Can AI change agriculture? Examining the use cases

Agriculture is a significant industry in the United States. According to Wikipedia, The output of America's farms contributed $134.7 billion—about 0.6 percent of GDP and is a net exporter of food.

As of the 2017 census of agriculture, 2.04 million farms were covering an area of 900 million acres (1,400,000 sq mi), an average of 441 acres per farm. However, 60% of farms are small than 100 acres. Here is the distribution, by acres:

• 1-9 - 11.2%
• 10-49 - 32.5%
• 50-99 - 16%
• Total "small" farms - 60.1%
• 100 - 2000 - 40.4%

Agriculture in the United States is highly mechanized, with an average of only one farmer or farm laborer required per square kilometer of farmland for agricultural production. However, vast productivity improvements are possible depending on the farm type. Still, the return on investment for AI-driven software and equipment is currently not feasible for small grain farms, predominantly corn, soybeans and wheat. Consider a 40-acre soybean firm: 

For 2021, a trend yield of 68 bushels per acre is used, and the projected soybean price of $8.35 per bushel, giving crop revenue of $568 per acre. With a $30 per acre ARC/PLC payment, gross income in 2021 is projected at $598 per acre

Profitability for a 40 acre farm

Soybeans have the greatest selling price for the farmer, at nearly $20/bushel in 2022, but this is highly volatile. In 2018, the price bottomed out at $8.42. The yield on average is 68 bushels per acre, except for a devastating drought in 2012. Therefore, gross revenue in 2022 is projected at $1,360 per acre. Average operating expenses, such as seed and fertilizer (which are fixed), account for about $650. Even without considering land cost, the farmer's return was $710 per acre, or $28,400. Considering soybean prices are at an all-time high, even that meager return requires the farmer to lease land to increase their harvest or to seek employment off the farm to supplement income. In either case, 60% of farmers are in no position to invest in AI. 

Stand-alone corn has a return of $237 per acre return on a yield of 220 bushels per acre.  A 5,000-acre farm could not operate with manual methods, but with a return of $1,185,000, it could afford the investment in the newest equipment like a John Deere combine, an advanced platform for both embedded AI systems and third-party add-ons. It sells for over $380,000 to $480,000. With add-on features, farmers might be looking at $500,000.

This is in the sweet spot for AI. The technology is increasing crop productivity and enhanced tracking, harvesting, processing, and marketing in real-time.

Above, we investigated small to massive grain farms, where planting happens all at once, and harvesting does too. But not all farms are the same. There are a lot of opportunities for farmers in the field because the investment in equipment is out of reach. However, AI software can help them predict the variables to increase their yield and timing. 

But there are other opportunities for farms producing fruits, vegetables, and nuts. Precision agriculture supports product harvesting, processing, storage, transportation and data analysis for more efficient agricultural production. On the modern farm, you can collect data with the use of advanced technology, such as:

• GPS-based soil sampling
• automated hardware
• telematics
• software
• sensors
• cameras
• robots
• drones
• GPS guidance
• control systems

In a previous article, I described a few innovations:

Here's one scenario for AI and agriculture: a drone intelligently scans the land looking for promising spots for planting. It carries the seeds contained in a shell of fertilizer and plants them. Such a system can handle up to six drones at once, and do the work of 20 people and locates viable spots in remote areas. But how efficient is it to maintain and harvest the plants spread out, possibly inaccessible areas? I've heard one company claim that the seedling sprouts into a plant 90% of the time. But the narrative goes like this: 

• Reducing pesticide and water usage. Once the seedlings are planted, they need water and pesticide. AI systems can use computer vision and only dispense where required, potentially using 50% less water and 90% less pesticide. Soil health sensors could give the farmer 10-12 measurements of moisture, humidity, chemistry, etc., to increase the farmers' yield and cash flow.
• Vertical farming with AI is also possible. Because the environment is closely controlled, AI models can constantly adjust the pertinent factors. The yields per "acre" (the vertical farm's footprint, with the help of AI, can produce 10x the yield). However, there is a clear difference in the quality of "hot-house" tomatoes and those grown in the field. Vertical farming is interesting, but I'm unsure if it will produce quality output.
• Harvesting - some products could be harvested daily based on ripeness. For the smaller harvest, like picking fruit, the machines are probably too costly to break even. There are quite a few harvester robot models, but their deployment is, at the moment, minimal.
• Fruit-picking robots. There are some trials using robots to pick fruit from trees. I have my doubts about this. A line of welding robots in an auto assembly line is one thing, but mother nature is a tough nut to crack. Sunlight can come and go from clouds of the trees which would distort the image recognition system. There are benefits if it can work. Laborers spend more time moving their ladders than picking fruit, and the rate of injury from twisting and falling is high, but the unemployment rate could be just as high.

My take

Bottom line, the very large and the very profitable (fresh fruit for example) agricultural businesses are already experimenting with AI, some with promising success. Autonomous combines monitoring plant health and judiciously applying fertilizer, water and pesticides, on a very large scale, are making progress. Likewise, small farmers using AI software to improve their yields are already seeing benefits at very low price points. Vertical farming and fruit picking robots have a long way to go. Also see my prior piece, Can AI help us address food insecurity and food waste?