Why General-Purpose AI Falls Short in Agriculture
GenAI knows farming. It doesn't know your field.
In July, a farmer in Chuzhou, China, turned to an AI chatbot for help regarding weed and pest control in his fields. The resulting plan, as described by a Futurism journalist, was a deadly cocktail of agrochemicals, leading to the destruction of all his sesame seedlings and a significant financial loss.
The chatbot admitted its mistakes; the recommended treatment was unsuitable for sesame plants and advised widespread spraying, which is not an optimal approach.
AI in Agriculture: A Missed Opportunity
Researchers from multiple countries tested whether generative AI could replace human agricultural advisors, concluding that it cannot.
They found that AI models like OpenAI's GPT-3.5 and GPT-4 provided ambiguous advice when asked about treating the fall armyworm, a global corn crop pest. When simulated questions from Nigerian cassava farmers were posed, ChatGPT recommended incorrect herbicides and application timings, posing a threat to food security.
Increasing Missteps and Growing Reliance on AI
With rising costs and a shortage of trained agricultural extensionists, many farmers are turning to digital assistants, particularly general-purpose AI chatbots like ChatGPT and Claude. These tools, while knowledgeable, are not tailored for specific agricultural needs. They provide answers that sound professional but lack practicality or even cause harm.
Training AI for Agronomic Purposes
The solution is to train AI models specifically for agronomy. This involves two steps:
- Data Collection: The model learns from real-world data, including satellite images, past harvests, current weather, and the crop's growth stage.
- Expert Guidance: It learns from working agronomists, who teach it to respond appropriately in various situations, ensuring recommendations are safe and effective.