming, pest control, and weed management.
After benefiting from previous recommendations, he gradually developed confidence in the tool and eventually turned to it for help dealing with weeds and pests affecting his sesame crop.
However, the recommendation that appeared suitable on his phone contained a fundamental error. It was applied across a large area of farmland without first consulting an agricultural specialist or testing the treatment on a small section.
A Weed Killer Turns Into a Crop Disaster
The application recommended a mixture of chemicals to control weeds and pests. Reports on the incident said the mixture included a herbicide designed to target broadleaf plants.
That was the critical mistake. Sesame is itself a broadleaf plant, meaning that a herbicide intended to eliminate broadleaf vegetation can also damage or kill sesame plants instead of protecting the crop.
After the mixture was sprayed, the sesame plants began to wilt and die. By the following day, extensive damage had spread across the field.
The losses reportedly affected around 25 acres, or approximately 150 mu under the Chinese agricultural measurement system. One report estimated the financial loss at around 150,000 yuan, equivalent to approximately $20,700.
Why Can AI Give the Wrong Agricultural Advice?
The problem is partly rooted in the way large language models work. They are primarily designed to understand language and generate coherent, plausible responses rather than independently make specialized agricultural decisions in the field.
An AI system can therefore produce an answer that sounds highly confident while missing a local condition or technical requirement that could completely change the appropriate recommendation.
A study published in Frontiers in Plant Science in July 2024 highlighted several challenges associated with using large language models for plant protection, including their limited ability to adapt to local agricultural conditions and the difficulty of using one general-purpose model across different crops, environments, and tasks.
Agricultural recommendations require consideration of multiple factors, including the crop type, target pest or weed, local conditions, plant growth stage, registered pesticides, chemical concentration, and application methods.
As a result, an answer that is appropriate in one situation is not necessarily suitable for another field. Soil conditions, weather, plant development, chemical concentration, crop variety, and local regulations can all change the correct course of action.
Experts Warn Against Relying on AI for Pesticide Decisions
The Chinese farmer’s experience is not the only example highlighting the risks of using AI to manage weeds and pests. Cornell University has warned against relying unconditionally on AI-generated pesticide recommendations.
The university has noted that some AI systems can provide inaccurate information or recommend pesticides that are not legally registered for use on a particular crop.
Applying a pesticide contrary to its label instructions can damage crops and the environment and may also create potential health risks for users and people in the surrounding area.
For this reason, experts recommend checking the pesticide label and consulting qualified agricultural professionals before acting on AI-generated recommendations involving chemical products.
AI Can Help Agriculture — But It Is Not a Substitute for Expertise
The incident does not mean that artificial intelligence has no place in agriculture. On the contrary, AI is increasingly being used for monitoring plant health, detecting pests and diseases, analyzing images, estimating irrigation requirements, and identifying agricultural stress.
However, safer and more reliable systems can be developed by combining specialized agricultural models with trusted databases and external sources rather than relying solely on a general-purpose language model.
Researchers at the Chinese Academy of Agricultural Sciences are exploring approaches that can make AI systems more specialized by training them on agricultural data and connecting them to structured knowledge sources. Such methods could produce recommendations that are more closely adapted to local conditions and easier to verify.
Other research has also demonstrated the potential of agricultural decision-support systems based on information retrieval. These systems can take into account factors such as plant growth stage, environmental conditions, and the specific context in which a recommendation is requested, rather than simply generating a generic response.
A Lesson That Extends Beyond Agriculture
The Chinese farmer’s experience illustrates a broader problem with artificial intelligence: a convincing answer is not necessarily a correct or actionable one.
The more an AI recommendation moves from general information to a real-world decision where an error could cause financial, health, or environmental damage, the more important it becomes to verify the information and consult qualified professionals.
In agriculture, the difference between a safe recommendation and a harmful one can sometimes depend on a small technical detail. But when that recommendation is applied across a large field, that seemingly minor mistake can result in the loss of an entire crop.
