Artificial intelligence (AI) is one of the tech sector’s newest innovations and will likely impact every industry. Beyond just a buzzword, this technology may be on your farm soon.
Artificial intelligence is loosely defined and can mean anything from machine learning to large language model (LLM) generative AI. Machine learning is already used in agriculture, such as on John Deere’s See and Spray system, using cameras to identify weeds to spray and obstacles to avoid.
Chances are, you have already heard of generative AI, and can already use it on multiple different web platforms and devices — Microsoft’s Copilot, Google’s Gemini, and Facebook’s Meta AI all are available for consumer use. Generative AI can create a variety of text, images, and audio, and is in the very early stages of video generation.
There are many AI business- and consumer-grade applications, including those for the agriculture industry. Before diving into AI’s potential benefits, it’s necessary to understand how it works, and its limitations.
How It Works
As one of the first AI models to reach the market, OpenAI’s ChatGPT is at the forefront of this technology and is already integrated into many business’s day-to-day operations. Users can ask ChatGPT to write text, summarize and organize it, and even have a “conversation.”
ChatGPT and many other popular AI services run on LLM to generate text from a user’s input. It does this by “reading” existing text on the internet, and recognizes how words tend to appear in context with other words, sort of like a highly capable auto-complete function when typing on a smartphone. It uses this to predict the next most likely word that might appear in response to a request, and each word after that.
According to the company’s website, ChatGPT is in active development, using three information sources: written work publicly available on the internet, content licensed from third parties by OpenAI, and data provided by users or human trainers. This leads to limitations, however.
Since the service relies on the publicly available internet — which may not always be accurate — AI has “hallucinations.” When an AI model hallucinates, it makes up incorrect responses based on this faulty information, or comes up with an answer entirely on its own, and presents it as factual. AI models also are incapable of providing sources to its output, even when accurate. As the technology is still in early development, the ChatGPT website recommends verifying any generated responses and providing feedback.
Retrieval Augmented Generation (RAG) is one way companies are avoiding AI hallucinations, says Patrick Walther, founder of Molewa AI Studio. Rather than using the whole database, a RAG-based AI model is trained on private information the user supplies, which takes priority over other sources of information. This way a company has more confidence over AI-generated results, since those are based on their own human-made information.
Environmental Impact
AI models have shown to be more energy-intensive than the data retrieval, streaming, and communication a majority of data centers have used since the boom of the internet.
According to a study conducted by the Electric Power Research Institute (EPRI), a Google search requires 0.3 watt-hours of electricity to retrieve results. Asking a single question in ChatGPT requires 2.9 watt-hours to generate a response — roughly 10 times more. With 5.3 billion global internet users, the EPRI says this could potentially lead to a dramatic increase in energy requirements.
The International Energy Agency (IEA) estimates global energy consumption in the AI and cryptocurrency sectors will more than double 2022’s estimated 460 terawatt-hours (TWh) to over 1,000 TWh by 2026. This is roughly the same yearly energy consumed by the entire country of Japan.
Google’s total greenhouse gas emissions were 14.3 million metric tons of carbon dioxide equivalent (tCO2e) in 2023 — a 48% increase compared with 2019, and a 13% increase year-over-year. That was “primarily driven by increased data center energy consumption and supply chain emissions,” according to Google’s 2024 Environmental Report. Google is adjusting for AI’s high-impact energy use by developing a new Tensor Processing Unit, which they claim is over 67% more energy-efficient than the previous generation. The company also says it has identified tested practices that can reduce the energy required to train an AI model by up to 100 times, and reduce associated emissions by up to 1,000 times.
Generative AI is also water-intensive. To generate results, large language models use graphical processing units (GPUs) in large, densely packed data centers. The GPUs get very hot in the process and need to be water-cooled to avoid overheating. In 2023, Google’s data centers consumed 6.1 billion gallons of fresh water — 17% more than in 2022. At its current growth rate, global AI demand may need to withdraw 4.2 to 6.6 billion cubic meters of fresh water in 2027 — half the United Kingdom’s total annual water withdrawal, according to a study published at the University of California, Riverside.
How AI Is Being Used
In 2023, the Farmers Business Network (FBN) debuted Norm, an AI-powered ag adviser, built on OpenAI’s ChatGPT model.
“When OpenAI released GPT-3, we saw this as an amazing development in the technology sphere,” says Kit Barron, vice president of technology at FBN. “We immediately started asking ourselves internally, ‘How can we leverage this technology to support our mission and support our farmer members?’ We started thinking about all of the areas in which a farmer needs to be an expert to be successful today.”
Barron says farmers need to be experts in plant genetics and biology, chemistry and application rates, weather, fertility, grain trading, and marketing. Norm is a potential solution to simplify the approach to these processes and broaden the farmer’s knowledge. In early testing, FBN found that farmers were using Norm to ask about crop protection. With a retail storefront for crop protection and a database going back 10 years, this seemed the right place to focus development.
FBN’s team of agronomists builds out recommendations for which adjuvants pair well with different types of crop protection, and have used this data to fine-tune Norm’s responses. While ChatGPT learns through user inputs and data, FBN determined that its farmers’ data needed to stay private, asking users only to share a ZIP code to provide location-based responses. For individual prompts, the more context a farmer can provide with their question, the more relevant Norm’s responses will be.
Norm can access the wider data from OpenAI’s base model of ChatGPT but focuses on providing responses based on FBN’s proprietary data such as agronomist recommendations, product labels and manuals, and relevant articles.
Barron describes these early days of AI development as the “wild west” and has made it clear that Norm is still experimental. For the first few months of availability, Norm would often hallucinate, recommending articles that didn’t exist or providing incorrect information. To prevent this, FBN has put up guardrails, having Norm decline to comment on a topic outside its agricultural knowledge base. Since then, responses have become less random and more accurate.
Norm’s current iteration is only for farmers who apply or blend their own crop protection. However, FBN consistently adds new tools, such as its new fertility model, to further assist farmers. Norm is also embedded into FBN’s online shopping experience, allowing farmers to ask questions or choose one of the pre-prompted questions to learn more about the products as they’re browsing. FBN also is looking into bringing grain markets and grain trading tools to Norm.
The Norm AI Ag Advisor is available for FBN members at fbn.com/norm.
Walther is focused on recommending AI-powered solutions to help agribusinesses enhance productivity, sustainability, and profitability. He works with clients that include crop protection firms, seed manufacturers, energy companies, and agronomists to help find opportunities to streamline their workflows with AI. When talking to an agronomist, for example, Walther breaks down their entire workflow into steps, from soil sampling to making recommendations to the farmer.
“It’s a lot of manual data entry processes that [the agronomist] has to do from step to step,” Walther says. “I started asking, ‘If we automate this, what would that impact to the bottom line be for you?’”
He says having AI analyze data and make recommendations and the agronomist double-checking the work not only could save them a few thousand dollars but also would bring their business to the next level. Spending less time with manual data entry could allow the agronomist to cover more acres at a faster pace.
AGCO is using AI-powered analysis for market forecasting, quality control, and customer support. This has helped AGCO streamline what once were time and resource-intensive processes into something a single person can resolve in a matter of minutes.
“There are huge [market] datasets available for many years — no one can crunch through that on their own,” says Seth Crawford, senior vice president and general manager for PTx at AGCO. “But with the processing power that’s available in the data, we believe that helps us get more proactive in planning for the ups and downs in the market.”
AI tools help handle heavy customer loads during seasonal peak times. The tool listens in to a support call and can pull information based on the customer’s issues such as related cases, relevant equipment manuals, and crop information. This can funnel down the problem to help customer support find the solution as fast as possible.
AGCO also uses AI in its quality control processes to analyze feedback on its machinery in fields across the world and identify potential problems that might plague the entire fleet. The quality control, manufacturing, and design engineering departments can evaluate the data and find the source of the issue, whether it’s a bad batch of parts or a problem with the design.
Walther predicts AI will probably be implemented nearly everywhere in the near future, with RAG at the center. Companies with decades of internal data will be able to sift through and analyze it faster than ever before. He says AI may help farmers manage their data as well — AI can now photograph documents and directly translate them into spreadsheets.
