Editor’s note: This story was updated on April 3, 2024, with a new name and link for Max von Olfers’ AI chatbot.
A Yakima, Washington, startup is inventing an AI chatbot for the tree fruit industry. They’re not alone.
Kragworks Ag Solutions is one of several farm service companies, startups and techno-tinkerers launching virtual consultants to answer questions about yield, weather and profitability in conversational style with generative artificial intelligence using large language models.
The technology — software that mines digital data and communicates via human-like language — quickly made news headlines and increased societal angst as it was adopted for everything from online tech support chats to teens trying to cheat on English papers.
The most infamous large language model, ChatGPT, was released by developers OpenAI in November and has an open platform that allows other companies to build their own chatbots using the underlying model. In the following months, bots held bizarre conversations with journalists, passed the bar exam and triggered copyright lawsuits from comedians and novelists. Alarmed AI company CEOs asked the U.S. Congress to discuss potential regulation of their own products.
The technology also prompted a pivot for Kragworks, which Matt Yeager and Justin Pease founded in 2018 to write logistics management software for warehouses. Now, they plan to maintain those clients but have otherwise shifted their attention to their new Q&A tool, Harvest Engine.
“We’ve just flipped everything on its head in our company and said, ‘OK, we’re completely changing, because the writing’s on the wall with this stuff,’” Yeager said. Yeager and Pease, who both have fruit backgrounds, are “training” their chatbot now with the help of Yakima Valley fruit growers and a data analytics company. They plan to release a beta version in the fall.
Several players
Removed from the fanfare and hand-wringing, several companies in the U.S. and Europe have harnessed the technology for agriculture. Some of them already answer questions from real growers every day. Among them is Norm (fbn.com/norm), unveiled in April by the South Dakota-based Farmers Business Network.
As of mid-July, Norm had fielded more than 5,000 questions, said Kit Barron, head of data science for FBN. Like the company’s 55,000 customers, Norm skews toward row crops but gets some apple questions, such as “What causes marginal chlorosis in apples?” Norm answered a variety of things, including iron deficiency, and recommended iron chelate treatments.
The program will not replace human agronomists, Barron said. Fed by both public information and proprietary data, it offers a quick first stop. Busy extension faculty and advisors can’t always answer questions right away.
Imperfection is part of the large language model landscape. Chat programs sometimes spout incorrect information, which coders call “hallucinating.” Most bots make disclaimers.
“It’s still very much an experiment,” Barron said.
The chatbot at agquestion.com adds a caveat of “less reliable answer” when it’s uncertain. Developed by Iowa-based Smart Agri Labs, the row-crop-focused bot named Centeōtl (after the god of maize in Aztec mythology) gave detailed answers to a couple of tree fruit questions.
Max von Olfers of France started his own AI chatbot, agri1, in March as a hobby and already has fielded thousands of questions from all over the globe covering 300 different crops, including alfalfa, apples and snail farming. He runs a multigenerational farm with his sister, an e-commerce cognac business and a farm tech blog.
Von Olfers may someday charge for the tool — currently on its fourth version — but for now he only wants to “democratize farming knowledge,” he said.
He’s learning, too. He recently asked his own tool how he should fertilize his alfalfa. An early version would have told him to up his nitrogen, but after some back and forth, the recent iteration advised him to leave well enough alone to align with his organic goals. Surprised by that wise insight, he bounced the advice off neighbors and human consultants without telling them it came from a computer. They agreed.
About 60 percent of his users are farmers, but consultants visit more routinely. He suspects that in the developed West, conversational artificial intelligence will not replace a human’s sage advice. Farming requires a skill set too broad for a computer, he said.
“No other job requires this much generalist understanding,” he said.
Relationships
Wilbur-Ellis, a Central Washington supply and service company, already uses an internal AI chat tool but sees the potential to build it out so field reps have access, said Mike Doerr, strategic business manager. Before that, however, the company’s technology team would want to feed it good data, such as peer-reviewed research papers and respected trade journals, not news headlines.
That would make advisors more efficient and accurate, Doerr said. Wilbur-Ellis would also need to hire more tech-savvy employees to keep up.
“I think in the end it will change our workforce,” he said.
Jeff Allen, who spent 25 years as a crop consultant, expects chatbots to make advisors better.
Field reps spend more time on paperwork than they would like to, said Allen, the director of technical services for G.S. Long Co., an agricultural supply and service company based in Yakima, Washington. Consultants complain more about their computer work than fickle clients or weather. If Allen has a concern, it’s that new consultants will rely on digital tools instead of clients who probably know more than they do.
“It’s very much a people job, a people-person sort of thing,” he said. “It’s very much a relationship.”
Kragworks is putting relationships to the test by training its prototype on orchard blocks operated by NWFM, a Central Washington farm management company. Keith Veselka, co-owner of NWFM, likes the conversational nature of the chatbot and the idea of making the advice specific to the farmer.
NWFM keeps careful, electronic records and is an early adopter of digital scanning technologies that track bloom and crop development. Accessing all that data is arduous.
“You’re in spreadsheet hell,” Veselka said. “You’re in silo data.”
To avoid requiring another app, Kragworks collaborates with Ultimate Analytics, a Carmel, California, purveyor of a data interpretation software with an office in Yakima. They may add Harvest Engine to Ultimate’s dashboard, which already integrates other platforms, said Nate Bostrom, chief technical officer.
The parties are in early discussions, but using the large language model’s conversational ability to answer questions seems “obvious” for the next generation of computing, Bostrom said.
“It’s like a tidal wave, everyone can feel it,” he said.
At Kragworks, Pease, Yeager and their three employees work in a former fruit packing warehouse office decked out with tech bro accoutrements like a pool table and hand-me-down couches. Historic fruit harvest photos adorn the walls.
They were motivated by a recent visit with software engineers at Microsoft, who showed them that even coding is being done by AI now. Yes, computers are programming themselves.
“That’s when we saw the writing on the wall,” Pease said. “It’s like, well, our jobs as developers are going to be going away.”
Yeager grew up in Idaho and worked at Yakima-area fruit company Allan Bros., in software development, where he met Pease, who grew up in nearby Selah in an orchard family.
They have recruited some investors and are looking for more. They also hope to build simulations into their program, so growers could ask the bot to rank outcomes of alternative options.
The Kragworks staff teaches Harvest Engine with extension newsletters, spray labels, growing degree-day models and the like. Yeager and Pease intend to take Harvest Engine further, tailoring it not only to the industry but to the client. Each grower’s chatbot would answer based on personal spray records, scouting reports and profit-and-loss statements. In AI, the partners see not just a tool but a way to archive the experiential knowledge that might be lost when grandpa dies, then deliver it in a tone grandpa would appreciate.
“Farming is questions and answers,” Yeager said.
—by Ross Courtney
Chatbot responses
We posed the same question to the three chatbots featured in this story. Here’s how they responded.
Question: Codling moth is making a comeback in my apple orchard in Washington. What should I do?
Centeōtl
You can try implementing integrated pest management (IPM) strategies to control codling moth in your apple orchard. This may include using pheromone traps to monitor and trap adult moths, applying insecticides at the appropriate timing, and practicing cultural practices such as pruning and removing infested fruit. It is also important to regularly monitor your orchard for signs of codling moth activity and adjust your control measures accordingly.
Norm*
Codling moth is a common pest in apple orchards. To control it, you can use pheromone traps to monitor the population and time your sprays. According to the Washington State University Extension, the best time to spray is when the first generation of moths is active, which is usually around mid-June. You can use insecticides such as spinosad or carbaryl to control the larvae. It’s important to rotate the insecticides to prevent resistance. Additionally, you can use cultural practices such as removing infested fruit and pruning to reduce the population. For more specific advice, I recommend consulting with an FBN agronomist at (844) 200-FARM.
agri1
I’m sorry to hear about the codling moth issue in your orchard. This pest can indeed be a challenge, but there are effective methods to manage it. Integrated Pest Management (IPM) is a comprehensive approach that could be beneficial in this case. IPM includes using pheromone traps to monitor moth populations, introducing beneficial insects that prey on codling moths, and applying organic or conventional pesticides when necessary.
Cultural practices are also important. Regularly inspect your trees, especially during the growing season and remove any fruit showing signs of codling moth damage. This will help reduce the moth population. After harvest, clean up any leftover fruits which could serve as a breeding site for the moths.
If you need more specialized advice, consider using software that can help monitor and manage pests. You can find a list of such software on the agtecher website: https://agtecher.com/products/agricultural-software/
It would be helpful to know a bit more about your orchard. How large is your apple orchard? Are you already using any pest management strategies? Are you leaning towards more organic or conventional methods of pest control? And are there any restrictions or certifications that affect your choices in pest management?
*Editor’s note
Norm’s carbaryl recommendation — a broad-spectrum insecticide that has fallen out of favor as apple growers invest in IPM — prompted Good Fruit Grower to call an entomologist for feedback. Here’s what Chris Adams of Oregon State University had to say:
“This is exactly why chatbots cannot replace crop consultants. You cannot get from a chatbot the implications for next year from a decision you make this year. We’re not recommending harsh broad-spectrum insecticides anymore because crop consultants understand the IPM implications of every decision. … It’s not that it’s wrong, it’s just not the full story.”
On the other hand, Adams liked how the agriGPT bot turned around and started asking its own questions. It’s going to take AI models that can navigate those specifics to give good advice, he said.
“You need to ask the right question so it can answer in a way that’s good advice for your orchard. It’s a skill people will have to learn — how to talk to the bot and to ask a bot the right questions.”
—Kate Prengaman
Leave A Comment