Chapter 13 of 14

The Data Farm: The Digital Age

About 22 minutes to read

The experiment

For one week in January 2012, Facebook altered what 689,003 of its users saw when they opened the site. For some of them, the company's software quietly reduced the number of posts from friends that contained positive words. For others, it reduced the number of posts containing negative words. Then the company's researchers, working with academic collaborators at Cornell University, measured what the users themselves posted afterwards. The users whose feeds had been made more negative wrote slightly more negative posts; those whose feeds had been made more positive wrote slightly more positive ones. The effect was small, but it was there. Emotions, the researchers concluded, could spread through a social network without any direct contact between people — and the network could adjust the spread.

The study was published in June 2014 in the Proceedings of the National Academy of Sciences, under the title "Experimental evidence of massive-scale emotional contagion through social networks". It caused an uproar. None of the users had been told they were part of an experiment. The company pointed out that its terms of service, to which every user had agreed, permitted it to use their data for research. The journal published an "editorial expression of concern". Facebook said it would review its procedures. And then the world moved on, and the number of people using Facebook continued to grow.

The episode was not, in itself, especially sinister. Companies test changes to their products on their users all the time. What made it so revealing was its scale and its frankness. A single private company had the power to adjust the emotional climate experienced by hundreds of thousands of people, measure the results, and publish them — and almost none of those people knew or could have known. Three years later, Sean Parker, Facebook's first president, said in an interview that the platform had been built to consume as much of its users' time and conscious attention as possible, by exploiting what he called "a vulnerability in human psychology": the need for social validation. "God only knows what it's doing to our children's brains," he added.

In every previous chapter, the farm has taken something from the 99%: grain, labour, rent, tithes, human beings, hours, interest. This chapter is about the newest crop of all. It is harvested from billions of people every day, mostly without their noticing, and it has made a small number of people richer, faster, than anyone in history. The crop is attention, and the data it leaves behind.

You are the product

In 2010, a commenter on the website MetaFilter, writing under the name "blue_beetle" — a man named Andrew Lewis — summed up the business model of the internet in a sentence that has been repeated ever since: "If you are not paying for it, you're not the customer; you're the product being sold."

The largest digital platforms are, for the most part, free to use. Their revenue comes from advertising. In 2024, according to its annual results, Alphabet, the parent company of Google, earned about 350 billion dollars, of which around 265 billion came from advertising. Meta, the parent company of Facebook, Instagram and WhatsApp, earned about 164.5 billion dollars, of which about 160 billion — roughly ninety-eight per cent — came from advertising. More than three billion people use Meta's apps every day.

What these companies sell to advertisers is access to their users' attention — and, crucially, the ability to target that attention with great precision, based on detailed knowledge of who each user is and what he or she is likely to do. That knowledge comes from data: every search, every click, every "like", every video watched and for how long, every location visited, every purchase made, every message's metadata. The scholar Shoshana Zuboff, in The Age of Surveillance Capitalism (2019), argued that the decisive discovery was made at Google in the early 2000s, when the company realised that the data its users generated as a by-product of searching — which Zuboff called "behavioural surplus" — could be used to predict what they would click on, and that these predictions could be sold to advertisers. From that discovery, she argued, a new economic logic spread across the digital economy: the harvesting of human experience as raw material for prediction products.

The scale of the harvest is easy to underestimate because it is spread so thinly across so many lives. Market research firms estimate that the average internet user around the world now spends more than two hours a day on social media alone, and considerably longer on screens in total. Multiplied across billions of users, that is a quantity of human time comparable to the entire working hours of the world's largest industries. In Chapter 9, the factory took the worker's time and paid a wage for it. The data farm takes its users' time and pays them in the service itself — and then sells the time on.

One need not accept every element of Zuboff's argument to see its relevance to this book. The farm has always been interested in what its herd produces. The data farm is interested in what its herd does — in every moment of its attention, and every trace of its behaviour — and it has developed technologies to capture these with a thoroughness that William the Conqueror, counting every ox, cow and pig in England, could not have imagined.

Cultivating the crop

The attention of three billion people does not harvest itself. It has to be cultivated.

The digital platforms employ some of the most talented engineers, designers and behavioural scientists in the world to make their products as engaging as possible — which, in the industry's language, means maximising the time users spend on them and the frequency with which they return. The techniques are well documented. The "infinite scroll", designed by Aza Raskin in 2006, removed the natural stopping point at the bottom of a page; Raskin later said publicly that he regretted the time it had cost people. Notifications, "likes" and personalised feeds deliver rewards at unpredictable intervals, a pattern that behavioural psychologists since B. F. Skinner have known to be especially effective at sustaining behaviour. Recommendation algorithms learn, user by user, what keeps each person watching, and serve more of it.

Many of these techniques were adopted for ordinary commercial reasons, and many people value the products they produce. But the companies' own research has sometimes shown the costs. In 2021, documents leaked by a former Facebook employee, Frances Haugen, and published by the Wall Street Journal revealed that the company's internal researchers had studied the effects of Instagram on teenagers. One internal presentation stated that the platform made body-image issues worse for one in three teenage girls. Other internal documents showed that changes to Facebook's feed algorithm in 2018, intended to promote "meaningful social interactions", had rewarded outrage and divisive content. The company disputed the characterisation of its research, and the scientific evidence on social media and adolescent mental health remains genuinely contested. But the documents showed that the company knew more about the effects of its products than it had told the public.

This is the data farm's version of the factory discipline we met in Chapter 9. The mill owners designed the working day to extract the maximum labour from their workers' time. The platforms design their products to extract the maximum attention from their users' time. The difference is that the users are not paid, and most of them experience the process not as work but as leisure.

The fence of the network

What keeps users on the farm? In part, they stay because they want to. But they are also held by a powerful kind of fence: the network itself.

A social network is valuable to each user in proportion to how many other people use it. A messaging app is useless if your friends are not on it. A marketplace is valuable to sellers because buyers are there, and to buyers because sellers are there. These "network effects" mean that digital markets tend towards monopoly: once a platform has most of the users, it is very difficult for a rival to attract them away, however good its product. Each user is free, in principle, to leave. In practice, leaving means losing access to friends, family, customers, audiences or work. The fence is made of everyone else.

The platforms have reinforced this fence in other ways. They have bought potential competitors before those rivals could grow: Facebook bought Instagram in 2012 and WhatsApp in 2014. They have paid to be the default choice: in August 2024, a federal judge in Washington ruled in United States v. Google that Google had illegally maintained a monopoly in online search, in part through payments to Apple, Samsung and others to make Google the default search engine on their devices — payments that totalled more than twenty-six billion dollars in 2021 alone. The European Commission fined Google more than eight billion euros in three separate antitrust cases between 2017 and 2019. In 2017 a law student named Lina Khan published an article in the Yale Law Journal, "Amazon's Antitrust Paradox", arguing that American competition law had been blinded by its focus on consumer prices to the power of platforms that kept prices low while dominating whole markets. Four years later she was chair of the Federal Trade Commission.

The new shepherds

Who runs the data farm? Its shepherds are, first of all, the engineers, product managers and data scientists of the platforms themselves — a highly paid professional class recruited, as Chapter 12 described, overwhelmingly from a small number of elite universities. They design the systems through which the attention of billions is captured, measured and sold. Most of them believe, often sincerely, that their work is making the world better. Their predecessors in this book — the scribes, the stewards, the overseers, the scientific managers — usually believed the same.

But the data farm has also introduced something new: the algorithm as manager. On platforms such as Uber, Lyft, Deliveroo and DoorDash, and in the warehouses of Amazon, much of the work once done by human supervisors — assigning tasks, setting pay, monitoring performance, disciplining and dismissing workers — is now done by software. The sociologist Alex Rosenblat, who studied Uber drivers for her book Uberland (2018), described how the app nudged drivers with messages, adjusted their pay through opaque "surge" pricing, and could deactivate them — in effect, fire them — on the basis of customer ratings, often with little explanation or recourse. Frederick Taylor, who wanted to take the knowledge of how work was done away from the workers and give it to the planning department, would have recognised the system. The planning department is now a server farm.

The gig and the warehouse

The companies of the "gig economy" have built their business models on the claim that their workers are not workers at all. Uber, founded in 2009, insisted for years that its drivers were independent contractors — self-employed entrepreneurs using its app to find customers — and therefore not entitled to the minimum wage, holiday pay, sick pay or the right to organise a union. The story was one of freedom: be your own boss, choose your own hours.

Courts and legislatures have increasingly disagreed. In February 2021, the Supreme Court of the United Kingdom ruled unanimously, in Uber BV v Aslam, that Uber's drivers were "workers" entitled to basic employment rights, finding that Uber exercised a high degree of control over how they did their jobs. In California, the legislature passed a law in 2019, AB5, that would have classified most app-based drivers as employees. Uber, Lyft, DoorDash and other companies responded by spending more than two hundred million dollars on a ballot initiative, Proposition 22, to exempt their drivers from the law — the most expensive ballot campaign in the state's history. In November 2020 it passed. The farm, as so often, could outspend the herd.

Amazon, the second-largest private employer in the United States, shows how the data farm and the old factory farm have merged. In its warehouses, workers' every movement is tracked by scanners and software that measure their "rate" — the speed at which they pick and pack items — and their "time off task". A coalition of trade unions, the Strategic Organizing Center, analysing the injury data that Amazon is required to report to the federal government, found that in 2023 the serious injury rate in Amazon warehouses was around six per hundred workers, roughly twice the rate in other American warehouses; Amazon employed about a third of all US warehouse workers, but accounted for more than half of the industry's serious injuries. Amazon disputes the methodology and says that it has invested heavily in safety. But the pattern — the precise measurement of human motion, the relentless pace, the bodies breaking under it — is one that Elizabeth Bentley, who testified to Parliament in Chapter 9, would have recognised.

The Turk in the machine

In 1770 the Hungarian inventor Wolfgang von Kempelen unveiled at the court of the Empress Maria Theresa a machine that seemed to play chess: a life-sized figure in Ottoman costume, seated behind a cabinet full of gears, which defeated many of the players who challenged it on its tours of Europe and America. It was a hoax. Hidden inside the cabinet was a human chess master, moving the pieces from below. The machine became known as the Mechanical Turk.

In 2005 Amazon launched a service for businesses that needed small tasks done that computers could not yet do well — labelling images, transcribing audio, checking data. It farmed the tasks out to large numbers of people around the world, who were paid by the piece, often a few cents per task. Amazon named the service Mechanical Turk. The anthropologist Mary Gray and the computer scientist Siddharth Suri, in Ghost Work (2019), documented the vast, invisible workforce that performs such tasks for the technology industry: labelling the data on which artificial intelligence systems are trained, moderating the violent and abusive content that users post, and correcting the errors that automated systems make.

Much of this work is done in poorer countries for very low pay. In 2023, Time magazine reported that workers in Kenya employed by an outsourcing company to label violent, sexual and abusive text, so that OpenAI could train its systems to filter such content, had been paid less than two dollars an hour. Content moderators for social media platforms, who spend their working days viewing the worst material on the internet, have reported severe psychological harm; in 2020 Facebook agreed to pay fifty-two million dollars to settle a class-action lawsuit brought by moderators in the United States who had developed post-traumatic stress disorder.

In Chapter 4, Aristotle imagined that if shuttles could weave by themselves, masters would have no need of slaves. The shuttles have begun to weave. But, as with Kempelen's Turk, there are still people inside the machine — and they are among the least visible and lowest paid workers on the data farm.

The hinterland of the cloud

The data farm likes to describe itself in weightless language — the cloud, the network, the virtual. But it rests on a physical hinterland that stretches across the world, and much of it would be familiar from earlier chapters of this book.

Every smartphone, laptop and electric car depends on rechargeable lithium-ion batteries, and most of those batteries contain cobalt. According to the United States Geological Survey, around three-quarters of the world's mined cobalt comes from the Democratic Republic of the Congo. Much of it is produced by large industrial mines, but a significant share comes from "artisanal" mining, in which men, women and — according to UNICEF's estimate in 2014 — tens of thousands of children dig by hand in dangerous, unregulated pits for a few dollars a day. The researcher Siddharth Kara, in Cobalt Red (2023), documented collapsing tunnels, toxic exposure and the chains of traders through which this cobalt reaches the supply chains of some of the most valuable companies in the world. The technology companies say they audit their supply chains and prohibit child labour, and several have invested in efforts to improve conditions. But the pattern is an old one. In Chapter 1 I mentioned, as an example of the variety within the 99%, a cobalt miner in the Congo. The device on which a reader in London or San Francisco scrolls through a social media feed may well contain the product of that miner's labour. The farm of the Atlantic age took sugar and cotton from the colonies to feed the factories of Europe. The data farm takes cobalt and coltan from Central Africa to power the devices of the rich world.

The data farm's harvest of attention, too, extends across the globe — and its costs are often borne most heavily by those with least power to resist. In Myanmar in 2017, the military launched a campaign of mass killing, rape and arson against the Rohingya Muslim minority that drove more than seven hundred thousand people into Bangladesh. In the years before, Facebook had become, for many people in Myanmar, effectively synonymous with the internet, and it had been used to spread hate speech and incitement against the Rohingya on a vast scale. In March 2018, the chair of a United Nations fact-finding mission said that social media had played a "determining role" in the crisis. Later that year, an independent assessment commissioned by Facebook concluded that the company had not done enough to prevent its platform from being used to foment division and incite offline violence, and the company acknowledged that it had been too slow to act. It had employed very few moderators who spoke Burmese.

The scholars Nick Couldry and Ulises Mejias, in The Costs of Connection (2019), proposed the term "data colonialism" for the way in which digital platforms, mostly based in the United States and China, appropriate the data of people around the world as a free resource, much as earlier empires appropriated land and labour. The comparison can be pushed too far: nobody is forced onto a smartphone at gunpoint, and the benefits of connection are real, especially for people in poor countries. But the flows are clear. The data, the attention and the raw materials flow from the many to the few, and from the periphery to the centre. The profits flow the same way.

Surveillance

The data farm does not only sell attention to advertisers. It also builds an archive of human behaviour of unprecedented scope, and that archive is available to more than advertisers.

In June 2013 Edward Snowden, a contractor working for the US National Security Agency, leaked documents showing that the agency had been collecting the telephone records of millions of Americans and, through a programme called PRISM, obtaining data from major internet companies. In 2018 it emerged that a political consulting firm, Cambridge Analytica, had obtained data on up to eighty-seven million Facebook users, harvested through a personality quiz app, and used it in political campaigns. In 2019 the US Federal Trade Commission fined Facebook five billion dollars for privacy violations — the largest such penalty in its history at the time. Employers, meanwhile, increasingly use software to monitor their employees' keystrokes, screens, locations and even facial expressions.

Surveillance is not new to the farm. Domesday Book, the Ming Yellow Registers and the Ottoman tax surveys were all instruments for knowing the herd. What is new is the granularity and the ownership. The medieval king knew how many oxen a village had. The data farm can know where each person is, whom they talk to, what they read, what they buy and what they are likely to do next — and that knowledge is held, in the first instance, not by states but by a handful of private companies.

The new 0.01%

The data farm has created fortunes on a scale and at a speed without precedent.

On Forbes's list of the world's billionaires published in April 2025, the richest person in the world was Elon Musk, with an estimated fortune of 342 billion dollars, built on Tesla and SpaceX. Most of the others at the very top of the list owed their fortunes to technology: the founders of Amazon, Meta, Google and Oracle among them. The total wealth of the world's 3,028 billionaires was estimated at 16.1 trillion dollars. As Chapter 1 noted, the World Inequality Report found that billionaires' share of global household wealth had risen from about one per cent in 1995 to more than three per cent by 2021.

Several features of these fortunes deserve attention. The first is their concentration of control. Many technology companies have issued two classes of shares, so that their founders keep control of the company even after selling most of it to outside investors. Mark Zuckerberg, for example, holds a minority of Meta's shares but controls a majority of its voting power. The owner of the farm can sell the herd and keep the farm.

The second is their relationship to the rest of the economy. Economists including David Autor and his colleagues have documented the rise of "superstar firms" — a small number of highly profitable companies that dominate their industries — and linked it to a decline in the share of national income going to workers. The digital platforms are the clearest examples: they generate enormous revenue with relatively few employees, and their profits flow overwhelmingly to their shareholders.

The third is taxation. In 2021, the investigative news organisation ProPublica obtained a vast trove of tax data on the wealthiest Americans. It reported that the twenty-five richest Americans had seen their wealth grow by a combined 401 billion dollars between 2014 and 2018, while paying a total of 13.6 billion dollars in federal income tax over those years — a "true tax rate", measured against the growth of their wealth, of 3.4 per cent. Jeff Bezos, the records showed, had paid no federal income tax at all in 2007 and 2011; Elon Musk had paid none in 2018. None of this was illegal. Under American law, the rising value of shares is not taxed as income until they are sold, and the very rich can borrow against their holdings rather than selling them. The ordinary worker's wage is taxed before it reaches her bank account. The billionaire's wealth grows untaxed until he chooses to realise it — which, if he is careful, may be never.

The story

The data farm's story is one of the most attractive in this book. Its slogans have entered the language. Google's early code of conduct urged its employees: "Don't be evil." Facebook's mission was to "make the world more open and connected", and its internal motto was "move fast and break things". The technology industry presented itself as the force of progress, disrupting old monopolies, connecting the world, democratising information, empowering individuals. Its products were free. Its founders were young outsiders in hoodies, not bankers in suits.

Much of this was true, at least at first. The internet did connect the world, and much of what it made possible — access to knowledge, communication across distance, the ability of ordinary people to publish and organise — has been genuinely liberating. But the story also performed the familiar functions. It presented the concentration of wealth and power in a few companies as the natural reward of innovation and genius, the modern form of the meritocratic theodicy described in the last chapter. It presented the gig worker's precarity as freedom. It presented surveillance as personalisation. And it presented each new technology, including artificial intelligence, as an inevitable force to which society must adapt, rather than a set of choices made by particular people in their own interests.

That last element matters now more than ever. In 1930, as the Great Depression began, John Maynard Keynes wrote an essay called "Economic Possibilities for our Grandchildren". Within a hundred years, he predicted, technological progress would make humanity so productive that his grandchildren's generation would need to work only fifteen hours a week, and the great problem of humanity would be how to use its leisure. The productivity growth Keynes expected has largely happened. The fifteen-hour week has not. Whether the new wave of artificial intelligence — trained, it should be noted, on the writing, images and conversations of billions of people, many of whom are now suing over the use of their work, as in the New York Times's case against OpenAI and Microsoft, and the authors' class action that ended in 2025 with a 1.5-billion-dollar settlement by Anthropic — will free people from labour or simply concentrate its rewards further is not a question of technology. It is a question of who owns the machines. Aristotle saw that in the fourth century BCE.

Breakouts

The data farm, like every farm before it, has met resistance, and some of it has succeeded.

In 2018 the European Union's General Data Protection Regulation came into force, giving people rights over their personal data and allowing regulators to fine companies up to four per cent of their global turnover. The EU's Digital Markets Act, which began to apply in 2023 and 2024, imposed new obligations on the largest platforms. In the United States, the Google search ruling of 2024 was the most significant monopoly judgement against a technology company in a generation.

Workers have organised too. In April 2022, workers at Amazon's JFK8 warehouse on Staten Island, led by a former employee named Christian Smalls who had been fired after protesting about safety conditions during the Covid-19 pandemic, voted to form the Amazon Labor Union — the first successful union election at an Amazon facility in the United States. In 2023, Hollywood screenwriters went on strike for 148 days and won, among other things, protections against the use of artificial intelligence to replace their work; actors followed with a strike of their own. Gig workers have won rights through courts and legislation in Britain, Spain and elsewhere, even as they lost in California.

Some of the most interesting breakouts have come from attempts to reclaim time itself. In 2017 France introduced a legal "right to disconnect", requiring larger employers to negotiate rules limiting the expectation that employees answer emails and messages outside working hours; similar measures have since been adopted in several other countries, including Portugal and Australia. It is a small law, but it addresses the oldest question in this book's account of the factory: who owns the worker's time? The ten-hour day of 1847 drew a line around the working day. The right to disconnect tries to redraw that line in an age when the phone in a worker's pocket can reach her at any hour.

Governments have also begun to test whether the platforms can be made to share the value they capture. In 2021 Australia passed a law requiring the largest platforms to negotiate payments with news publishers whose journalism appeared on their services. Facebook responded by briefly blocking all news for Australian users — including, for a few days, the pages of health services and emergency agencies — before reaching a compromise with the government. The episode showed both the platforms' power and its limits: a single company could switch off the news for a whole country, but it could not, in the end, ignore that country's law.

Others have tried to build platforms owned by the people who work on them. The scholar Trebor Scholz has given the movement a name — "platform cooperativism" — and examples range from driver-owned ride-hailing cooperatives to cooperatively owned stock-photo and music-streaming services. Most remain small. But they are the digital descendants of the cooperative stores and mutual societies that nineteenth-century workers built in the shadow of the factory, and of Meidner's wage-earner funds in Sweden: attempts to change not only how the farm treats its herd, but who owns it.

And the digital world has also produced its own commons. Wikipedia, founded in 2001, became one of the most visited websites in the world while being written, edited and governed by volunteers and owned by a non-profit foundation. Open-source software, built collaboratively and shared freely, runs much of the internet's infrastructure. These are, in the terms of Chapter 7, commons — shared resources governed by their communities, as Elinor Ostrom described. They show that the digital economy does not have to be a farm.

What the data farm teaches

The fence is the network itself: platforms whose value to each user depends on everyone else being there, reinforced by acquisitions, default agreements and lock-in; and, for workers, the algorithm that assigns, measures and dismisses.

The feed is attention and data — the hours and the behaviour of billions of people, harvested continuously and sold as prediction — together with the labour of gig workers, warehouse workers and the invisible "ghost workers" who keep the machines running.

The shepherds are the engineers and data scientists who design the systems, and increasingly the systems themselves: the algorithm as manager, carrying Taylor's scientific management to its logical conclusion.

The story is progress, connection, disruption and freedom: free products, flexible work, the genius founder and the inevitability of technology.

The breakouts are data protection and competition law, the first unions at Amazon, the courts that have recognised gig workers as workers, the strikes that won limits on artificial intelligence — and the digital commons, from Wikipedia to open-source software, that show another way is possible.

Bridge: the grandchildren

We are, almost exactly, the grandchildren Keynes was writing about in 1930. The productive power he predicted exists. The leisure he predicted does not, and the wealth that power has produced has been distributed, as this chapter and the three before it have shown, in a way that would have looked familiar to the temple accountant of Uruk with whom this book began.

That leaves one question, and it is the one the final chapter must try to answer. Across five thousand years, the farm has been challenged thousands of times. Most challenges failed. Some succeeded, at least for a while. What, in the end, has actually changed the ratio between the 0.01, the 1 and the 99 — and what, if anything, could change it again?