Ryan Rakestraw, whom The Mixing Bowl has collaborated with in the past, has teamed up with Amrit Acharya from Mckinsey to write the following on AI in agriculture, and we would like to share it with you.
When Norman Borlaug, the father of the green revolution, won the Nobel Prize in 1970, the Nobel Committee remarked that "more than any other single person of this age, he has helped provide bread for a hungry world." Borlaug’s introduction of disease resistant high-yielding crop varieties and advanced agricultural practices was a game changer, as agriculture yields increased tremendously and helped save millions from starvation.
Half a century after Borlaug received the Nobel Prize, we live in a world where yield growth is plateauing and the total land under cultivation is decreasing. Changing weather patterns and water availability is altering productivity in certain agricultural regions. At the same time, world population continues to grow and is projected to reach at least 9 billion people by 2050, much of the growth is clustered in developing countries, where rapid economic expansion is allowing for increased calorie availability and consumption with an increased demand for protein. As these two forces of population growth and demand for food gain momentum, is a risk of reaching a Malthusian doomsday—a scenario where population growth outpaces the growth in food supply resulting in large-scale famines—becoming increasingly likely? Preventing this may very well be one of the most important challenges of the 21st century.
Increased consumption drives increased demand for agricultural production. Growers around the globe are meeting this challenge, but they must do so in a manner that does not irreparably strain the planet’s resources. To balance the sometimes-opposing goals of increasing production and conserving resources, researchers and entrepreneurs are working on ways to sustainably intensify agriculture on its existing footprint. Like Borlaug, these researchers and entrepreneurs have access to the tools of plant genetics, chemistry, agronomics, and machinery. However, today there is also a new tool, Artificial Intelligence (AI).
The Emergence of AI
Despite being ranked near the bottom in industry surveys on the state of digitization, Agriculture is rapidly becoming more digital.
- The adoption of high-speed variable rate planting equipment is providing accurate “as planted” information and yield monitors are supplying granular information about production at harvest. This is fundamental data (input and output) is a key to building predictive algorithms.
- Farmers are using sensors and soil sampling to gather data on soil moisture and nutrient levels across their fields.
- There are a variety of farm information management systems, that make inputting operational and financial data easy.
- Farmers today have access to software tools to assist in in-field scouting. From mobile apps to unmanned aerial vehicles, these tools collect data that can be used to assess crop health and monitor pest and disease conditions during the season.
With this transition, farm data is becoming both more rich and robust. The availability of this data is paving the way to develop and deploy AI in agriculture.
Applications of AI today are primarily being driven by the tech sector, with use cases ranging from enhanced information security to mobile ad placement, to autonomous vehicles.
Five years ago, Google sponsored researchers made a breakthrough in AI when their neural network software taught itself to detect shapes of cats and humans with >70% accuracy. Today, multiple groups have exceeded human image recognition capabilities in the annual ImageNet Challenge, with less than 3% classification error. Google with the IEEE Computing Society is sponsoring a contest called iNaturalist Competition which hopes to train AI algorithms to identify more than 5000 different species of plants and animals.
The power of these algorithms also extends to the interpretation of language. Utilizing AI, Microsoft’s speech recognition system is now at an error rate of 5.1%, an accuracy on par with professional human transcribers. The accuracy of their system has improved substantially every year.
According to Pitchbook, over the last decade, over $17b has been invested in AI related startups in the U.S. There have already been 200+ AI-related acquisitions since 2012. This activity has mostly been led by the big tech giants like Google, Facebook, Microsoft, and Amazon as they look to gain access to capabilities to help transform industries as diverse as transportation, healthcare, retail, and manufacturing.
Is agriculture next in this transformation?
Headwinds for AI in AgTech
While AI has become a mainstay of the tech community, many of the major ag input companies, equipment manufacturers, and service providers have yet to vigorously pursue AI applications in agriculture.
Part of this hesitancy may be the overall lack of familiarity with AI advances and the potential applications, which this article hopes to at least partially remedy.
Additionally, the development of AI algorithms can be challenging in an agricultural setting. AI applications require large amounts of data to properly train the algorithms. In agriculture, while there is a significant amount of spatial data, much of the data is only available once per year during the growing season. Thus, it can be years before a statistically significant temporal data set about a given field or farm is collected. Often, the data collected in the fields needs extensive pre-processing (cleaning up) before it can be reliably used as input to AI algorithms.
Today, there continue to be challenges associated with getting connected to the data. The Wall Street Journal recently wrote about how cell phone reception is spotty or nonexistent in farms, which makes it hard to transfer the data to a location where it can be analyzed.
A lack of standards, perceived poor transparency around data use and ownership, and the difficulty of gathering and sharing data has lead to a situation where AI algorithm developers in Ag are still starved for data. Luckily, products like the Climate Corporation’s FieldView Drive, John Deere’s JD Link, and Farmobile’s PUC are aiming to make the collection and transfer of equipment data easy and seamless.
Emerging agricultural technology (AgTech) companies developing AI algorithms may also be exacerbating the problem. Many startups are building decision automation tools while there still exist large gaps in data collection, preparation, and benchmarking capabilities. Farms have historically lacked the information technology infrastructure and data warehousing systems that Silicon Valley tech firms have relied on to develop and implement AI applications. The data infrastructure on the farm will need to become more robust before large scale agricultural AI deployment can be successful.
Further, some of these emerging companies have tended to avoid the use of scientifically validated, statistically controlled field trials to quantify the benefits of their products. Instead, these companies have used “lean” methods to get to market quickly with a small subset of customers, following the playbook for building a tech startup. While the lean method has worked well in software, in agriculture, a grower simply can’t risk adopting a new technology across their whole farm that may not work. Before launching a product, major agricultural companies put their products through years of field trials to ensure consistent performance and clear benefit. Even with this testing, many growers will want to see new products perform well on a subset of their own acres before complete adoption. Thus, the pervasive “get to market fast” and “scale quickly” mentality may need to be tempered by a more progressive and incremental product launch strategy.
The final headwind is the fact that competition for AI talent is fierce. A common complaint amongst the ag-tech startup community is that it is often extremely hard to find AI talent, in light of competition with employers in the software, internet, and autonomous vehicle sectors. Further, it has been a challenge to retain these individuals after they join a company. A machine learning specialist at one of the MGV portfolio was recently recruited away by a tech giant for over $700k in annual compensation.
The Promise of AI in AgTech
While the headwinds for AI in AgTech are valid, there is reason to believe that AI’s success and large-scale adoption in agriculture are on the horizon. Below are a number of AI enabled technologies poised to transform parts of agriculture.
Abundant Robotics, a spin-out from Stanford Research Institute, has developed technology to autonomously harvest firm fruits. Abundant, a Google Ventures funded company, uses machine vision to detect the location of apples grown on a trellis then targets a vacuum system to pull the apples off the branch.
Resson, a Monsanto Growth Ventures (MGV) portfolio company with offices in Canada and San Jose, has developed image recognition algorithms that can detect and classify plant pests and disease more accurately than a trained human. Resson has partnered with McCain Foods, to help minimizes losses in their potato production supply chain.
AgVoice, a Georgia-based startup, is developing a natural language processing tool kit for crop scouts and agronomists. AgVoice’s approach is context relevant for agriculture. The system interprets sudden death syndrome for the soy fungal disease and prompts for the location and severity of the observation.
Startups such as Orbital Insights, Descartes Labs, Gro Intelligence, and Tellus Labs are developing yield prediction algorithms based on satellite imagery, weather information, and historical yield data. Tellus Labs, claims to be more accurate than USDA reports for everyday of the growing season with predictions available one month prior to the first USDA report in July.
Some have criticized AI as being too rigid for agriculture environments, indicating that there is just too much variability. To some extent this is true, however with advances in computing power AI algorithms can be quickly retrained with additional data.
Such was the case with Slantrange, a San Diego-based startup, which has developed a machine vision system to measure crop populations and detect weeds. The company’s plant counting algorithm was initially developed for the Midwest growing region. This algorithm didn’t perform well when first tested in a South African field, which had lower planting densities and sandier (more reflective) soils. However, overnight the Slantrange team re-trained their algorithm with the new data. The updated version of their software was deployed back in South Africa a mere two days from the issue being first reported. Slantrange recently announced a major partnership with Bayer Crop Science to aid in plant breeding.
Perhaps, the best example of successful implementation of AI in agriculture is MGV portfolio company Blue River Technologies (BRT). BRT is a Sunnyvale, California-based company that was founded in 2011 by two graduate students at Stanford: Jorge Heraud, already an accomplished agriculture executive, and Lee Reden, who had a deep background in AI and computer vision. Initially, BRT focused on using robotics to thin lettuce populations, a process largely done with hand implements today. Now the sixty-person company is applying their See and Spray robotic spraying system towards eliminating weeds in cotton fields. They have shown that they can reduce herbicide usage by 90% by moving from a broadcast spraying methodology, to a highly precise and targeted spray application. This See and Spray technology utilizes AI to analyzehigh-resolutionn images and detect the presence and location of weed species.
Not Just for Precision Agriculture
While the on-farm AgTech applications of AI are certainly important, the application of AI to the discovery and development of new, more efficient agricultural inputs is equally important. However, until very recently, AI systems have not been tuned to analyze data about chemical and biological systems. Thus, there are tremendous untapped opportunities for leveraging AI in plant breeding, biotechnology, agrochemical discovery, and supply chains.
Indeed, AI may see a more rapid adoption for development of new seeds, fertilizers, or crop protection products than for in-field precision agriculture applications. There are two reasons for this thesis. First, agriculture input developers have been meticulous about collecting and storing data over the past decade. This data ranges from sequence information on soy varieties to the structure-activity relationships and environmental biodegradability of synthetic compounds. Second, the financial value of increasing the effectiveness or accelerating agricultural R&D efforts could be quite large.
Per a 2016 Philips McDougall analysis, bringing a single new crop protection product to market now requires over 11 years of discovery and development, with the analysis of 160,000 compounds and expenditures of over $280 M per commercial product. The industry collectively spends over $2.6 B annually in the development of new agrichemical products. The utilization of AI can provide efficiencies in this process.
For example, Monsanto and Atomwise, a startup utilizing AI to develop novel therapeutics for hard to treat diseases, formed a unique research collaboration to increase the speed and probability of discovering new crop protection products. This collaboration is leveraging AI based pattern recognition to reduce the amount of trial-and-error based laboratory testing in early stage chemical discovery.
For the application of AI in crop biotechnology, Monsanto is collaborating with Second Genome. Based in San Francisco, Second Genome is a venture backed company that is developing new drugs based off of insights derived from analysis of the human microbiome. To accelerate discovery of novel proteins for next-generation insect-control solutions, Monsanto is providing access to its extensive genomic databases and utilizing Second Genome’s expertise in analyzing microbial function through big data metagenomics, machine learning, and predictive analytics.
The benefits of AI are also applicable to plant breeding. Monsanto evaluates corn hybrids for many years in the field before bringing them to market, this process can take up to 8 years from discovery to commercialization. Corn breeding has often been compared to finding a needle in a haystack, a 32,000-gene haystack, representing a difficult search problem for generations of breeders. Historically, a breeding program could select around 500 combinations annually for trials from a set of hundreds of thousands of available options. This selection is constrained by the logistics and costs associated with managing the field testing program. To help alleviate these constraints, AI researchers at Monsanto have developed an algorithm that can evaluate breeding decisions and predict which hybrids will show the best performance in the first year of field testing. This algorithm was trained using the past 15 years of molecular marker and field trial information. Mike Graham, Global Breeding Lead at Monsanto, has indicated that this algorithm has taken a year out of the breeding process, enabling breeders to get their best ideas into large scale field trial more quickly. This algorithm not only accelerates the breeding process, but also has enabled Monsanto to scale the size of its corn breeding pipeline by five-fold versus the conventional methods. A breeder can accomplish more utilizing AI tools than they ever could before.
Similarly, Syngenta has recently announced a partnership with the “AI for Good” foundation to bring AI-based tools to seed breeding and improve the efficiency of current crop production methods. Syngenta provides AI researchers with datasets that include seed genetic information as well as soil, weather, and climate data. The goal is to develop algorithms to determine which crop variety or varieties should be planted in a given area.
One emerging company, St. Louis based Benson Hill Biosystems, is also applying AI to plant breeding and biotechnology. They are utilizing their proprietary CropOS platform to find gene candidates to enhance photosynthesis in several crops. CropOS utilizes data from a variety of sources such as DNA and RNA sequence information, field trial observations, and imaging analytics to predict patterns of gene expression needed to get a specific phenotypic response. With each new data set, the CropOS platform re-calibrates and learns, improving its predictive capability.
On the academic front, researchers at Carnegie Mellon University are developing a new initiative called FarmView which uses AI tools to combine plant phenotype data with genetic and environmental data to help breeders and geneticists better understand the relationships between genetics, environment, and crop performance.
AI’s Impact on the Farmer
Over the past 60 years, the number of farmers in the U.S. (as a percent of the total population) has decreased dramatically. These remaining farmers have and will continue to use advanced technology in seeds, crop protection, and automation to increase agricultural productivity.
For the near to medium term, possibly even for the long term, AI in agriculture will require a farmer to actively participate for AI to be successful. Farmers and their advisors are currently well suited to benefit the most from this emerging technology. AI will be a powerful tool that can help organizations cope with the increasing amount of complexity in modern agriculture. Farmers will benefit not only from the direct on-farm applications of AI, but also from the use of AI in the development of improved seeds, crop protection, and fertility products.
About the Authors
Ryan Rakestraw is a principal at MGV, the venture capital arm of Monsanto Company, and is based in St. Louis, MO. He invests in a broad range of companies and technologies that can have a major impact on the future of agriculture. Prior to MGV, Ryan help lead investment activity at Cultivation Capital, an early stage fund investing in healthcare, pharmaceutical development, and enterprise software. During this time, he also founded and launched the Yield Lab, an accelerator program focused on agriculture. Ryan spent the early part of his career in aerospace where helped to design and implement military platforms ranging from undersea to space applications.
Amrit P. Acharya is an Agribusiness expert, currently a consultant at McKinsey & Co.'s Silicon Valley office and previously with Monsanto's Corporate Venture Capital Group and at the Agribusiness group of India's largest consumer goods company (ITC Ltd) where he helped modernize Indian agriculture through technology. He holds an MBA from the University of California Berkeley's Haas school of business and an Electrical Engineering degree from the Indian Institute of Technology Madras.
