VWV Volumes #6: ML Legend (and also VWV Associate) Danny Adkins '21 gives a short intro to AgTech
A Brief Intro to AgTech
Agricultural technology, also known as AgTech, is a rapidly growing field (pun not intended). At the same time, there are relatively few engineers and students who have a priori expertise in agriculture. I’m part of Van Wickle Ventures, a student-run venture fund that seeks to enable innovation across industries at Brown and RISD.
Agriculture is hard to “break into”, exceptionally complex, addresses global problems, and has historically been under-addressed or wrongfully addressed by technology. In this post, I’ll give my (humble) understanding of the problem space, existing market, and emerging technologies in the space.
Every person in the world needs food to survive. This is a problem that has been persistent throughout history, yet grows more pressing as our population grows. Furthermore, we face an existential threat of the so-called Malthusian Catastrophe, a theoretical situation in which we reach the asymptotic capacity of food production and a mass starvation event occurs. In order to serve our growing population, we need to adapt our systems to be able to break free from the pre-technological bounds of food production.
Whether food is understood as a human right or (per the regrettably majoritarian view) a marketized commodity, the need for technological progress to serve a growing population is dire.
I. Problem Space
Let’s build a (very) rough map of the problem space. Writing this post, I found that the problems were seemingly endless. Here, I’ve tried to list some that I think can be addressed with technology.
Internalproblems, which pertain to a farm’s own material interests, include problems of minimizing cost and maximizing the farm’s long-term prospects. Externalproblems, which affect the broader population beyond just individual farms, are plentiful and include issues of food security, supply, access, environmental impact, and food quality.
Production Cost
Minimizing the cost of food production is clearly essential for any farms. To do this, technology should be used to decrease natural inputs, human inputs, and any other production inputs.
Natural inputs
In terms of natural inputs, we can look towards three fundamental resources that every farm (currently) uses: land, water, and nutrients. The amount of food produced per unit of land is called yield. Similarly, the efficiency of water and nutrient inputs are called water use efficiency and nutrient use efficiency. The game? Figure out how to make more food while minimizing the usage of these resources. Maybe we can even eliminate them -- but we’ll get to that later.
[Crop yield is highly correlated to nutrition and irrigation. Nature Magazine]
Lots of factors prevent farmers from, to venture capitalists’ chagrin, just fixing it with money. One economic factor is that they quickly reach diminishing returns, with the cost becoming too high. Non-economic factors that limit scalable access to these resources include:
Regional land-management policies
Lack of access to sustainable water
Political instability
Human inputs
Labor and management practices are the key human inputs to crop productivity.
Management practices are the different human-mediated approaches to solving all of the problems at hand. There’s lots of choice on how to grow, even when you have all the resources. How should one rotate or select crops? What micronutrients should be used? How can we best manage pests for our specific climate? How do we make sure the seeds we’re using are top-tier? If we can systematize and create infrastructure to analyze the efficacy of different strategies, we can optimize for the best management practices.
Other inputs
While there are plenty, one other notable input is logistics. This encompasses transportation, accounting, and more. These issues are highly-specific to farm types and distribution strategies.
Resilience
Farms also need to improve their long-term prospects. There’s a considerable tradeoff between short-term cost factors and long-term efficiency in agriculture.
Yield saturation
Long-term yield needs to be optimized in the same way that short-term yield does. Factors such as yield saturation contribute to long-run inefficiencies that have little to do with improved unit costs.
The common mechanism causing yield saturation is the degradation of soil quality. Soil quality is determined by both moisture and nutrient content.
George Washington Carver might come to mind here: his idea of crop rotation utilized peanuts to help replenish nitrogen in the soil, making cotton production more efficient in subsequent years. The method has since been generalized and canonized. Tillage is another traditional approach.
Geographic variability in production
Geographic variability (precipitation, temperature, soil moisture, arable land, type of crops) cause yield gaps, which are the “differences between observed yields and those attainable in a given region” [https://www.nature.com/articles/nature11420].
Yield gaps not only lead to inequitable outcomes and food access -- they’re also a clear indicator of production problems to be solved.
Climate resilience
Climate change is a crisis that will affect every walk of life, and its comprehensive impact on agriculture is difficult to foresee in entirety. We do know that it will be exhaustive and drastic. While solving climate change or even understanding it is beyond the scope of this post, we need to ensure that farms can be as resilient as possible to its impact.
Climate change is increasing the incidence of “natural” disasters, especially in coastal communities. Climate change looks to destabilize political regimes in conflicts for increasingly scarce resources. It also heavily affects production inputs to farming. Temperatures, precipitation, and access to cropland, all previously taken as certainties, will become radically unstable. Beyond this, second-order effects like migration of wildlife and increased likelihood of pandemics are likely to disrupt agriculture even further.
[Precipitation is expected to change drastically over the coming century. MIT]
Security, Supply & Access
Issues of food insecurity are incredibly complex. Beyond production issues already named, MIT Prof. Dennis McLaughlin identifies key systemic failures in ensuring all demographics have access to food [Land, Food, and Water]. Socioeconomic factors are a strong predictor of access to food, and existing market solutions can even exacerbate the crisis.
Food deserts are a massive problem.
[Getting food is incredibly difficult for many. StreetsBlog USA]
Across the United States, where 50% of all produce is thrown away, communities are still systemically underserved by food banks and markets [The Atlantic].
Notably, existing incumbents like Feeding America represent the complexity of the issue. While successfully moving hundreds of millions of pounds of food per year, these incumbents have failed to combat food insecurity nor address food deserts across the country. Furthermore, they are sustained by corporations like Walmart, which contribute to the economic precarity that causes food insecurity. Philanthropy isn’t a perfect fix: it provides a charitable vessel for exploitative corporations to receive tax and PR benefits. We can do better.
Safety
Quality of Food
One large indicator of the issue is the obesity epidemic in America. The nationwide crisis causes undue suffering, diverging health outcomes, and nearly $210B/year in healthcare costs in the United States [Pewtrusts].
The issue is a confluence of factors: predatory marketing practices, behavioral shifts, and increased access to low-quality food. It’s clearly a political and economic problem. But if we can make more nutritious food (less sugar-dense) less expensive to manufacture and distribute, we can help slow the growth of this crisis.
Environmental sustainability
Finally, the behemoth: environmental externalities. I can’t begin to broach the extent of the Herculean task, and it deserves essays multiple scales of magnitude larger than the rest of agriculture. Some current practices are wildly unsustainable yet the population continues to grow rapidly. Every solution needs to consider its effects and nuances with regards to environmental footprint.
Here are some immediate, broad-strokes questions:
How can we decrease the per-unit environmental impact of water usage? Or, how can we minimize the need for water usage? [https://www.nature.com/articles/nature11420]
How can we minimize food loss and the related carbon emissions?
How can we farm at large scales without the colossal carbon footprint?
How can we avoid pollution of nearby areas and eliminate runoff?
Who is being affected by unsustainable practices? What will the future look like if we don’t fight the issue?
II. Markets & Agrosystems
This is a massive space, and simplifying it down is nearly impossible. Every continent has different needs, systems, and strategies for agriculture. For example, the problems that Africa faces are centered around consolidation of fragmented farms and improving crop yield, since it has 25% of the world’s arable land and only 10% of its agricultural output. In North America, agriculture looks very different, characterized by massive farms.
[Distribution of farm sizes. MIT]
This variance has left many farmers’ practices underserved or disrupted by international attempts at innovation. Furthermore, the rise of international credit markets and trade has led to destabilization of agriculturally-dependent economies. Some liken it to a neocolonialist expansion by Western nations [Yanaizu]. These problems are not unique to farming, but they are tremendous. Any optimal solution will consider existing practices and expertise alongside new technologies.
III. Technology
Finally, we get to the technology. This, admittedly, is a topic I’m more (humbly) comfortable writing about.
Agriculture has, quite literally, been under disruption since before history was written. Humans formed collective societies around crops during the so-called Neolithic Revolution 12,000 years ago. Just about everything we know about agriculture has been engineered and re-engineered countless times.
Most recently, genetically modified organisms (GMOs) and the Green Revolution have catapulted crop yield and mechanization to new levels. If this post so far hasn’t convinced you of the scale of agriculture, look no further than the following statistic: Norman Bourlag, the “father of the Green Revolution” and Nobel Prize recipient, is believed to have saved 1 billion people from starvation [Science Heroes].
Future technologies
Sensors
The trend across all sectors is the same: collect more data, drive better business outcomes. In farming, we’re still at the early stages of this transformation.
Sensors come in two main categories: airborne and land-based. They’re usually optical, but can also be magnetic, ultrasonic, electronic, gyroscopic, and more [Bogue]. Think: drones, tractor sensors, cameras, or other small IoT devices.
Overall, sensors are being used to record numerous physical and chemical factors, such as moisture content, seed count, crop weight, crop loss, etc.
In conjunction with sensors, enormous investment is being poured into computer vision for agriculture. Once 2D/3D visual data is captured, we can infer important information from it.
One final development in this vein, which displaces the need for human labor, is the advent of autonomous robots. As we see more advances in this field, the technology encroaches further towards replacing humans entirely. Fully-autonomous factories are being tested across the globe. For agriculture, this could mean automatic crop-harvesting, planting, maintenance, and everything a worker could do. It should drastically increase agricultural output, but not necessarily change food security in the short-term unless further measures are taken to ensure economic security of workers whose jobs are threatened.
Genetic Engineering
The past 20 years have seen extraordinary strides in nanotechnology and genetics.
In the year 2000, the Human Genome Project sequenced the entire human genome for a whopping $3B. Today, we can sequence the human genome at nearly 200,000 times cheaper cost.
Furthermore, advances in genomics and deep learning allow us to predict the expression and downstream effects of different genetic edits [one example].
All signs point towards a radical increase in our capability to use genetic engineering on a massive scale for specific target outcomes.
For example, CRISPR-Cas9 is a well-known gene-editing technology that allows easy splicing of RNA. Using gene-splicing techniques and analytical inference, we can predict how a given edit will affect crop yield and easily manufacture the seeds in a lab.
Risks include but are not limited to a decrease in biodiversity, unknown health effects, and adverse effects on external ecosystems [Wolfenbarger & Phifer].
Other synthetic biology
I’m sure you’ve heard of the Impossible Burger, a plant-based meal that tastes like authentic beef. The rise of synthetic biology, beyond just genetic engineering, is increasingly enabling scientists to produce food in the lab. The underlying technology is accelerating at a breakneck pace with recent advances like AlphaFold 2. The possibilities for creating realistic flavors from cheap components are boundless.
Lab-grown meat is a clear example of how to mitigate the harmful effects of factory farming. If we can scale the production of lab-grown meat, we’ll rid the need for excess carbon emissions and animal suffering.
Plant-based substitutes are another approach that solves the issue of factory farming. Similarly, synthetic biology may help us unlock new keys to healthy nutrition.
Precision fertilizers can increase crop yield. Better understanding of chemical and biological interactions will augment their production.
Vertical Farming
Capital-rich R&D labs like Google X are investing heavily into vertical farming. This is a process by which crops are grown in vertical stacks. Vertical farming rigs allow crops to be grown with less land required. They’re able to produce up to 10x greater yield when optimally implemented [Benke et al.].
Currently, key bottlenecks include cost of installation and production of vertical farms. Furthermore, these strategies haven’t yet been adapted to work for a wide variety of crops.
Generative Design
Generative design has drawn no shortage of skeptics from across industries. Long seen as a myth that could never replace humans, the paradigm seeks to design new systems automatically. This could include illustrations, new drugs, buildings, or, you guessed it: mechanical systems for farms.
OpenAI’s recent release of DALL-E, using a similar architecture to the now-legendary GPT-3, showcased the rapid advancement in generative design. The model is able to create entirely new images from English specifications.
[Images generated by DALL-E with the specification “an armchair in the shape of an avocado”.]
So what else can it do? As multimodal transfer learning continues to advance, we could see real-world low-cost applications of generative design for engineering. Bespoke farming setups could be designed to minimize cost and maximize yield, customized to a given farm and its environment.
Logistics
We glossed over the logistics in agriculture -- mostly because it’s a gargantuan problem itself -- but there is an amazing opportunity to improve processes. Think Flexport, but for farms.
Other approaches, like VeChain, are addressing issues in robustness & reliability in global supply chains.
IV. Student Impact
The capital-intensive nature of AgTech can make it daunting for students to try to make an impact. However, this is becoming less and less true as low-capital tools like software and data play a more salient role.
For example, computer vision allows low-capital entrepreneurs to make highly differentiated models and analytics without ever having to build a sensor themselves. Logistical problems can be streamlined with software, and agricultural logistics
A Brown-founded company, Cloud Agronomics, recently raised a $6M seed round for their imaging, data collection, and machine learning inference platform. Machine learning models can build key, differentiated value on agricultural data, allowing farmers to understand the plentitude of data at an actionable level. Cloud Agronomics, for example, helps farmers understand “chemical fingerprint of a target property” in order to motivate usage of precision fertilizer.
Other satellite-based approaches have exploded as well. For example, fire prediction startup Fion was recently founded by a Clemson student to detect wildfires from satellite data.
A new non-profit, The Farmlink Project, was co-founded by Brown and Stanford students to address the massive food surplus in farms across America. The issue is endemic to market-based agriculture: farms consistently over-grow and have to destroy surplus food, yet markets underserve food-insecure populations. Farmlink, fueled by donations, solves the logistical challenges of sourcing and organizing transportation from farms to food banks.
If this post has shown you anything, I hope it’s that the problem space, markets, and potential solutions are fantastically vast. Make sure to consider the disparate and secondary impacts your technology could have, and know that the best practices and insights come from farmers themselves. And, of course, reach out to me if you’re building here.