A recent study highlights significant concerns about the reliability of ChatGPT, particularly versions 3.5 and 4.0, in providing agricultural advice. Dr. Asaf Tzachor, founder of the Aviram Sustainability and Climate Program at Reichman University, along with an international team of researchers, scrutinized the chatbot's guidance to farmers in Africa. They found that the inaccuracies in its advice could lead to serious agricultural missteps and potential food crises.
In their article for Nature Food, the researchers caution against the unmediated use of generative AI models in agriculture. They emphasize the risk of farmers implementing flawed recommendations, which could trigger widespread crop failures. Instead, they advocate for a more rigorous development process for AI models in agriculture, including thorough monitoring and testing before widespread deployment.
Shortly after ChatGPT's launch in early 2023, Dr. Tzachor assembled a team from agricultural research centers in Nigeria, Kenya, Colombia, France, England, and the United States. They observed that farmers in developing countries had started using the AI model for professional advice on agronomy and botany. These farmers, from small to medium-sized farms across equatorial Africa, Southeast Asia, and South America, have internet access and utilize the OpenAI interface.
The research aimed to assess whether ChatGPT could replace agricultural extension agents—professionals who traditionally provide training and consultancy to farmers. These agents, known as "extensionists," include agronomists, botanists, plant disease experts, and advisors on various agricultural practices.
"Extensionists are vital in disseminating advanced agricultural knowledge and guiding small farmers in sustainable crop intensification," explains Dr. Tzachor. "They conduct seminars on new herbicides and pesticides, provide irrigation and fertilization strategies, plan field experiments, and advise on marketing and export strategies."
Globally, about 570 million small and medium-sized farms require training in various agricultural fields. However, extensionists face significant challenges, especially in developing countries, due to language barriers, poor infrastructure, and outdated communication networks. These obstacles make it difficult for agricultural consultants to reach remote farms and for farmers to attend professional seminars.
Given these challenges, the research team explored whether an AI model could fill the gap in agricultural advisory services. However, their findings revealed significant flaws in the chatbot's advice.
For instance, when tasked with recommending control measures for the fall armyworm—a pest that causes extensive damage to corn crops—ChatGPT provided ambiguous advice on pesticide use. Similarly, for cassava root farmers in Nigeria, ChatGPT incorrectly advised on the timing of herbicide application, which could lead to crop damage and food crises.
"The issue extends beyond algorithmic errors," says Dr. Tzachor. "The fundamental problem is the lack of safeguards against the widespread use of Large Language Models in sensitive systems like agriculture. There is no oversight, no evaluation of context-specific suitability, no accountability for incorrect use, and no responsibility for the consequences of flawed recommendations."
Dr. Tzachor emphasizes that while the allure of AI is clear, the risks are substantial. "We are dealing with food security and farm management, not composing songs or screenplays. The consequences of inaccuracies can be dire for vulnerable populations."
In response, the researchers propose an idealized development and deployment process for AI models in agriculture. This includes thorough testing, monitoring, and safeguards to ensure accurate and reliable advice.
Dr. Tzachor, also Acting Dean of the School of Sustainability at Reichman University, concludes, "We see the potential for AI in agriculture, but the current usage involves errors that farming cannot tolerate. The issue of liability and ensuring safe use remains largely unexamined. Addressing these concerns is crucial for protecting the food security of vulnerable populations."
More: https://phys.org/news/2024-05-gpt-inaccuracies-agriculture-crop-losses.html
