Implications of ChatGPT in Ag

MT. JULIET, Tenn. (DTN) -- Marc Arnusch first turned to ChatGPT for help evaluating a potential business opportunity.

It took him several hours of researching the correct types of prompts to get helpful responses from the artificial intelligence chatbot, but once he did, it gave him a very thorough look at not only the pros and cons of the deal, but also strategy on how to approach the negotiation.

"And holy smokes, what it told me was nothing short of brilliant," the Colorado farmer said. One of the biggest pieces of advice was to be direct and honest. "Sometimes in agriculture, we tend to want to hold some of our cards closer to our chest, and ChatGPT told me to forget that. Just get right to the point, and as odd as that sounded, it was the correct strategy."

ChatGPT is a kind of artificial intelligence called a large language model. It uses machine-learning algorithms to process and generate text that appears natural, like it was written by a human being. It's trained on billions of pieces of data, creating a neural-like network that enables it to identify patterns and relationships and then make predictions and generate responses.

The potential applications in agriculture are vast, but like most new technologies, the current version has limitations. It's only as good as the information it accesses. A good query gets a good response, and that's sometimes trickier than it seems. It cannot make decisions for you.

"I hope we get to the point where it becomes that predictive analytic tool that can really help us make decisions on our farm," Arnusch said. He's used it to analyze weeds in early stages of development, draft a marketing plan for his current crop and create a spreadsheet to analyze part of his business.

It has nuances, and there are things it just doesn't understand, but Arnusch thinks production agriculture needs to become familiar with this technology.

"You don't necessarily have to adopt it in your operation but know what it does. I think the cause-and-effect relationship of ChatGPT will impact a lot of farmers and ranchers, whether or not they've ever logged on. It's clear the agriculture industry is going to adopt this technology at certain levels."

GROUNDBREAKING TECH STILL HAS LIMITATIONS

While University of Delaware postdoctoral researcher Eugene Law primarily works with a different kind of artificial intelligence in his work with GROW, an integrated weed management research organization composed of USDA and university scientists, he's watched the development of large language models with interest. For more details on GROW's AI projects, please visit: https://growiwm.org/…

"What goes in to train an AI model heavily impacts what comes out the other side," he told DTN.

ChatGPT, for instance, was trained on the entire content of the internet prior to 2022. It's a self-supervised model, meaning it's been given minimal direction to what's correct or most important. It's constantly learning from user feedback, which could make it more or less coherent over time. Its responses don't cite sources, so it's hard to know whether it's getting information from a scientific study or a farmer's personal blog. It makes mistakes and occasionally makes things up.

"I just want farmers to be cautious. You might have this sense that it knows what it's talking about because of the way you're interacting with that information. I wouldn't necessarily treat it as a reputable source," Law said.

How you query, or prompt, programs like ChatGPT make a difference in the response you get. If you were to ask it how to address Palmer amaranth or another weed species, you might get a response that's logical. But just like you would if you were to insert that question into Google and comb through responses yourself, "You need to then use your own knowledge and understanding of who to trust to vet that information," Law said. "The context matters in agriculture."

Arnusch said that grammatically, ChatGPT does a good job, but it's up to the user to determine fact from fiction. What it does well is filter out a lot of the noise, distraction and emotion. "There's not a lot of emotion, and what it spits out is very black. It's very white, and it's very direct. And that's a blessing and curse all at the same time."

FARMER'S BUSINESS NETWORK LAUNCHES AI AGRONOMY ADVISER

Companies are taking this technology and building on it. Bloomberg is developing a large language model specifically for the financial industry. Coca Cola, Salesforce and Instacart are putting it into action.

Farmers Business Network launched a ChatGPT-based agronomic adviser, named Norm, on Tuesday. Norm is built on ChatGPT 3.5 but has additional "training."

Kit Barron, head of data science at FBN, said they've trained Norm to look at FBN's library of published blogs as well as its product warehouse before searching publicly available data such as weather insights, soil monitoring, application rates, product labels, current events, university research and grower commentary. During a demonstration of Norm's capabilities, the chat identified herbicides that work on specific weeds and provided a link to FBN's store. It also produced a list of 10 movies based on farming, listed factors to consider when determining the return on investment of owning a self-propelled sprayer versus hiring a custom applicator and wrote a poem about the virtues of glyphosate in iambic pentameter.

It's still a beta version, which means it's under construction and will produce errors. Barron said the development team at FBN has been working with its agronomy testers to reduce Norm's inaccurate responses and eliminate some of the quirks. He said feedback from farmers will make the program better.

Barron said it's unlike any other software because instead of telling it exactly what to do, it comes up with answers on its own.

"There's still an element of randomness that feels like magic," he told DTN. "We're learning how to influence that, and our plan is to add more novel data sources over time. But we really focused on our crop protection offerings for this first release."

Norm's knowledge of weather stops when the GPT dataset stopped in 2021, so he's not able to discuss forecasts, but Barron said they're working on a way to bring localized weather into the model.

Barron added that the only user data that Norm can access is the zip code of the FBN membership account. That helps prevent him from recommending an herbicide that can't be used in the FBN member's state, for example.

"While Norm will reference FBN reports that aggregate (farmer) data, he will never reference individual growers' data. You can imagine in the future, where if a farmer gives permission, Norm might be able to answer even more highly specific questions tailored to that very farm, but we're a long way off from that," Barron said. "I think we're just scratching the surface, and I'm really excited to see what our farmer members ask for next."

FBN members can access Norm at www.fbn.com/norm.

DECISION-SUPPORT TOOLS RIPE FOR AI INTEGRATION

Law sees an opportunity for ChatGPT-driven applications in the decision-support arena, particularly if it can analyze large amounts of scientific, local and grower-specific data.

"Farmers are great at what they do, but as soon as you start to add layers to the management, it can become more difficult to get them to adopt certain practices," he said. Many of the AI solutions in the market or in development are aimed at simplifying management. "We want to counteract that complexity by having it be aided by a computer."

Barron said farmers have to be experts in everything from crop genetics to tractor mechanics, grain marketing to soil biology. He thinks AI tools can help compile all the related information around agriculture and make it easily accessible. He hopes FBN's tool helps farmers gain familiarity with chat AI technology so that it doesn't feel so alien.

"My personal belief is that this technology is here to stay. The genie is out of the bottle," he said. "There are so many different applications for this that I think we're going to see a big difference between the businesses that choose to adopt tools like this and the ones that don't down the road."

Arnusch said artificial intelligence has been and will continue to creep into our daily lives a little bit more with each passing day, whether it's through ChatGPT or other means.

"If we don't take an interest in ChatGPT, it's going to take an interest in us," he said, referring to the way it learns from its users. "We also need to recognize what it's going to do in society. One of the things I've noticed is that sometimes ChatGPT doesn't necessarily paint agriculture in the correct light. So, it makes you wonder how it's being influenced, what it's drawing from, who's influencing it, and how do we make sure that our story is still factual, correct, and represents our industry without these AI bots doing it for us?"

If you'd like to try ChatGPT for yourself, you can find it at https://openai.com/….

Katie Dehlinger can be reached at katie.dehlinger@dtn.com

Follow her on Twitter at @KatieD_DTN

(c) Copyright 2023 DTN, LLC. All rights reserved.