The People's Republic of Walmart

by Leigh Phillips

  Uber’s drivers, on the other hand, are poorly-paid “contractors.” No longer classified as workers (except in the UK where courts reinstated their rights as workers), they can make below minimum wage and have few labor rights. As with more and more workers in a range of sectors, they are under constant, nigh on panoptical, surveillance via data. Uber uses a five-star driver rating system in which drivers must maintain an average rating of 4.6 stars to keep driving for the company. Uber can “suggest” certain norms for its drivers to follow (how much to smile, what kind of extra services to offer, and so on), but in reality it is the risk of even one bad rating that quickly prods them to fall into line. Yet there is no top-down rule; when businesses can constantly collect and analyze information, strict management happens from the bottom up. Uber’s business model is to use the economics of information to do more than just sell ad space. The company’s ability to make people do things without telling them explicitly is not unique and is but a refinement of capitalism’s ability to make people complicit in their own unfreedom—a refinement made possible by a greater amount of and greater control over information.

  On the other hand, rather than the herald of dystopian workplaces everywhere, Uber is also a natural candidate for a worker co-op. All that Uber provides, after all, is an app; the company is nothing but a middleman. A cooperatively owned network of drivers using a similar app could set pay rates and work rules democratically, in the here and now. A drivers’ co-op would be far superior to the venture capital–fueled behemoth we have today, even if this is a form of enterprise that, while introducing more workplace democracy than is normally possible under capitalism in the short run, is still subject to the same profit-seeking imperatives as any firm within capitalism—imperatives that will prompt self-exploitation in order to compete with other enterprises, thus ultimately undermining these very same democratic impulses.

  Similarly, social networks could be run as public utilities rather than as private monopolies—remember that we created public electricity or water works after the failures of nineteenth-century robber baron capitalism. One of the big questions of the twenty-first century will be, who owns and controls the data that is quickly becoming a key economic resource? Will it be the fuel for democratic planning, or instead for a new more authoritarian capitalism? These questions require that we recognize the immense challenges posed by data-driven twenty-first-century capitalism: How could we nationalize multinational corporations that span and disregard national borders, and often play jurisdictions off of one another? How would we ensure privacy with so much data under collective, state control?

  Privately held data is making possible more efficient production, but at the same time it is enabling closer supervision, and modern corporate planning is only starting to take advantage of all this newly available information. One outcome is illusory freedom for workers. If we constantly produce information both at and outside of work, we don’t need to be supervised so directly—but the boss is still watching, and doing so more closely than has ever before been possible. Data and metrics speak for themselves: managers can see how many parts a worker assembled per minute or how many packages a driver delivered per hour.

  Increasing self-management at work—ostensibly without managers, but still closely surveilled—is a symptom of bigger changes. As wages, both in the United States and across much of the global North, have grown at glacial pace or outright stagnated since the 1970s, workers have taken on more personal debt just to keep up. At the same time, governments have cut public benefits, leaving workers more vulnerable when they are laid off or injured. Even Alan Greenspan, the former head of the US Federal Reserve, called today’s workers “traumatized.” Translated, this means that pressures to fall into line now exist outside the explicit top-down hierarchies.

  Capitalism is stuck with planning even though it regularly transmogrifies its techniques of planning. Today, capitalist planning exists both in the old, hierarchical sense that Coase studied as well as in new, more roundabout ways that take cues from the economics of information.

  Opening the Gates to the Future

  There’s an old quip among historians of economics that a PhD-level microeconomics textbook from the 1960s could be mistaken for a textbook at the department of planning at a university in Havana. In the microeconomics textbook, the free market generates the prices that dictate how much of everything is produced and how things are distributed; in the planning textbook, a planner solves the same equations by coming up with the equivalent proportions of production and distribution. Oskar Lange’s version of socialism and the economic orthodoxy of the twentieth century shared the same flawed assumptions. Over time, as outlined in this chapter, many poked holes in these assumptions: Markets are costly, said Coase. Human beings are not infinitely powered, all-knowing calculators, argued Stiglitz. Even Hayek was right: capitalism is dynamic, not static, and rarely in the sort of equilibrium imagined by Lange and conventional economics.

  But the economics of information also challenges Hayek’s counterargument to Lange, that the market is the only means we have to produce all the information that planning would require in the first place. For markets sometimes fail to discover the right information, and that which they do reveal can be false. Also, the enormous amount of economic activity that continues to take place outside the market—within the black boxes we call Walmart or Amazon or General Motors—is evidence against Hayek. At the same time, the rise of information technology shows just how much information it is possible to have at our fingertips. Hayek describes prices as “a system of telecommunications”; today, we have telecommunications far more precise and powerful that can communicate information directly without it being mediated by prices. Hayek’s arguments may have worked to disarm some of Lange’s vision for planning, but they shouldn’t stop contemporary socialists from arguing for democratic planning that is also a process of discovery.

  Economics, if it is to be of use to those who desire an egalitarian society, needs to leave behind fantasy worlds. Paul Samuelson, one of the most influential mainstream economists of the postwar era and author of the economics textbook used in most graduate programs from the 1950s well into the 1970s, observed that in the idealized vision that animated both sides of the calculation debate, it doesn’t matter whether capital hires labor, or labor hires capital. The dense web of abstractions completely obscures what it means to be a boss or a worker, an owner of resources or an owner of a body and mind that can be put to work for a wage.

  The economist Duncan Foley describes this lacuna in the calculation debate: “The real import of the historical social choice between socialism and capitalism is precisely what is left out of the socialist calculation debate: the social relations through which people organize themselves to produce.” When we say that we are interested in how things are distributed, we mean that we are interested in how society is organized. Who makes the orders, and who follows them? What counts as “work,” and what is part of the household? Who rears the children, and who does the dishes?

  These are only some of the big questions with which any economics of equality will have to grapple. Planning is not only possible, but is already all around us, albeit in hierarchical and undemocratic forms. What a very different, democratic planning will look like is a question a new generation of progressive economists needs to begin today to discuss, debate and test through modeling.

  But to the question of whether information should be discovered and created via a system that inevitably creates huge social disparities while depriving a majority of people of a say in how they work, or rather via one of democratic deliberation that fosters equality, the answer should be obvious.

  4

  MAPPING THE AMAZON

  Amazon is on its way to developing psychic powers. Or at least, such was the fantasy that one could be forgiven for believing, based on the hosanna-filled, adrenalized newspaper column inches that appeared in the summer of 2014 when the online bookseller-turned-“ev
erything store” filed a patent application for a new process it called “anticipatory shipping.” Amazon would soon know what you wanted to buy before you knew it yourself. When you placed an order for the latest John Green young adult novel for non–young adults, another jar of artisanally brined lupini beans, or that Instant Pot wonder–pressure cooker that produces pulled pork faster than the speed of light, the package would already be on its way.

  As those journalists less prone to the confection of hyperbolic clickbait pointed out at the time, what this patent describes is in truth a very small step from what Amazon already does. It is a minor extension of the kind of data the company already collects and of the colossal, tentacular logistics operation it already runs. Amazon, building its retail market position on the back of the internet revolution, is the largest technology company using the fruits of modern IT to distribute consumer goods. In short, Amazon is a master planner. It is these sorts of logistical and algorithmic innovations that give the lie to the hoary free market argument that even if planning can deliver the big stuff like steel foundries and railways and healthcare, it would stumble at the first hurdle of planning for consumer items. A fortiori, Amazon offers techniques of production and distribution that are just waiting to be seized and repurposed.

  What Amazon Plans

  Since its late-’90s dot-com beginnings selling only books, Amazon has expanded to potentially fulfill a large part of a household’s everyday consumption. Echoing Walmart’s horizontal integration, the company has even started to incorporate producers of the things it sells into its distribution network by placing its own workers at the factories and warehouses of some of its key suppliers. Under what the company calls its “Vendor Flex” program, the number of Band-Aids that Johnson & Johnson produces, for example, can depend in part on Amazon’s need. It gives the retail behemoth a role in managing production that extends beyond its own corporate borders.

  Beyond simply distributing products, Amazon is, like Walmart, “pulling” demand. In fact, in its early days, Amazon headhunted so many top-level managers from Walmart for their logistics savvy that the Waltons sued. The untold billions of gigabytes of customer data that Amazon collects and the algorithm marvels it uses to parse this data give it an incredibly detailed picture of what people want to buy, and when. Meanwhile, integrating operations with producers ensures that products can be ready in sufficient quantities. Here too, given the sheer scale of this economy, we see the fits and starts of a more integrated model of production and distribution planning, however hierarchical and servile toward its bosses it may be. We might describe Jeff Bezos as the bald, moustache-less Stalin of online retail.

  Yet at heart, Amazon remains (for now) a giant distribution network for consumer goods. The internet age has enabled the rise of a new type of retail model for moving goods from producers to consumers, and Amazon took advantage of this opening better than any of its rivals did. Amazon now controls nearly half of total online retail in the United States. So when Amazon plans, it plans big. Some of Amazon’s planning problems are the same as those faced by other major distribution networks; other problems are entirely novel. In essence though, Amazon’s story is another tale of getting the logistics right—in other words, getting things from point A to point B as cheaply as possible. While this task sounds simple enough, it demands plans for everything from warehouse siting and product organization to minimizing the costs of delivering customers’ packages and shortening delivery routes. Wired magazine describes the company as “a vast, networked, intelligent engine for sating consumer desire.”

  Add to this the fact that Amazon, as with every internet company, collects improbable amounts of data on its consumers. A conventional brick-and-mortar store doesn’t know which products you look at, how long you spend looking at them, which ones you put in your cart and then put back on the shelf before arriving at the checkout, or even which ones you “wish” you had. But Amazon does. This data tsunami not only involves consumer information, but stretches throughout the supply chain, and the company uses this data to its advantage wherever it can. Its planning problems are no longer the pedestrian optimization challenges faced by any large company before the internet age, but rather the optimization of “big data”—sets of data that are produced at such gargantuan volumes, varieties and velocities that traditional data processing techniques and software are insufficient.

  Amazon’s scale—its ambition to be the “everything store”—introduces significant problems for its IT systems. It is one thing to deliver even a thousand products to a hundred or a thousand retail stores, as would a traditional seller. It is another to deliver millions of products to millions of customers. The problems that Amazon has to solve to be the most efficient it can be are very hard, even if they may not appear so at first glance.

  The warehouse and transport problems mentioned above are a particular class of mathematical challenge known as “optimization problems.” In an optimization problem, we aim to do something in the best way possible, subject to a number of limits on our action, or “constraints.” Given three different possible routes through a city to deliver a package, say, which is fastest given the number of traffic lights and one-way streets? Or more realistically for Amazon, in delivering some daily number of packages, the company is limited by the schedule of delivery flights, the speed of airplanes, the availability of delivery trucks and a host of other constraints, in addition to city traffic. There are also random events, such as bad weather, that can shut down airports—and while these are sporadic, they are also more likely in some places and at certain times than others.

  Every day when you commute to work, you are solving a relatively simple optimization problem. But the math behind optimization is very complex for problems with more than even just a handful of constraints. Given enough variables (conditions that can change) to be optimized and enough constraints, even the most powerful supercomputer we can currently construct, armed with the best possible algorithm we can design, would be incapable of solving some of these problems within our lifetime, and some even within the lifetime of the universe. Many of Amazon’s problems fall squarely into such categories.

  So while patents for drone delivery get all the media attention, the true wonders at the heart of its operations are actually the esoteric mathematical techniques that help it manage and simplify its optimization problems. To give one example, these key patents help Amazon plan how to best move items between warehouse shelves and customer doorsteps. Part of solving this problem involves “load balancing”: the same way that your computer shifts tasks so as to not crash any single system, Amazon decides where to build its massive warehouses and how to distribute products between them to make sure no part of its system gets overloaded.

  To be clear, Amazon’s planning methods are not complete solutions to optimization problems that might take the lifetime of the universe to solve, but instead simply best approximations to get around exploding mathematical complexity. Yet Amazon still chooses to plan rather than leave optimization up to price signals from the market. Amazon’s engineers break down problems into smaller pieces, simplifying them or finding other ways of giving a computer a chance at solving them in seconds, rather than eons. What Amazon looks for is traction; the aim is to make problems tractable rather than to solve them with absolute precision.

  Again, take the problem of shipping orders at the lowest cost. Even precisely answering the seemingly simple question of finding the lowest cost shipping method for a day’s worth of orders can quickly grow out of hand. There is no single best way to ship one order out of thousands or millions shipped on a given day, because each order’s cost depends on all the others. Will the plane from the UPS “Worldport” hub in Louisville, Kentucky, to Phoenix be full? Did your neighbor down the street order her electric toothbrush with express shipping, or can it be delivered with your book order tomorrow? The complexity ratchets up still further when Amazon considers not only all the possible alternative routes—which
it controls—but also adjusts for the possibility of random events such as severe weather and tries to predict the next day’s orders. This “order assignment” optimization problem has hundreds of millions of variables, and no easy solution. The problem is so complex that there are not even approximations that can take every aspect of the problem into account.

  But despite such problems, the planning process within Amazon does not fall apart. While Amazon may depend on horrible working conditions, low taxes and poor wages, it nevertheless functions. The planning problems faced by individual corporations under capitalism do have approximate, “good enough” solutions. As this book argues again and again, planning exists on a wide scale within the black box of the corporation—even if it is “good enough” rather than perfect.

  That’s the trick: to find the best possible, even if partial, approximations. Amazon’s modelers work to bring intractably complex problems down to size, to build plans that neither stretch into infinite time, nor respond to all the possible random events that could happen at every step, but that simply work. This means coming as close as possible to the true answer of a planning question within a realistic time frame and with the use of available computing power. When it is impossible to use an “algorithm of algorithms” to mechanically find the algorithm that best approximates the original problem, creativity then comes into play.

  As computing power increases and mathematical science advances, our solutions to optimization problems become better and better. The planning problem is not one of 100 percent precision, but of efficiently using the available computing power to get to 80 percent or 95 percent of the way there. And remember that the market isn’t 100 percent precise either; prices are constantly in flux, and the economy is constantly adjusting. Far from the Econ 101 fantasy of economic equilibrium, the market is never anywhere close to a perfect synchronization of what we want and what is produced.