Blockchain Chicken Farm

by Xiaowei Wang

  4.

  Human farmers are inefficient in an optimized world. Human farmers are subject to “bounded rationality.” The term was coined by the economist Herbert Simon (who also coined the word “satisficing”), and it describes how individuals are subject to information and time constraints in decision-making. These constraints have enormous impact when you’re talking about unanticipated weather events that are only set to increase under climate change. So why not replace the farmers with AI models, which have access to endless data and computation time?

  Alibaba is proposing just that, with its ET Agricultural Brain—a hulking new product that uses AI to transform agriculture in order to help create China’s pork miracle.

  On a gray, chilly day in Hangzhou, I visit Alibaba Cloud to try to understand the company’s pledge of using artificial intelligence to help raise pigs in partnership with the Sichuan-based Tequ Group, a sprawling food company with a focus on industrial agriculture. Tequ had pork-yield plans of ten million pigs by 2020 (though they were stymied by failure to contain ASF and labor disruptions from COVID-19). Alibaba Cloud’s new campus is a half hour outside the city center, in a place called Cloud Town. The lush green setting reminds me of the Amazon Web Services (AWS) campus in Seattle, including the rain that occasionally pours down in sheets.

  The main building is generic, just like the main AWS building, with gray carpets, convenient beverage fridges, and uninspired office furniture. In the main lobby is the Cloud Computing Museum, showcasing Alibaba Cloud’s technical achievements of the past decade, which parallel AWS’s trajectory.

  Online shopping has been the biggest catalyst for innovation over the past twenty years. It’s because of online shopping that we have targeted ads, recommendation algorithms, hypnotic social media, and, of course, technical infrastructure for rent from Alibaba Cloud (Aliyun) and AWS. Both these companies started off as e-commerce companies. They leverage shopping lulls on their own platforms to rent out computers, or servers that they aren’t using, making money off their unneeded computing power.

  Despite increased automation, online shopping still requires legions of engineers. Even the speedy fetching of a high-resolution color image of a product is the result of years of technical innovation. Early e-commerce websites were pretty simple, some text and an image or two. Websites today are increasingly more complex, loaded with 3D videos of products, countless images, interactivity, and algorithms that suggest other products for you. And for a small tech company, instead of running your own servers to host your complex website, renting servers from Aliyun or AWS makes more sense. These costs can balloon into the millions for startups, allowing Alibaba Cloud and AWS to make enormous profits from renting out excess server space. Increasingly, Aliyun and AWS also rent out other tools they’ve developed internally—voice-recognition tools, satellite imagery, prebuilt AI models. These timeshare computing setups were just as important as venture capital funding in creating the late-2010s tech boom in the United States and China—it’s estimated that 40 to 60 percent of all traceable internet traffic now comes from a rented cloud server.

  The first wall display at the Cloud Computing Museum describes the platform’s initial technical setup. In the early 2000s, a handful of computers sat in the Hangzhou office running Alibaba.com and Taobao.com. While Alibaba.com connected the rest of the world to China’s bulk sellers, Taobao.com is a consumer-oriented online shopping platform that is now twice the size of Amazon.

  The early Alibaba systems borrowed other people’s technology, says the panel, including Oracle databases. The next panel shows a photo taken in 2009 of the smiling faces of Jack Ma and a few engineers holding a computer. The old Oracle databases were replaced by Alibaba’s own framework, the Apsara framework, named after the Buddhist goddess of clouds. A towering server sits behind a glass pane, with a printed poem: CODE / LINE BY LINE / BUILDS THE FOUNDATION / FOR ETERNITY / JUST LIKE SAND / GRAIN BY GRAIN / CALMS THE ROARING SEA. Opposite the server are the first lines of code ever run on the Apsara system, configuring logging, heroically presented under a spotlight: Void InitLoggingSystem(conststd::string&configFiles="").

  In a brightly lit area painted white, with a half-dead orchid, I sit with Jintong, a stoic Aliyun expert. A few glass-walled conference rooms are down the hall. Jintong tells me that raising pigs using AI was a natural opportunity. The farming structure was already in place; Aliyun just helped optimize it.

  Large pork farms already have closed-circuit televisions and sensors, monitored by humans. For a few hundred pigs, a human might do reasonably well overseeing operations. But for hundreds of thousands of pigs, where do you even begin? And in order for China to achieve its pork miracle, millions of pigs must be farmed.

  Aliyun offers a way to help sort through data using AI. In these large-scale farms, pigs are stamped with a unique identity mark on their bodies, similar to a QR code. That data is fed into a model made by Alibaba, and the model has the information it needs to monitor the pigs in real time, using video, temperature, and sound sensors. It’s through these channels that the model detects any sudden signs of fever or disease, or if pigs are crushing one another in their pens. If something does happen, the system recognizes the unique identifier on the pig’s body and gives an alert.

  Certain machine-learning models, like the one used by ET Agricultural Brain, require massive amounts of training data in order to work. It’s only after collecting three months’ worth of training data (where cameras sit and record data, without analysis) that the AI model is actually useful. Only then can it be effective in diagnosis.

  Jintong explains that, beyond detecting porcine disease, ET Agricultural Brain makes decisions based on data, and offers a precision that is beyond human capacity. ET Agricultural Brain is like a Swiss Army knife of models—these models are fed training data from specific clients, big industrial farms that raise pigs and grow melons, or even agricultural drone companies like XAG, which it helps crunch through sensor data to finesse autopilot capabilities. It can determine the best time to plant, based on the weather, or when to pick fruit for optimal sweetness. ET Agricultural Brain also conveniently plugs into Aliyun’s other offerings, like ET Logistics Brain, which can perform complex calculations on the cold chain during food delivery. The problem Matilda posed would be gone with ET Logistics Brain, which would calculate the amount of refrigeration a truck driver had used by sensing how much gas was left in the tank by the end of a trip.

  And where are all the human farmers in this scheme? Are they relaxing, eating peanuts as the machines do all the work?

  It turns out that humans are still needed. Aliyun works with farmers to formalize their knowledge for the machine-learning models through the Alibaba Knowledge Graph. ET Agricultural Brain can “see,” but that is a generous term, given how much effort had to be put into teaching it, and how it can see only a limited set of objects.

  But the payoff is enormous—the production of millions of pigs at a low price. Jintong is optimistic that trickle-down innovation can happen. He believes that “dragon-head” agricultural companies, large national conglomerates that rely on a network of smaller farmers, will share certain innovations with their small farmers.

  Given the computation time, the data required, the hardware infrastructure needed, and the cost, it currently makes sense to utilize AI only if you are raising millions of pigs, not just one or two. Other companies are also trying to cash in on the AI pork-farming business, using technologies like pig facial recognition.

  The logic is striking. A demand for pork drives industrialized farming of pigs, which increases disease transmission. The constant emergence of diseases drives the implementation of new technologies like AI pork farming. These technologies go on to make pork cheap, driving even more availability and demand, as people start to believe pork is a necessary part of their diet. AI is not the balm to any problem—it is just one piece of the ever-hungry quest for scale.

  5.

  If pig life can be optim
ized through gene editing and automation, can human life be optimized as well? The concept that human life can be optimized, of human actions being calibrated toward better performance, is a central belief of the ET Agricultural Brain project: it may eventually replace human farmers with AI farmers.

  The optimization of life is a distinctly modern endeavor. Some proponents of a world run by artificial intelligence (AI, when a computer program can perform defined tasks as well as humans can) and artificial general intelligence (AGI, computers more powerful than AI, with the ability to understand the world as well as humans can) present an optimized version of human life that is very seductive: rational, error-proof, and objective. Others have similar convictions: if we can quantify human consciousness and emotions through mechanisms like AI, we might be able to reduce suffering by optimizing our world to decrease those emotions. One machine-learning engineer I met at a tech salon in San Francisco eagerly described the dawn of this AI world, one without the “clumsy irrationality of meat machines.” AI would teach humans how to live ethically and in accordance with reason. “Just imagine,” he said, lowering his voice to a hushed tone. He sat uncomfortably close to me, holding a Fibonacci sequence–inspired cocktail, eyes cast intently at my face. “No more irrational things like sexism,” he whispered.

  Artificial intelligence is a broad category, and that broadness makes it susceptible to slippery usages, to being malleable to any kind of political or economic end. AI is technically a subset of machine learning. And within artificial intelligence, one of the most exciting areas over the past ten years has been work done on neural networks, which are used in deep learning. These artificial neural networks rely on models of the brain that have been formalized into mathematical operations. Research into these “artificial neurons” began as early as 1943, with a paper by Warren McCulloch and Walter Pitts on the perceptron, an algorithm that modeled binary (yes/no) classification, which would serve as the foundation of contemporary neural networks.

  Yet the neural networks of today’s AI haven’t caught up to the latest neuroscience research on how our brains function and process information. And the way brains learn and encode information are still emerging areas of research. One theoretical neuroscientist I spoke to, Ashok Litwin-Kumar, explained that studies and experiments on animal brains are still being done in order to understand more complex, generative brain functions—like constructing new meanings and relationships, or interpreting new experiences. Neurons can be artificially “created” and modeled on a computer, but we still do not know how to regenerate human neurons once they die off. While artificial neural networks often assume there are only a few types of neurons, human neural networks consist of thousands of different types scattered across the body, existing even in places like the stomach. Just replicating a single brain using computer neural networks doesn’t guarantee an exact mimicking of brain function. After all, the process of learning doesn’t reside solely within our brains; it’s environmental, physical, and, most of all, social, carried out through interaction and dialogue.

  The seduction of AI is already palpable in China and the United States, across the political spectrum, as people advocate for a fully automated world. The attraction is not simply about rationality and the level of control provided by making systems automated. It’s also about scale: once implemented, certain applications of deep learning, like image recognition, have been shown to be faster and more accurate than humans. It’s no surprise that these qualities make AI the ideal worker.

  Many of us live in a world where machine learning and forms of artificial intelligence already pervade our everyday lives—recommendation algorithms, fun cosmetic and face filters on Snapchat and Meitu, automated checkouts using image-recognition cameras. Since “artificial intelligence” is a vague term, it has become a catchall to instill deep fear of a blurry future. Some radical proponents of AI claim we are on “the edge of a revolution driven by artificial intelligence.”4 These same proponents of the AI revolution espouse the belief that this optimized version of human life will take over, replacing humans in the workplace, as caregivers, or even in romantic relationships. “Artificial” will no longer sit in the term as a dirty caveat. AI will farm greenhouses with data-based decision-making, will drive better, with fewer accidents; AI will make sandwiches and pack boxes. AI will do all this without complaining or needing to sleep.

  The philosopher and theorist Sylvia Wynter writes, “The struggle of our new millennium will be one between the ongoing imperative of securing the well-being of our present ethnoclass (i.e., Western bourgeois) conception of the human, Man.”5 Her work deconstructs the way “human” was created as a category. This concept of the “human” was tweaked throughout history to serve the projects of colonialism, slavery, racism, and subjugation. Through religious and economic institutions, the idea of who is considered human and what it means to be human has for hundreds of years been a political project by those in power. Wynter gives the example of colonial subjects and slaves being designated as nonhuman, with submission leading to salvation, allowing “inferior subjects” to become human.

  I see the myth of automation replacing humans as yet another attempt by those in power to sharply define the boundaries of what being human means, elevating AI to a form of power that seems to have a righteous, natural force in our lives. This myth defines being human as simply being a rational, efficient worker. The fear instilled by these radical proponents of AI is ominous and forceful, and it implies an inevitability written by those in charge—leaders in the tech world, owners of companies that are building this scary AI. The same fear of automation drives a public discourse that glints with a subterfuge: that being human is the only thing that makes us special.

  The project of making AI a natural, evolutionary force continues. In this state of optimized life, we are told humans will be free from work. Silicon Valley claims it has anticipated this mass unemployment by automation, with places like Y Combinator piloting universal basic income programs. Individuals would get a monthly stipend to pay rent and purchase things, keeping a consumer-driven economy afloat. The promise being advertised to us about an AI labor force is that we will be free, and we will also be able to optimize our own tiny human lives—maybe for freedom, for true happiness.

  6.

  On a subway ride about an hour away from the center of Shanghai, I’ve struck up a conversation with a kind stranger. I’ve left Hangzhou, and am headed back to Shanghai for a few days, stopping in villages along the way. Most of my time is spent like this—countryside trips buffered by stops in cities, where I gorge on meals that cost as much as a few months’ income for a farmer.

  My new acquaintance, Shan, and I sit under neon lights, entranced, watching a video on the subway TV showing how to cook red braised pork. Screens are unavoidable in contemporary Chinese life—they proliferate everywhere, as rampant as the video cameras that are always recording, always watching. For every video camera in a public place, for every surveillance lens watching you, there’s a mirror, a screen placed for you to watch ads, cartoons, and news in a hypnotic glaze of content. More than a government conspiracy of surveillance, it ends up feeling like a hardware conspiracy to sell as many video cameras and screens as possible.

  Shan is only a few years older than me and has a fifteen-year-old daughter. Shan lives in the center of Shanghai but commutes to the outskirts every day for her job as a database administrator at a motor factory. She’s not Christian but she’s taking the rest of the day off to make Christmas Eve dinner, for the holiday spirit. She might even make red braised pork.

  “Honestly though, over the past few years, Christmas has come and gone. For a while we all celebrated it, even though none of us would call ourselves Christian. But it’s different this past year. The government has been seeing it as a Western influence, a religious influence, so you know, they are trying to tamp it down.” As rationality and control pervade everyday life in urban China, as life becomes optimized, religion is makin
g a resurgence. Faith can take on new significance in the suspended, static realm of everyday urban life.

  In an early work, Understanding Computers and Cognition, the computer scientists and AI pioneers Terry Winograd and Fernando Flores point out our tendency to ascribe rationality to computers. We do this when a physical system is “so complex, and yet so organized, that we find it convenient, explanatory, pragmatically necessary for prediction, to treat it as if it has beliefs and desires and was rational.”

  But neither computers nor humans are rational actors—and this is not a problem. They continue, “We treat other people not as merely ‘rational beings’ but as ‘responsible beings.’ An essential part of being human is the ability to enter into commitments and to be responsible for the courses of action that they anticipate. A computer can never enter into a commitment.”6

  The version of life under AI being sold by big tech companies presents a reassuring, controlled world, where unfettered optimization and automation are inevitable. Those in control, those who built the closed systems of control, unsurprisingly purport to predict the future. As ASF and the number of emerging pathogens climb, it becomes obvious that it is impossible to predict the future because we live in an open system. The imperative for these companies then becomes creating a tighter control loop over all the variables that might exist, making an ever more claustrophobic system. This is the self-fulfilling prophecy of life in an AI world: a static, closed world.