Artificial intelligence is already changing how people work, communicate online, create art and manage businesses. Now the technology is being used in every aspect of our food systems.
AI holds the promise of making agriculture more efficient and sustainable, yielding healthier food with less impact on the planet, according to Ilias Tagkopoulos, director of the , or AIFS, at 69成人.
Take the humble tomato, for example. In California, where some were harvested last year, researchers from 69成人 and other institutions are using AI to reduce loss of tomatoes as they are trucked from field to cannery. AI can help us develop new tomato varieties adaptable to a changing climate and screen fruit for quality in the processing plant.
AI algorithms can make predictions and recommendations based on very large amounts of data. in Woodland, California, uses data every step of the way including sorting the fruit 鈥 and AI can lead to even more efficiency, said Dan Vincent, who retired last year after 19 years as cannery CEO.
鈥淲e鈥檙e already using 鈥業,鈥 in AI. The 鈥業鈥 are people, and they do a very good job, but they鈥檙e overwhelmed with data. So can AI help those people do their jobs better?鈥 Vincent said.
![Red tomatoes on the plant](/sites/default/files/media/images/tomatoes.jpg)
Decisions from big data
AI is essentially a set of tools that creates decisions from data, said Tagkopoulos, who is also a professor in the 69成人 Department of Computer Science and Genome Center. Those tools can support human decision-making as our systems become more complex.
鈥淭he more complex the system becomes, the more the analysis of data and finding the best point of operation becomes difficult. And the food system is really, really complex,鈥 he said.
The seed for the AIFS began with a White House initiative on artificial intelligence in 2019. In 2020 the National Science Foundation and partner agencies issued a call for proposals for institutes working on AI in various fields, including food systems. The AIFS, led by 69成人 and including UC Berkeley, the University of Illinois Urbana-Champaign, Cornell University and the University of California鈥檚 Division of Agriculture and Natural Resources, was established that year with a grant from the U.S. Department of Agriculture鈥檚 National Institute of Food and Agriculture.
In its first three years, the institute has funded research projects on AI technology across the food system, from cheaper sensors for agricultural production to digital simulators to manage indoor farming and technology to predict the nutritional content of food.
As large-scale data collection and AI become routine at different steps in the food supply chain, those steps can themselves be integrated by AI.
鈥淲here the opportunity lies is in integration 鈥 connecting the dots throughout the supply chain,鈥 Tagkopoulos said. In 2015, Tagkopoulos founded , an AI company to help ingredient and consumer packaged goods companies come up with better products from farm to fork.
For example, information gathered by AI about crops growing in a field can be integrated with the AI for plant breeding. That integration up and down the supply chain is a key goal for the AIFS as it enters its fourth year, Tagkopoulos said.
![Artificial intelligence in plant breeding](/sites/default/files/media/images/breeding.jpg)
Making the seeds: AI in plant breeding
One of the original innovations in agriculture was to select and breed plants that make good crops. Today, plant breeders and geneticists continue to push for plants that are tastier, hardier, and more drought or pest resistant. With the rise of modern genetics, breeders can use genetic information to select plants to cross with each other and create new varieties.
鈥淭he genetics helps inform the breeding,鈥 said Christine Diepenbrock, assistant professor of plant sciences at 69成人. 鈥淲e can find different regions of the genome that are favorable for traits of interest, then cross the best lines together and make sure we get all those different favorable regions.鈥
The more information breeders have about the plants, the more quickly they can create the best crosses. Artificial intelligence can support those decisions, Diepenbrock said.
鈥淪o we can get gains more quickly if we can accurately predict how the next generation is going to perform, or how a given crop variety might perform in a new environment where we haven鈥檛 tested it before,鈥 she said.
Artificial intelligence is particularly useful for finding complex, nonlinear relationships, Diepenbrock said. For example, recent work by one of Diepenbrock鈥檚 students studied ways to use DNA sequences to predict carotenoid (provitamin A) concentrations in maize.
Plant breeders often must consider several traits at the same time. Your tomato may be drought tolerant and pest resistant, but you also want it to be nutritious and flavorful. AI can be useful in understanding and taking advantage of the complex relationships between these traits, Diepenbrock said. Combined with DNA sequence information, AI can speed up the breeding cycle and bring new varieties forward more quickly.
![Artificial intelligence in agricultural fields](/sites/default/files/media/images/In-The-Fields-section.jpg)
In the fields: bolstering ag-tech
Sown into fields, farmers must nurture their plants into a crop. That presents many opportunities for AI to increase yields and quality while saving resources, said Mason Earles, assistant professor in the Department of Viticulture and Enology and Department of Biological and Agricultural Engineering, who leads the agricultural production cluster of the AIFS.
The ag-tech industry is already taking notice. Precision agriculture uses agricultural technology, including AI, to monitor and cater to the needs of crops in real-time. Precision agriculture can help farmers minimize waste, use resources efficiently and support sustainable farming practices by providing just the right amount of water, fertilizers and protection against pests to the plants that need it, exactly when they need it.
鈥淭here are a number of businesses all the way from startups up through large scale producers and machine manufacturers who are doing this sort of thing right now,鈥 Earles said.
Major manufacturers like John Deere already produce 鈥渟ee and spray鈥 technology, which can automatically and selectively deliver fertilizer or pesticides where most needed, for row crops like wheat and soybeans. A challenge for California agriculture is that speciality crops 鈥 fruits, nuts and vegetables 鈥 vary a lot in their growing conditions and requirements.
鈥淏uilding tools around AI that work in almond is very different from grape, is very different from strawberry; it鈥檚 different from lettuce,鈥 Earles said.
Using data is not new in agriculture, Earles noted. But now AI could take different kinds of data 鈥 images of plants from video cameras on tractors; water and fertilizer inputs; soil, weather and other environmental conditions 鈥 and combine it to produce forecasts to support farmers鈥 knowledge and experience.
鈥淚t鈥檚 sort of giving them a superhuman suit, to be able to go out and do better the things they鈥檙e already doing,鈥 Earles said.
Precision agriculture technology developed in Earles鈥 lab at 69成人 is being commercialized by Scout, a startup company based in Napa County and Davis. Instead of a farmer walking through their vineyard and inspecting vines in person, Scout鈥檚 technology uses AI to analyze thousands of images of grapevines collected at ground level and provides insights.
![A hypothetical tomato truck using artificial intelligence](/sites/default/files/media/images/tomato-quality.jpg)
Onto the truck: AI minimizing food loss
Early summer in the fields around 69成人, mechanical harvesters are separating fruit from tomato plants and loading it into open trucks 鈥 called gondolas 鈥 for transport to nearby canneries. Along the way, some of the tomatoes may get split or squashed and leak fluid, known as serum. Others may be moldy or unripe. All of these are losses for the producers.
Christopher Simmons, professor and chair of the Department of Food Science and Technology at 69成人, has used AI to study the factors that give rise to loss of quality before tomatoes get to the cannery gate.
鈥淲e wanted to predict when these losses might occur, in order to design interventions that would maintain quality and preserve tomatoes,鈥 he said. While the number of tomatoes rejected is small compared to the size of the harvest, it鈥檚 still a concern for processors, he said.
鈥淚t鈥檚 also a loss of all the resources that were used to produce those tomatoes 鈥 all the water, the energy, the fertilizer, other inputs are lost each time a single fruit is rejected.鈥
Big data on tomatoes
Simmons and a multidisciplinary team at 69成人 and UC Berkeley were able to draw on years of data collected by the Processing Tomato Advisory Board on every load of tomatoes arriving at a processing plant, including variety, season, where it was grown and quality measures.
Their first challenge was to get all this data 鈥 from archives, from documents, on different scales in space and time 鈥 aligned and in a format that could be used to train an algorithm.
鈥淭he quality of the algorithm is going to reflect the quality of the data used to train it,鈥 Simmons said.
Over the course of the project, which was supported for two years by the AIFS, the model did identify some weak but significant predictors for likelihood of mold, specifically the time of season and the acidity of the tomatoes.
New sources of data could make the model more robust. Engineers Ana Arias and James Evans at UC Berkeley, also collaborators at AIFS, have developed a sensor array that could be mounted in the gondola to measure the forces on tomatoes as they are trucked to the processing plant, as well as the level of serum leaking from damaged tomatoes.
The data so far are preliminary, Simmons said, but if these kinds of sensors were in use throughout the industry, they could generate a layer of data that would strengthen predictions.
![Tomato canning](/sites/default/files/media/images/tomato-canning.jpg)
Into the can: AI transforming the cannery
Once tomatoes enter a plant like Pacific Coast Producers鈥 cannery, they enter a high-tech, highly automated environment. Former CEO Vincent said he has seen a lot of change at the company since he started working there in 1983.
鈥淎t the time it was very labor intense 鈥 many people working there, pretty much hand sorting, hand stacking of cans,鈥 said Vincent. Today, the Woodland plant is highly automated with a small, highly skilled workforce.
Pacific Coast Producers is a cooperative owned by about 170 farmers. It operates several plants in California and Oregon, supplying store-brand canned fruit and tomatoes to major retailers across the U.S.
For Vincent, AI is another tool to help processors deliver products to their customers.
鈥淲hether it鈥檚 tomatoes, diced tomato, tomato paste, or sliced peaches, pear halves or maraschino cherries, our job is to take that raw product and get it to the state the customer wants, at the lowest possible cost, at the highest quality,鈥 he said. To that end, the industry already uses high-speed color sorters that use cameras and a rudimentary AI to accept or reject fruit as it passes by.
Pacific Coast Producers sells about 200 different product types (SKUs, or 鈥渟tock keeping units鈥 in industry parlance) to most retailers and food service distributors in the U.S. Those products leave the cannery unlabeled, or 鈥渂right,鈥 and go to the company鈥檚 distribution center in Lodi, California, for labeling.
There, staff have to figure out the most efficient way to label all the different products for all the different customers while minimizing the number of times they have to change over the equipment.
鈥淔or a company like ours, it鈥檚 14,000 data intersections, so 14,000 theoretically unique items you have to manage,鈥 Vincent said. 鈥淪o that鈥檚 where I see a real opportunity for AI, because we have very good data; it鈥檚 just a lot of data.鈥
![Hypothetical scenario where you can scan your food with your phone for nutrient values](/sites/default/files/media/images/scanning-2.jpg)
Plates and stomachs: AI in nutrition and diet
Imagine having an expert dietitian at your side who could tell you what鈥檚 in your meal and what effect it will have on your body. That鈥檚 the potential of an AIFS project led by Danielle Lemay, a research scientist at the USDA鈥檚 Western Human Nutrition Center in Davis and an adjunct professor at the 69成人 Department of Nutrition.
鈥淸Our] main project with AIFS is trying to predict health from food photos,鈥 Lemay said.
Nutritionists use food diaries, questionnaires and peoples鈥 recollection to try to track what people eat. These methods are tedious and not very accurate.
鈥淲ouldn鈥檛 it be great if we could just use our phones to capture what we鈥檙e eating in order to better inform dietary assessment?鈥 Lemay said.
Take a picture of your meal, and an AI could give you a complete breakdown of the ingredients for nutritionists to study.
Alongside the photographic food diary, Lemay is collaborating with Tagkopoulos on an AI-based dietary recommendation system that can start with an image of a food and come up with a detailed list of ingredients and micronutrients, its health effects and interactions. Lemay and Tagkopoulos are working with Hamed Pirsiavash, associate professor in the Department of Computer Science, on algorithms that can generate information based on an image.
鈥淭here鈥檚 this whole world of thousands of molecules in each food that we鈥檙e starting to map with analytical chemistry,鈥 Lemay said. 鈥淚f we were able to map foods to databases that tell us about these thousands of molecules, then that opens up the opportunity to understand more about how the molecules and what we eat impact things like your gut microbiome or other aspects of health.鈥
Human nutrition is staggeringly complex. You have to consider all the molecules that make up the ingredients of your meal, most of them still unknown; all the variation between individuals, in age, weight and genetics; and the millions of microbes that live in our gut and help us digest different foods.
Finding a path through the millions of possible interactions between all these variables is very difficult for human brains, but it鈥檚 the kind of big data problem that AI and machine learning can do easily.
From seeds to fields to stomachs, AI technology is already having an impact on our food supply.
鈥淲hether we realize it or not, AI is becoming interwoven into many of the tools, software and hardware that we rely on. I think we鈥檙e going to see the same thing in the food system,鈥 Simmons said.
![Staff lab manager Matharin Lee (wearing black shirt).](/sites/default/files/media/images/winery-food-science-major-uc-davis.jpg)
What Can You Do With a Food Science Degree?
Do you know where your favorite foods come from? Future food scientists have the answer. Find out what it's like to major in food science at 69成人.
Media Resources
Media Contacts:
- Andy Fell, 69成人 News and Media Relations, 530-304-8888, ahfell@ucdavis.edu
- Amy Quinton, 69成人 News and Media Relations, 530-601-8077, amquinton@ucdavis.edu