Agricultural Disruption

Qual Dairy in Lisbon, North Dakota, is one of the first in the nation to install a robotic rotary milking system. Sixty robots milk 60 cows at a time—at a rate of one cow every 16 seconds. Photograph: Mikkel Pates, Agweek, used with permission. Artificial intelligence (AI) is poised to dramatically change agriculture around the globe. New technologies are likely to increase food production and sustainability. But millions of small-scale farmers and seasonal laborers could lose their occupations to robots that would perform repetitious, routine tasks. How these farmers and migrant workers will find new livelihoods is not addressed by agricultural disrupters. “Robots won’t take all jobs at once,” says Evan Fraser, a food-security researcher and geographer at the University of Guelph in Ontario, Canada. Workers will collaborate with automating machines for simple tasks and later for more complex ones. “Automation will supplement the tasks of many workers at first; only later will machines replace them.” Agriculture is probably the least automated economic sector in the world. But just as advanced digital tools are transforming healthcare, they will revolutionize farming, initially in high-income countries, where the process is just beginning. “Even in Canada and the US, we’re just scratching the surface of what can be done,” says Fraser. Digital agriculture—or “smart farming”—is advancing through the use of improved sensors, more-accurate computer vision systems, and powerful AI. Someday, large-scale farms will use remote and built-in sensors, cloud storage in digital warehouses, AI software to analyze huge volumes of data, and algorithms to guide machinery. Automated farming systems could grow more bountiful crops on the same acreage at lower cost while using smaller volumes of pesticides, fertilizers, and water. Smart farming, then, is expected to help meet the rising demand for food in more sustainable ways. Global demand for agricultural products is expected to grow 50 percent from 2005/2007 levels to 2050 levels to meet population growth, a rising middle class in many nations, and greater demand for meat and other animal protein, according to the United Nations report, “World Agriculture towards 2030/2050: The 2012 Revision.” Artificial intelligence could be an important tool in addressing this looming agricultural crisis. At the same time, many of the benefits of these new technologies could go to the wealthiest farmers. For others, including millions of small-scale farmers and migrant workers, the impact could be devastating. More than 50 percent of farm tasks are both physical and predictable—and likely to be automated, according to “A Future that Works,” a December 2017 report by McKinsey and Company. But societies both past and present have experienced similarly wrenching workforce challenges and adapted successfully. In 1850, agriculture employed about half the US workforce; in 1970, that figure was only about 5 percent. New kinds of jobs in expanding industries, particularly urban manufacturing, eventually replaced most farm work. A Mexican migrant worker picking tomatoes in Fort Blackmore, Virginia. Although many farm workers are in the United States on H-2A visa, many are undocumented. The US food system depends on immigrants more than any other sector of the economy. Photograph: Laura Elizabeth Pohl/Bread for the World (https://creativecommons.org/licenses/by-nc-nd/2.0). Other countries have transitioned from rural to urban societies even faster. One-third of China's workforce moved out of agriculture between 1970 and 2015 when young people flocked to cities to find low-skilled manufacturing jobs. But that work rung could weaken as factories automate and urban employment requires further education and training. Some developing economies are struggling to create jobs for growing urban populations of more educated workers. Smart farming will change what farm workers do as well as where many do it. “Automation in agriculture is taking away low-paying jobs and replacing them with good-quality jobs in Canada,” says Fraser. A growing number of Canadian agriculture-related jobs are in information technology (IT), marketing, and telehealth veterinary services. Says Fraser, “If I’m a farmer with more spare cash because I’ve reduced my labor costs and increased productivity, I can do a better job of marketing my product and spending money on advertising or social-media marketing to give my milk, for instance, or other produce a more distinctive appearance to the public.” But as agricultural work becomes increasingly autonomous, some tasks could be done remotely, which might not help rural economies that fail to attract workers with digital skills. As automation improves, many smaller-scale farmers will probably continue to struggle, but migrant workers could be hit hardest first. “Automation could disadvantage people who are already very disadvantaged,” says Fraser. “These are typically very vulnerable people who depend on seasonal work as their main source of livelihood.” Will workers arrive in time for harvest? That is the question that keeps many farmers up at night. Seasonal labor in high-income countries is becoming scarcer and more expensive because of tougher immigration laws and migrant-labor crackdowns. Many growers hope that advanced sensors and robotics could help augment a shrinking labor force. Growers meanwhile are being squeezed by changing demographics, says Avi Kahani, CEO of Israel-based FFRobotics, which is one of the few companies developing apple-picking robots. “From Washington State and California to Argentina, to Israel and Poland and Italy, and even to India and China, I’m told, “Think about your own son. Would you let him go and pick apples or would you send him to the university? ”” Picking soft fruit is one of the hardest automation challenges in agriculture. Soft fruits—such as apples and pears—are easily bruised, so every apple sold on the global fresh market is still picked by hand. If robots became cheaper and more reliable, though, they would represent a major advance in farm automation. But many migrant workers would lose their jobs. This apple-picking robot, developed by FFRobotics in Israel, uses cameras and facial-recognition software to identify fruit (left). To use the robots, orchards are transitioning to fruit-walling systems (right), rather than allowing tree branches to grow randomly. Photographs: FFRobotics. Each autumn in the United States, about 40,000 people pick and pack apples. Temporary migrant laborers constitute about one-third of this peak-season labor force. Most arrive on H-2A guest visas, though an unknown number of seasonal workers are undocumented. Growers complain that the guest visa system is confusing, bureaucratic, and unreliable. After years of field tests, FFRobotics plans to introduce its apple-picking robots into Argentinean orchards in March 2020 and in Washington State for autumn harvest. The six-arm robot, sitting on a human-driven tractor, is expected to pick 90 to 95 percent of apples, and small numbers of human laborers would follow behind to pick the rest, says Kahani. Kahani's apple-picking robot uses cameras and facial recognition software to identify individual fruits. Nearly all of today's commercial facial-recognition systems are based on artificial neural networks or “deep learning” software. FFRobotics trained its software by feeding it huge numbers of labeled digital images of apples. The software learns how to recognize certain features—color and shape and orientation—of each apple's “face.” Drones are being used in agriculture to identify crop diseases, apply agrochemicals, and assess harvests. Photograph: Agridrones Solutions Israel. The software is also trained to identify leaves and branches and other orchard background details. Over time, the program learns to differentiate confounding backgrounds from apple faces. Eventually, the system begins self-teaching, improving its own ability to recognize an individual apple. Finally, programmers train the robot's arms to move toward identified apples and pick them with gently grasping “hands.” Robot pickers are feasible only in modernized orchards. A traditional apple tree has branches that grow randomly and crookedly in many directions. Many apples hide in shadow or behind limbs and leaves—details that confuse facial recognition systems. Over the past decade, geneticists have improved designs of dwarf tree varieties that growers prune and thin to create simpler canopies. Limbs grow in vertical planes or V shapes along orchard rows. In a “fruiting wall” canopy, apples are less obstructed from view and reachable within the length of the human arm. These simplified canopies also allow a robot a better chance of identifying and picking apples. Orchards annually remove some aging trees and plant new ones, but it can take decades for a grower to transition entirely to a fruiting-wall system. To afford high-tech robotics, some larger orchards are likely to buy up smaller ones, consolidating production into fewer hands. Joshua Haslun, lead agriculture analyst at Lux Research, a technical innovation consulting firm, is doubtful about FFRobotics’ 90-to-95 percent picking efficiency hopes for 2020, even in modernized orchards. “Fifty percent of the harvest by a commercially viable apple picking robot is the best-case scenario by 2020,” he says. Changing sunlight, shadows, weather, and dust tend to confuse even sophisticated computer vision systems. “When you’re out there in the field, the vision system is difficult to control and it misidentifies things and makes mistakes,” says Shriram Ramanathan, who leads the big data analytics practice at Lux Research. “Most success stories with deep learning systems and robotics in agriculture are in controlled settings. Operating inside a greenhouse is a much, much better option.” Indeed, smart farming's leading edge is indoors—in milking barns, cow sheds, chicken houses, and particularly in industrial-scale greenhouses. Facial recognition software and cameras mounted in cow sheds are already helping farmers identify individual cows and indicators of sickness or injury. Farmers can treat livestock sooner, protecting their investments. Around 50 percent of all European herds will be milked by robots by 2025, according to a 2018 whitepaper by the UK-RAS Network, an academic–industry collaboration. Robots are starting to remove waste from cattle cubicles and move feedstuffs, as well. Cameras and other sensors in industrial greenhouses monitor temperature, plant and stem density, soil moisture, and other parameters. By manipulating climate factors, managers can increase yields with fewer resources, growing up to 10 times more fruits and vegetables than open fields, with much less water. Some advanced greenhouses deploy AI and sensors to analyze which seeds produce healthier yields under different climate conditions. Someday, self-learning software could largely take over greenhouse management, crunching data to determine the optimum climate and other needs for particular crops. Computers would instruct robots when and where to plant and pick produce. Growers could manage greenhouse systems from anywhere in the world, although IT workers would be needed on site to tune algorithms and monitor and repair equipment. Human farm labor is becoming more expensive and harder to find while some robots are becoming less costly to operate per acre. This chart shows that a lettuce thinning robot is expected to be cheaper than human labor by 2024 to 2027, depending on how rapidly the robot improves efficiency. Chart: Lux Research. Farm automation outdoors is moving in a similar direction, albeit more slowly. In 2016, autosteer tractors accounted for 10 percent of the US market, and that share is expected to increase, according to Lux Research. (The report is proprietary. This information is from a press release.) Small robots are thinning young lettuce buds and trimming wintertime grape vines. Robotic vision systems can “see” spring lettuce plants more accurately against dirt backgrounds that are free of early weeds. Wintertime grape vines have lost leaves that can confound a robot's vision. For the past decade, farmers have increasingly used data from aerial sensors—drones, airplanes, and satellites—to identify crop diseases and estimate potential harvests. For instance, hyperspectral images from aerial sensors can illustrate ground moisture and heat as color-coded images. The most advanced aerial sensors are typically used by very large agricultural operations on broad, flat fields of monoculture row crops such as corn or soybeans. Advanced sensors typically are not affordable for small, isolated farms on varied or hilly terrain that grow a patchwork of different crops and livestock. “Larger farms,” says Haslun, “are more likely to have the funds, insurance, and credit to decrease their risk as they try new technologies.” The shift by individual growers to automation will be determined largely by costs and availability of labor for different crops and places. For instance, if shared by multiple farms, experimental strawberry-harvesting robotics in Japan nearly match the cost of human labor. But robots there will have an advantage. The average Japanese agricultural worker is 67 years old—and more than 70 in highland regions. Some nations are consolidating farmland to improve agricultural efficiency, guarantee future food supplies, create more profitable exports, and set the stage for automation and robotics. China, for instance, has undertaken a sweeping land-consolidation program to improve food production, and its universities are among global leaders in artificial intelligence research in agriculture. After the collapse of the Soviet Union, Russia's farm sector continued to sputter. In the early 2000 s, the Kremlin began an initiative to turn Russia into a global superpower in agriculture. Over the past 15 years, Russia's federal, regional, and municipal governments and private companies have merged thousands of former collective farms into nearly 800 megafarms, each of which can comprise hundreds of thousands of hectares. Megafarms have become more productive, more profitable, and economically far stronger than midsize and smaller ones, according to an October 2018 study in the Journal of Agrarian Change by Stephen K. Wegren, a political scientist at Southern Methodist University. The Virginia Tech team won first place in 2018 for this watermelon harvester, in the annual Agbot Challenge, held at an Indiana farm. The machine uses computer vision and machine learning to locate the melons. It can even test for ripeness, using a slapper that hits the melon and analyzes the sound. After measuring how much current it takes to lift the melon, the machine provides a size and weight estimate. The process takes only a few seconds. Photograph: Sam Teller, Virginia Tech. McKinsey Global Institute. 2017. A Future That Works: Automation, Employment, and Productivity. (www.mckinsey.com/mgi/overview/2017-in-review/automation-and-the-future-of-work/a-future-that-works-automation-employment-and-productivity) McKinsey Global Institute. 2017. Jobs Lost, Jobs Gained: Workforce Transitions in a of the and UK-RAS The Future of Robotic are to political both and and they the of and says “These farms are both and so they control all of the production for instance, a when a farm or has control over other of food production such as operations and In are to further to robots for and Each robot is to cost about for the of the average small or midsize farm The program to IT to help robotics into Russia meanwhile is in greenhouse In the a of all fruit and from European countries being with the of Kremlin is developing industrial greenhouses to produce and and tomatoes that they won’t have to says The a program for high-tech greenhouse it for to up to percent for their although the will this This robot, developed by at plant The robot is of an initiative artificial intelligence plant and agriculture at the global food crisis. Photograph: University. industrial greenhouses are taking over market share from A farmer in the as by to smaller than which is the most farm size in many around the world. In are typically by who have other of small in they have an in growing many of the But their share of production has in says. industrial I how they will be to The global industrial crop and to and which further global market in fewer hands. of global food and agriculture there are only and merged and became the and and merged to become by of other and merged into which could become of These companies control most of the that farmers to including 67 percent of global market share of seeds and 70 percent of Other companies control percent of farm according to the a 2018 report by the a Canadian and European large many of them and have as an up smaller operations and consolidating into in and But these are not and the growing of industrial have small farmers for says of the a based in In the s, the of many small farmers. to buy volumes because that them But smaller farmers produce so they the same To some farmers out their and that it for them to and credit But many small farms have their costs of production their US farm has by half and half of all farmers have says global agricultural companies farm data and Global and have or with the farm as the very digital that farms to in industrial agriculture. Smart farming about data “A of these data are based in the tractors and other and they are to and and says a food-security scientist at the University of in Ontario, Canada. By a farm with companies and and also with Someday, could of data analytics with to crop and Farmers might into crops if they take such Says this global companies can to “If you this can guarantee the size of your and that as a of for farmers. But will farmers to the data that from their own If they own these data, what kinds of could farmers Farmers in the future might have fewer to farming such as and crop that and increase soil is a potential that all rural agriculture will become says that about the of the of for farmers and rural If small afford digital they out by larger who and that can to out the livelihoods of many millions around the Smart farming could potential in data and says. At one a farmer would be to give up all to data controlled by a small number of At the other open data systems would allow farmers more to among technologies and Global agriculture is moving in the first are of these data and says Fraser. also have companies in the sector that are not and because not present in in that but they are also data though many people Some are less about global every data of farm they are more that many rural will behind in In the rural that are more likely to have or to help farmers advanced technologies and to workers with computer skills. percent of in the United States affordable The and private to according to a 2017 the Digital by of the of and in is likely greater than from most other of Global have in fewer companies more data and developing tools to help farming advance more other major food farms and commercial also automate their systems as continue in sensors, robotics, climate and AI software. In the who might most and midsize or global companies that data and control data Smart farming could between small farms and large-scale The expanding food could help small to stronger for their produce. But it that many farms and rural will to from the in food based in