Storm Warnings

Cloud computing’s hidden environmental and human costs

The computer industry generates buzzwords faster than an extroverted 20-year-old sends status updates, and “the cloud” is one of the more evocative ones. It is where your digital music, photos, e-books and personal documents live (or will live). No longer stored on a computer in a corner of the living room, Amazon and Apple and Microsoft now keep your stuff for you so that it can be fetched over the internet whenever you need it, right to whichever device you are using. It is the infrastructure that lets people flit from smartphone to tablet to e-reader to personal computer and have access to their data wherever they are.

To mix metaphors, there is a second string to the cloud’s bow. Amazon has always seen itself as a technology company rather than a bookseller, and was the first to grasp that it could make a commercial offering out of the infrastructure it uses to run its own global business. The last few years have seen a huge growth in major companies renting out access to computing resources in the form of individual computers as well as databases, messaging services, large-scale permanent data storage, and more.

Amazon’s computers rent out for less than 10 cents per hour, so computations demanding thousands of computers have become widely affordable for the first time. I am an occasional (and not expert) user of these services and like many others have come away marvelling at the power that is on offer for pennies at a time.

Amazon’s cloud services are now a big business in their own right. Its cloud data centres are relied on by many of this generation of start-ups, and ironically have been used by both Wikileaks and the CIA. They are even used by competitors such as Netflix, whose movies are streamed from Amazon’s computers. While Amazon remains several lengths ahead of the field, Microsoft, Google and others are spending massively on their cloud services to catch up.

Real clouds are nebulous and fuzzy, and the metaphor suggests that you do not need to worry about the concrete details of where and how all this computing is happening. But Vincent Mosco does worry, and the theme of To the Cloud: Big Data in a Turbulent World is the gap between the digital promise (what Mosco calls “the digital sublime”) and reality.

Mosco focuses on three “dark clouds” of the industry. One of those is the gap between our image of a pristine post-industrial technological world and the environmental reality. The cloud metaphor evokes an ethereal alternative to grimy industrial technologies. No forests are harmed in the making an e-book; no exhaust-spewing trucks are needed to deliver email. But cloud computing does have its own massive industrial infrastructure, even if most of us never see it. It has energy-intensive data centres, power-hungry computer networks, sprawling warehouses and a global workforce of millions who mine exotic minerals, assemble our gadgets and deliver online orders to our door.

Mosco starts with the “cloud” itself: the buildings where our data is stored. The size of today’s data centres is startling. The largest are run by the big internet companies and house hundreds of thousands of computers while consuming “more power than a medium-sized town.” And of course they are getting bigger all the time. Mosco highlights Chinese company Baidu, which is investing $1.6 billion to build the world’s largest data centre, where 700,000 computer chips will work tirelessly in a building the size of 15 football fields, making it one of the largest buildings in the world by some measures. And that is just phase one.

Locating these massive projects is becoming a political issue. Companies search far and wide for the right location and drive hard bargains to get cheap rates for power and water (for cooling) in return for their often much-needed investment. Rural locations with plentiful water and a cold climate have obvious appeal, and Canada is becoming a popular destination.

As Mosco describes, the tensions do not stop when the building is complete. Locals do not appreciate clouds of black smoke from diesel-powered backup generators, and companies haggle over contract disputes and fines. While some data centres run on renewable energy sources, Greenpeace has been leading efforts to “name and shame” companies with data centres that run on fossil fuels. This is a story that will become more important in the coming years.

Unfortunately, Mosco skims over the many advances in data centre design and efficiency, from plumbing to power distribution to low-power servers. These improvements lead proponents to claim that moving computing activity into cloud data centres cuts overall energy use, so major data centre companies position themselves as friends of the environment.

Mosco’s concern is with the other side of the efficiency coin: as computing gets cheaper, low prices drive increasing demand. Put the exploding power consumption together with the consumer side (charging all our gizmos, and the increasing power demands of always-on devices) as well as the power needed to deliver all that data over the networks themselves, and Mosco says the computing landscape will consume 30 percent of the world’s electrical grid in eight years’ time, and 45 percent within 20 years.

What will happen to overall power consumption as the tug-of-war continues between efficiency gains and increasing demand? Or the similar tension between the efficiency of sharing computing resources among many users and the requirements that those resources be continuously powered on, ready for use at all hours of every day?

Mosco is explicit about his critical agenda, setting out to highlight “the major problems associated with cloud computing.” He sees cloud computing as “a prism through which to view problems facing societies confronting the turbulent world of information technology.” In tackling the environmental story behind our sleek new computers, Mosco raises an important issue, not widely discussed in a trade book format. If he does not convince me that, as he quotes one power industry consultant, “it’s just not sustainable. They’re going to hit a brick wall,” he does at least convince me that there are problems here that will be the site of struggles and conflict in the coming years, and that makes the book a valuable contribution to the debate.

The scale of today’s computing infrastructure is new, but the inspirations behind cloud computing have a history that Mosco, a sociologist with a long career studying these developments, recounts ably. One vision is of computing as a utility, like gas or electricity, accessible where you want it and when you need it. Another is the promise of central planning made possible by smart algorithms and ubiquitous data collection. Not surprisingly, these visions have socialist origins: the USSR, the cybernetics project of Allende’s Chile and France’s Minitel service. It is ironic that these visions are now being realized by private industry.

The world of work is the second of Mosco’s dark clouds. The workforce of modern computing goes far beyond highly paid programmers in modern, light-filled buildings. The 1.4 million employees of Chinese gadget-maker Foxconn are nothing if not an industrial workforce, and there is nothing green about the mining of tin and tungsten, gold and tantalum, which are welded into our integrated circuits. Amazon’s warehouses are gaining reputations to match Walmart’s superstores as low-wage and high-pressure workplaces. There is even a new precarious workforce in the shape of a freelance contracting landscape where individuals can hire themselves out over a website to carry out more or less menial tasks.

Again, these are uncommon topics for a technology book (although Simon Head does cover overlapping ground in his recent Mindless: Why Smarter Machines Are Making Dumber Humans). Mosco highlights the conflicts arising around these new labour forces in one of the best parts of the book—the militancy of Chinese labour and the rise of an international trade unionism are particularly striking. The outcome of these conflicts may shape how technology’s role evolves. As with his environmental concerns, raising these issues is a timely reminder of continuity: old issues such as labour standards are resurfacing in new guises.

Not all Mosco’s subject choices are so fortunate. He devotes significant space to the language referring to the cloud. He investigates the selling and marketing of cloud offerings, and takes us through a reflection on the cloud as metaphor, from Aristophanes’ comedy The Clouds to a guide for 14th-century monks called The Cloud of Unknowing to David Mitchell’s novel Cloud Atlas.

The material is original, but for me it gives too much weight to the metaphor. Despite the occasional foray into mainstream advertising (the book’s title comes from a Microsoft TV ad), “cloud computing” remains mainly an industry phrase. Internet companies try to sell us on the services they offer, not the infrastructure behind it. Mosco’s take is that the cloud metaphor is used to sell an aspirational future, but more often than not such considerations are not mentioned at all in consumer advertising. Although he claims that “the cloud is an enormously powerful metaphor, arguably the most important developed in the short history of the IT world,” it may already be falling from favour to be replaced by “Big Data” (of which more below), the “Internet of Things” and more.

In his take on marketing and selling the cloud, Mosco focuses on the world of business computing: the information technology departments and computer systems that companies use to run their accounting, human resources, supply chain and so on. He is interested in the way industry analysts and others push companies to move their operations from their own data centres to the cloud—basically, to outsource and automate their IT operations. And he is concerned with the possible “demise of IT labor” if these efforts are successful.

I should acknowledge here that my employer is a large company that supplies business software and that is mentioned three times in this book (views expressed here are, of course, my own). But I am not here to defend the practices of business computing. I share To the Cloud’s suspicions about where digital technology is taking us. No, Mosco’s focus in these sections misses the bull’s-eye because personal computing, rather than business computing, has been setting the pace. Business computing demands the patching together of complex and disparate systems around the world in an environment with little room for error (accounts must balance, sales transactions must be recorded).

Business’s move to the cloud is not without consequences—retail companies will gather social media data and mine it in targeted marketing efforts for example—but it is playing second fiddle. The consumer world is, if anything, driving the business computing world in what is called, to use another buzzword, “the consumerization of IT.”

More on target is the material on privacy and related concerns, the third of Mosco’s dark clouds and also the subject of his final chapter. All that data about our location, our habits, our interests, is collected in those data centres, and the collectors find it irresistible to do things with it, whether it is selling it to advertisers or, in the case of the National Security Agency’s very own cloud data centre in Utah, using it for surveillance.

Beyond the whole issue of surveillance, Mosco fears the way that cloud computing brings with it particular ways of knowing, which are now collected under the umbrella of Big Data. Mosco is concerned with a growing “digital positivism,” with the consequence that “large data sets and massive computational power will … replace narrative with correlation and … ask only or mainly those questions that big data can handle.” He sees this happening in public policy, in business and in academia too, where the “digital humanities” and social media data analysis threaten to displace more established techniques and approaches.

Probably the most widely read introduction to Big Data is a 2013 book of the same name by Viktor Mayer-Schönberger and Kenneth Cukier. They claim that today’s huge data sets are often a substitute for traditional data collection and analysis. They also argue that, even if correlation is not causation, a focus on correlation is good enough for many purposes.

For example, when Walmart looked through its sales data it found that shoppers would stock up on Pop-Tarts just before a storm. The authors argue that it is not so important to understand why people do this, just that they do, and Walmart takes advantage of the knowledge by displaying Pop-Tarts conveniently near the checkout when a storm is forecast. Other prominent examples of Big Data analysis are Google’s use of a massive corpus of translated texts (instead of sophisticated grammatical analysis) in its machine translations, and the same company’s 2009 prediction of flu incidence in North American cities from the flu-related search queries that users submit.

These claims are provoking significant pushback from social scientists, and Mosco joins in with an effective critique, especially on the dangers of spurious correlations and overfitting. His arguments are bolstered by the fact that the Google flu experiment has succumbed to these problems: in more recent years its predictions of flu incidence were way off. As political scientist David Lazer has written, “the initial version of [Google Flu Trends] was part flu detector, part winter detector.”

As a broad critique of the technological tide from a sceptical standpoint I did prefer Astra Taylor’s The People’s Platform: And Other Digital Delusions, reviewed in these pages a couple of months ago. But despite its shortcomings, To the Cloud is a valuable contribution to digital debates, and Mosco’s decision to highlight emerging labour and environmental conflicts in the world of cloud computing promises to be farsighted.