WESTERN PRODUCER — Over the past six months, powerful systems based on artificial intelligence have become available to the public, and they will have a profound effect on agriculture.
AI has already been deployed by agricultural companies, from autonomous tractors and sprayers capable of green-on-green herbicide applications to crop phenotyping and disease risk forecast models.
However, the list of AI-based products in the agricultural industry is about to skyrocket with the wide-scale availability to developers and farmers of these new tools.
Last November, OpenAI released an AI chatbot called ChatGPT that’s built on the company’s language models and fine-tuned using both supervised and reinforcement learning techniques.
Since then, this program has upended the global tech industry with competing companies, including Google and Facebook, racing to release their own AI-based systems to the public.
One of the first agricultural companies to use a ChatGPT is Farm Business Network (FBN), which launched an AI powered agronomic adviser named Norm that’s built upon ChatGPT.
The name Norm was chosen to pay homage to Dr. Norman Borlaug, an agronomist who helped usher in the Green Revolution.
Kit Barron, head of data science at FBN, said Norm is available on the FBN website to provide farmers with a wide array of agronomic intelligence.
“I noticed that certain topics in the base model of ChatGPT don’t quite get deep enough or rely too heavily on guessing or probabilistic responses. So, we started thinking, how might we turn this into something of value for our farmers,” Barron said.
Norm accesses publicly available data such as weather insights, soil monitoring, application rates, product labels, current events, university research, grower commentary and FBN proprietary data feeds.
“We started off by feeding it data and information from our blog, since we have an extensive agronomy blog with hundreds of articles about different specific facets of agriculture. We’ve also fed it our product label database that basically powers the back end of our store,” Barron said.
Users can ask Norm on the FBN website a wide variety of questions, although Barron said they installed limits for the chatbot so that it only responds to questions related to agriculture.
For instance, Norm can recommend a crop protection product or practice for a particular type of problem on a specific field.
“Norm would take the location information (of where the product will be used) and whatever issue that you’re trying to address and then query the product label database and hopefully respond with some relevant recommendations,” Barron said.
“Another example of an area where Norm is quite well-trained at this time, is he knows an awful lot about equipment and tractors and loves spewing off loads of detail about different types of agricultural equipment.”
Norm can also help farmers with nozzle selection, seed varieties based on geographic region, soil type, climate, market demand and livestock and animal health including diseases and pharmaceuticals.
Barron invites farmers to experiment with Norm, and he said the chatbot will improve overtime by using these interactions to refine the model.
Guy Coleman is a PhD student at the University of Sydney who works in precision weed control, including computer vision, robotics and laser weeding.
His research now focuses on how crop-weed biology can be incorporated into weed recognition for site-specific weed control.
He said some of the AI systems recently made available to the public are powerful tools that will help researchers and start-up companies develop new agricultural techniques and products.
For instance, Coleman has been experimenting with Meta’s Segment Anything Model (SAM) released in April, which uses AI for image segmentation.
This system can quickly separate objects in a photo or video, including a person, from the rest of the image, which may be helpful for researchers.
“The whole process of segmenting and annotating your images is critical for training algorithms to recognize weeds and all ranges of plant diseases or insects in fields. That has been a major bottleneck to get some of these algorithms working. The first step has been collecting enough data, including images, and then the next step is annotating them,” Coleman said.
“This algorithm, SAM by Meta, could automate some of that annotation process and cut out some of the tedious work behind it.”
Previously, researchers would do this by bounding box detection, which is done by drawing a box around the weed, disease or insect in the image within the program. Then the algorithm would predict similar boxes in other images.
Researchers can also perform manual segmentation: specific pixels on images are annotated where a weed, disease or insect is located.
“Things like laser weeding or that type of a more advanced weed control often needs a more precise level of information about where the weed is. But this is all manual, you have a little touch screen tablet, so you draw around the edges of the weed, and it takes hours to do. So, any way that automates that is a big advancement,” Coleman said.
With the use of a program such as SAM, it may soon be possible for users to just click an object on an image and the system will immediately segment it.
This could quickly improve the quality and availability of green-on-green spray systems into more growing regions because the regional and crop specific algorithms capable of identifying weeds within crop canopy is the last major hurdle before wide-scale adoption of green-on-green systems is possible.
Coleman said new AI tools will help with product development across the agricultural sector.
“The rapid change in AI tools, I think, will really increase the speed of development because all of the sudden these ideas that people have, they might need engineering expertise just to get off the ground,” Coleman said.
“Things like ChatGPT, SAM and DINO (another recently released program) can help leapfrog that first stage so you can get to a prototype much faster. People with ideas can then start prototyping quicker as opposed to waiting for people with engineering expertise.”
He said commercial agricultural companies will also integrate these new tools, which will soon help them make rapid gains in their technologies.
Many AI experts are ringing alarm bells because they say unleashing some of these programs before establishing robust limitations and best practices will have serious consequences.
However, countless companies and people are already building new products based on recently released AI models, especially ChatGPT, and farmers could be major beneficiaries of this movement.
This is because of the wide variety of topics producers must understand to be successful.
From marketing and finance to chemistry and biology to mechanics and computer programs, new language models will soon help farmers quickly access information they need.
And it won’t be long until farmers struggle to understand how their predecessors got the job done without guidance from AI.