societies and economies.

Each company
A Big Choice for hoards the information it collects and
WORLD WAR WEB

uses centralized systems to run its huge

Big Tech businesses. That hoarding has hampered
innovation and allowed the companies
to abuse user data, and their centralized
Share Data or Suªer the systems leave online markets vulnerable
Consequences to unexpected shocks, posing risks to the
wider economy.
Viktor Mayer-Schönberger and The most common answer to the
problem of overly powerful firms is to
Thomas Ramge break them up, as U.S. regulators once
did to Standard Oil and AT&T. Yet that

O
ver the last two decades, a few would destroy much of the value that
technology giants have come these digital giants have created and
to dominate digital markets. probably do little to improve competition
Google performs about nine out of every in the long run, since without structural
ten Internet searches worldwide. Face- reforms, killing today’s digital superstars
book, the world’s leading social media would simply generate opportunities for
platform, has well over two billion users. new ones to emerge. There is a better
Together, the two companies have seized solution: a progressive data-sharing
well over half of the online advertising mandate. This would leave these com-
market. Apple, originally a computer panies intact but require them to share
manufacturer, now runs the world’s anonymized slices of the data they collect
largest mobile app store in terms of with other companies. Such a mandate
revenue, with about 80 percent of the would decentralize digital markets and
market, and the second-largest music spur innovation as companies competed
streaming business, approaching a third to extract the best insights from the same
of the market. And Amazon captures data. Much is at stake; if governments
close to every other dollar spent online fail to act, they will leave key parts of
in the United States. These companies Western economies and democracies
are what the economist David Autor calls vulnerable to sudden failures.
“superstar firms,” able to gain huge market
shares and translate their market power MARKETS IN EVERYTHING
into enormous profits. What sets the new digital superstars
Their success has brought tremendous apart from other firms is not their market
benefits to users—and grave dangers to dominance; many traditional companies
have reached similarly commanding market
VIKTOR MAYER-SCHÖNBERGER is
Professor of Internet Governance and Regulation
shares in the past. What’s new about these
at the University of Oxford. companies is that they themselves are
THOMAS RAMGE is Technology Correspondent
markets. Amazon operates a platform
for brand eins and writes for The Economist. on which over $200 billion worth of
They are the authors of Reinventing Capitalism goods are bought and sold each year.
in the Age of Big Data. Apple runs gigantic marketplaces for

48 f o r e i g n a f fa i r s
 A Big Choice for Big Tech

Your data or your life: at the Google office in Zurich, Switzerland, August 2009
music, video, and software. As the their customers. In traditional markets,
world’s largest music-streaming service, informing everyone about which prod-
Spotify provides the biggest marketplace ucts they are likely to prefer is cumber-
for songs. The Chinese e-commerce some and costly, so the information is
behemoth Alibaba manages the world’s usually condensed into a single figure:
largest business-to-business marketplace. price. Relying on prices provides the
Mean­while, Google and Facebook are grease that enables conventional markets
not just the dominant search engine to work, but it omits a lot of detail. If a
and the dominant social media platform, potential customer, after learning the
respectively; they also represent two of the cost of an item, decides not to buy it,
world’s largest marketplaces for advertising the seller will not know why. Was the
space. And home-rental companies such price too high? Or was the product just
as Airbnb and ride-sharing companies not what the buyer was looking for?
such as Uber and Didi Chuxing match That loss of nuance hurts both sides of
workers with customers. the transaction.
Markets have been around for Prices can also be deceptive. They
millennia; they are not an invention of are easy to compare, so customers often
the data age. But digital superstar firms believe that doing so is an appropriate
C H R I ST IA N HART MA N N / R EU T E R S

don’t operate traditional markets; theirs way to choose the products that will suit
are rich with data, which they use to them best. But behavioral economics
improve transactions and thereby make has long shown that humans are poor
consumers better off. More data about judges of their future needs and that
the products on offer and the preferences their ability to make objective compari-
of buyers and sellers helps people find sons can easily be skewed, resulting in
what they are looking for and allows many transactions that should not happen.
businesses to figure out how best to serve By contrast, the data-rich markets oper-

September/October 2018 49
Viktor Mayer-Schönberger and Thomas Ramge

ated by digital superstars offer lots of and the wider economy. As data-rich
information about products and prefer- markets replace conventional ones, compe­
ences, along with tools to search and tition withers. On Amazon, for example,
filter the available goods. On Amazon, consumers can still choose from a variety
buyers can tick options to quickly iden­ of products, but Amazon decides what
tify the kinds of products they are looking brands to carry—and it pushes its own
for. Several studies have shown that products, from batteries to food. If Apple
Amazon rarely offers the cheapest choice, wants to ban a competitor from its App
but buyers value the ability to find Store, it can. (Some argue that it has
something easily that closely matches already done so by delisting apps that
their needs. offer services similar to Apple’s own,
Digital superstars also use data in although Apple disputes this.) Market
another important way. They offer shop- dominance isn’t illegal, but if history is
pers digital “decision assistants,” which sift a guide, it commonly causes problems.
through huge amounts of information to As Microsoft discovered in the late 1990s,
provide recommendations that are often a thin line separates market dominance
superior to the choices that consumers and abuse of power. If a monopoly’s
might make on their own. Automated stranglehold on the market enables it,
assistants suggest music on Spotify, that monopoly will almost always raise
movies on Netflix, and apps in Apple’s prices when it runs into trouble. Con-
App Store. A third of Amazon’s retail sumers may pay only a small extra cost
sales reportedly result from customers per transaction as a result, but the profit
following the advice of the company’s difference for the monopoly can be
venerable decision assistant. sub­stantial. That is one reason why many
In each of these cases, the use of economists worry about the ascend­ancy
data improves market matches, and as of digital superstars. Market dominance
customers keep shopping, the assistants also allows companies to squash com-
learn how to make even better recom- petitors, either by undercutting them
mendations. Most of this learning takes on price before they have a chance to
place completely or largely unassisted by grow sufficiently large to survive a price
humans, as data is fed into machines that battle or by simply acquiring them.
continuously update their algorithms. It Market dominance is not the only
is the combination of vast markets and problem raised by the ascent of digital
decision assistants driven by the data superstars. These companies gather
those markets produce that makes these comprehensive personal data on their
companies such enticing places for customers and store that information in
customers to spend their money. At least databases that are enticing targets for
so far, consumers, and not just the digital hackers. Reports of major data breaches
superstars, seem to be winning. are mounting, and many people rightly
worry about how safe their personal
THE RETURN OF CENTRAL information is.
PLANNING But digital superstars pose an even
Yet these gains in consumer welfare bigger, more systemic threat to the wider
mask dangers for both individual users economy. By hoarding data and operating

50 f o r e i g n a f fa i r s
 A Big Choice for Big Tech

their own, dominant decision assistants, aware of doing so, or pay too much for
they undo the resilience of traditional, what they purchase. They may never be
decentralized markets. In a typical market, shown the search results that best answer
if a buyer or a seller makes a mistake, it their queries and never see the news that
may hurt that person or that company, but really matters. This not only exposes
it will not bring down the market as a millions of consumers to the risk of
whole. In contrast, in many of the data- wasting their money and their time but
rich markets run by digital superstars, the also endangers the market as a whole.
company controls all the information
about customer preferences and transac- THE FAILURE OF ANTITRUST
tions, and the company uses that data to Existing regulatory tools fail to address
help its own decision assistant learn. the risks posed by today’s superstar firms.
Buyers still make individual decisions, but Antitrust laws prohibit certain kinds of
they are greatly influenced both by explicit anticompetitive behavior. Companies
recommendations and by the ways in can be fined for fixing prices or bundling
which options are filtered and presented. products together to stifle competition.
These markets are much closer to centrally But such laws do not regulate market
planned systems than to resilient and power directly; superstar firms operat-
decentralized traditional marketplaces. ing data-rich markets can exercise a
Digital superstars don’t want to dangerous degree of control without
operate a centrally planned economy. violating any antitrust laws.
But given the complexity of the algo- The legislators who wrote the existing
rithms that recommend products, it’s laws assumed that competition would
always tempting for these companies make it hard to concentrate market power
to manipulate the markets they operate. for long. To be sure, as companies grow,
Last year, for example, the European the marginal cost of production usually
Commission fined Google 2.4 billion decreases. As long as there are sufficient
euros after finding that the company had buyers, it pays for a company to get big-
distorted its search results to favor its ger. And for some products and services,
own shopping service. And a company’s new users improve the value for everyone
motives don’t always matter: the hoard- else: when more users join a social media
ing of data makes it far easier to system- platform, the existing users have more
atically influence consumers’ decisions people to interact with. But the advan-
even without intending to do so. tages of being a big, well-connected
Central control of data means that company have long been balanced by the
faults in a system have the potential to ability of start-ups to take on established
influence not just a single consumer but firms by developing superior products or
also every market participant. If there is better production processes. Antitrust
a systemic problem in Amazon’s recom- laws have been needed to stop only the
mendation engine, everyone heeding its most egregious abuses; innovation has
advice may make needlessly poor deci- eventually taken care of ordinary market
sions. People may buy products that don’t concentration.
really match their preferences, shun a That is changing. Innovation is
particular brand without even being shifting to data-driven machine learn-

September/October 2018 51
Viktor Mayer-Schönberger and Thomas Ramge

ing. Insights are no longer the product learn from, leading to inferior products
solely of human ingenuity. They are now and services overall.
the result of the automated analysis of Similarly, although restricting the
huge amounts of data. More and more, ways digital superstars can collect or
the success of a firm rests on its ability to use data—through tougher privacy laws,
use the information it possesses. Because for instance—might fragment markets
only the largest firms have access to and thus improve their resilience, the
enough data to compete, innovation is quality of recommendations would dete­
losing its power to make markets fairer. ­riorate absent sufficient data, leading
To solve this problem, some experts to inefficient transactions and reduced
have suggested breaking up digital super­ consumer welfare.
stars, so that they no longer control the
marketplace, the information that flows TRUSTBUSTING FOR THE
among market participants, and the DIGITAL AGE
decision assistants. The model would Luckily, regulators do not have to
be the robust antitrust enforcement that choose between structurally vulnerable
led to the breakup of Standard Oil, in but efficient markets and resilient but
1911, and AT&T, in 1984. A less drastic inefficient ones. There’s an easier way
alternative might draw inspiration from to foster both market diversity and
the steps taken by regulators in the 1990s resilience: a progressive data-sharing
to force Microsoft to stop bundling a mandate. Under this system, every
Web browser with its operating system company above a certain size, say, those
and, more recently, to prevent Google with more than a ten percent share of
from favoring its own services in its the market, that systematically collects
search results. and analyzes data would have to let
But by reducing firms’ ability to use other companies in the same market
large amounts of data, such measures access a subset of its data. The larger a
would reduce market efficiency and leave firm’s market share, the more of its data
consumers worse off. If, for instance, others would be allowed to see. Data
Amazon were broken up into a market- would be stripped of personal identifi-
place and a separate tool to provide ers, augmented with metadata to make
recommendations, the latter would no clear what sort of information the data
longer have access to the huge streams provided and where it came from, and
of data generated by the former. Nor selected randomly to prevent companies
would a breakup improve competition. from gaming the system (by granting
Alternative recommendation engines access only to largely useless data, for
would not see the market data either, so instance). Participants would have to
their suggestions would be no better. It agree to certain restrictions, including
would not really matter how regulators rules against sharing data with third
broke a firm up—whether they created parties. The role of regulators would be
many little Googles, for instance, or limited to assessing market share, an area
split YouTube from Google Search— in which they have already accumulated
because after the breakup, all the new expertise. If necessary, regulators would
entities would have less information to also enforce access to data, but they

52 f o r e i g n a f fa i r s
 A Big Choice for Big Tech

would not actively organize or operate different degrees. That would support
the sharing system. diversity, innovation, and competition.
Eventually, data sharing should be Once companies had access to the
mandated across the board. But countries necessary raw material, they would
should start with online markets, as these launch alternative decision assistants.
are particularly vulnerable to the dangers People might still shop on Amazon or
of concentration. In the United States, listen to music on Spotify, but they
Congress would have to amend the might use a third-party recommenda-
country’s existing antitrust regime to tion tool to choose products and songs.
develop a comprehensive data-sharing Today’s decision assistants mostly serve
regulation, and in Europe, the eu would the digital superstars. Tomorrow’s more
have to act as a whole, but a transatlantic independent decision assistants could far
consensus would not be necessary. Both more convincingly represent the interests
the United States and the eu have enough of consumers. Price-comparison sites
regulatory power and important enough already let people find the seller offer-
markets to make a mandate enacted in ing the lowest price for a wide range
either jurisdiction effective. of products. Independent decision
A progressive data-sharing mandate assistants would help them identify the
would offer several advantages. Unlike a best product match, as well.
tax, it would not impose any direct cost Creating competition among assistants
on firms; companies would remain free and markets would eliminate the need to
to use the data they collect, just as they break up digital superstars, because they
do now. It would allow many firms and would no longer enjoy an insurmountable
people to use the same data, which would competitive advantage. And because the
spur innovation; today, although huge shared data would be chosen randomly,
quantities of data are collected, it remains each competitor would train its systems
underused. If a wide variety of firms had on slightly different data sets, reducing
access to market data, a firm’s competitive the risk of systemic failures.
advantage would rest on its ability to Some may worry that mandated data
extract insights, encouraging companies to sharing would only boost the power
develop smarter algorithms and analytics. that firms have over consumers. But
The policy would not differentiate so far, regulators have mostly failed to
between different players that crossed counter power imbalances between
the necessary threshold; even Amazon users and companies by strengthening
would have access to data from smaller individual privacy rights. Even in Europe,
competitors as long as their market which has enacted the strongest privacy
shares were greater than ten percent. protections, most people routinely click
But since smaller firms would have less “OK” and accept companies’ privacy
data to share and machine-learning terms rather than exercise their right to
algorithms produce diminishing returns choose what information they share. The
for each new data point, a company problem is that current privacy regula-
like Amazon would gain far less than tions focus too narrowly on the relation-
its smaller competitors. A data-sharing ship between each consumer and each
mandate would lift all boats, but to firm, rather than considering the market

September/October 2018 53
Viktor Mayer-Schönberger and Thomas Ramge

as a whole. Mandating progressive data legislative aim is similar: to break the

sharing would not solve the privacy stranglehold incumbents maintain on data
challenge. But preventing a small number and create a competitive market full of
of digital superstars from monopolizing innovative data-driven services. Both the
data would better distribute the power European Commission and members of
that flows from exclusive access to infor- the European Parliament are studying
mation. Requiring firms to strip obvious how to legislate progressive data-sharing
personal identifiers from their data before mandates, and Andrea Nahles, a leader of
sharing it would also greatly limit the Germany’s ruling coalition, has thrown
potential exposure of their users, spread- her support behind the idea.
ing power more equally among companies Requiring companies to share their
without compromising privacy. data will come with challenges. Govern-
ments will have to overcome several
MAKING IT WORK organizational and technical hurdles,
The idea of forcing companies to share and making such a policy effective will
their data may seem novel, but it has been require some fine-tuning. But the success-
employed successfully around the world. ful cases of regulated data sharing in
In the United States, the federal govern­ the United States and Europe, and the
ment sometimes requires firms to share rising interest in such policies, suggest
data as a condition of allowing an acquisi- that regulators will be able to find prag-
tion. For example, in 2011, when Google matic solutions.
bought the airline reservation company If, on the other hand, governments
ita Software, the Justice Department leave accelerating market concentration
forced Google to continue to offer access untouched, they will threaten more than
to ita Software’s travel data to third competition. They will allow the most
parties, including Google’s competitor efficient digital markets to become also
Microsoft. The situation is similar across the most vulnerable, both to deliberate
the Atlantic. In Germany, for instance, attacks and to accidental failures that
large insurers must let smaller insurers cause systemic breakdowns. Allowing
access some of their actuarial data, so that superstar companies that run centrally
the smaller firms have enough informa- planned systems to maintain their domi­
tion to calculate risks. nance will erode the most important
Recent eu legislation has boosted the quality of capitalist markets, the decen-
idea of data sharing. Thanks to new rules, tralized way they make decisions. Worse
a consumer can now force companies to still, this centralization has begun to
provide competitors with access to his or spill over into politics. Just as healthy
her banking data (as part of so-called open markets rely on people making inde-
banking), as well as any other personal pendent decisions, democratic systems
data (as part of the new General Data are founded on the individual choices
Protection Regulation). This approach, made by voters. Yet today, much of
which the eu has dubbed “data portability,” public discourse is shaped by a few
will not have the same effect as a data- companies with exclusive access to vast
sharing mandate, since it requires each quantities of voter data. Ignoring the
consumer to act independently. But the problem is no longer an option.∂

54 f o r e i g n a f fa i r s