A group of LSU students have developed a large language model that will simplify the process of herbicide application for farmers.
Completed this semester as part of the HNRS 3035 course, the LLM tool aggregates herbicide labels, federal and state regulations and the hundreds page long LSU AgCenter research booklet to give farmers simplified and custom information.
“Ideally for this model you’ll be able to put in what you’re growing, where you’re at, what weeds you’re trying to get into and it will be able to coalesce that information and quickly and, importantly, accurately give advice for how to get rid of different weeds and things,” said mechanical engineering senior Colin Raby, one of the students that worked on the project.
The tool not only saves time but can potentially prevent the overuse of herbicides, a practice sometimes motivated by risk of lower yields, according to Raby.
“Instead of having to spend a lot of money to work with a farming consultant, or do the research that could take days, weeks, months, comparing everything that’s available online, the hope is this tool can save that time and energy.”
Other project members agreed that the accuracy of the tool was vital to its success and applicability.
“You’re recommending something that could kill somebody’s crop and hurt their livelihood,” said computer science junior Cody Flurry. “You have to make sure you are giving great information.”
The tool is programmed to report and ask for any insufficient information. This is to avoid the LLM giving inaccurate recommendations with confidence, also known as “hallucinations.” Sourcing of the recommendation is also programmed, so any discrepancy can be caught.
“If this says seven milligrams instead of six milligrams, that could mean thousands of dollars and an entire crop,” Raby said. “Ensuring the accuracy of this system and ensuring our model doesn’t hallucinate is something we’ve been working on.”
Further upgrades to the LLM are planned, including the ability to use the tool with just a photo of the weed.
Expected to release in the spring, the LLM is one of many new devices set to incorporate AI into agriculture.
“In a field like agriculture, a larger portion of the research ongoing does include some sort of AI component that has enhanced the ability for these researchers to do their job,” said research coordinator for the LSU AgCenter Brayden Blanchard.
AI machine learning has been used for years to make predictions with data, Blanchard said. However, its recent popularization has made its utilization easier for agricultural research through things like greater stakeholder funding.
“The way I’ve seen it applied in improved estimates and data driven, knowledge-based decisions, if that type of use can complement any industry, I think it’s an excellent thing,” Blanchard said.
Dr. Thanos Gentimis teaches EXST 7005, a class covering machine learning, data analysis and image processing in agriculture.
“What I see is agriculture turning into this fully automated process where everything is handled by some sort of smart machine,” Gentimis said in an interview with LSU President William F. Tate IV.
AI will eventually help us optimize water consumption and fertilizer application, Gentimis said in the interview.
“These technologies have the potential to be applied in more ways than we can think of going forward,” Raby said. “If you want something that can both, put you ahead career wise and that’s really interesting and can improve society, I would highly recommend people get involved in this atmosphere and learn as much about it as possible.”
