83
Chapter 4
White-Collar
Jobs at Risk
On October 11, 2009, the Los Angeles Angels prevailed over the Boston
Red Socks in the American League play-offs and earned the right
to face the New York Yankees for the league championship and entry
into the World Series. It was an especially emotional win for the
Angels because just six months earlier one of their most promising
players, pitcher Nick Adenhart, had been killed by a drunk driver in
an automobile accident. One sportswriter began an article describing
the game like this:
Things looked bleak for the Angels when they trailed by two runs in
the ninth inning, but Los Angeles recovered thanks to a key single
from Vladimir Guerrero to pull out a 7–6 victory over the Boston
Red Sox at Fenway Park on Sunday.
Guerrero drove in two Angels runners. He went 2–4 at the plate.
“When it comes down to honoring Nick Adenhart, and what
happened in April in Anaheim, yes, it probably was the biggest hit
[of my career],” Guerrero said. “Because I’m dedicating that to a
former teammate, a guy that passed away.”
9780465059997-text.indd 83 2/12/15 12:46 PM
RISE OF THE ROBOTS84
Guerrero has been good at the plate all season, especially in day
games. During day games Guerrero has a .794 OPS [on-base plus
slugging]. He has hit five home runs and driven in 13 runners in 26
games in day games.1
The author of that text is probably in no immediate danger of receiving
any awards for his writing. The narrative is nonetheless a remarkable
achievement: not because it is readable, grammatically correct,
and an accurate description of the baseball game, but because the
author is a computer program.
The software in question, called “StatsMonkey,” was created by
students and researchers at Northwestern University’s Intelligent Information
Laboratory. StatsMonkey is designed to automate sports
reporting by transforming objective data about a particular game
into a compelling narrative. The system goes beyond simply listing
facts; rather, it writes a story that incorporates the same essential
attributes that a sports journalist would want to include. StatsMonkey
performs a statistical analysis to discern the notable events that
occurred during a game; it then generates natural language text that
summarizes the game’s overall dynamic while focusing on the most
important plays and the key players who contributed to the story.
In 2010, the Northwestern University researchers who oversaw
the team of computer science and journalism students who worked
on StatsMonkey raised venture capital and founded a new company,
Narrative Science, Inc., to commercialize the technology. The company
hired a team of top computer scientists and engineers; then it
tossed out the original StatsMonkey computer code and built a far
more powerful and comprehensive artificial intelligence engine that
it named “Quill.”
Narrative Science’s technology is used by top media outlets, including
Forbes, to produce automated articles in a variety of areas,
including sports, business, and politics. The company’s software generates
a news story approximately every thirty seconds, and many
of these are published on widely known websites that prefer not to
9780465059997-text.indd 84 2/12/15 12:46 PM
White-Collar Jobs at Risk 85
acknowledge their use of the service. At a 2011 industry conference,
Wired writer Steven Levy prodded Narrative Science co-founder
Kristian Hammond into predicting the percentage of news articles
that would be written algorithmically within fifteen years. His answer:
over 90 percent.2
Narrative Science has its sights set on far more than just the news
industry. Quill is designed to be a general-purpose analytical and
narrative-writing engine, capable of producing high-quality reports
for both internal and external consumption across a range of industries.
Quill begins by collecting data from a variety of sources,
including transaction databases, financial and sales reporting systems,
websites, and even social media. It then performs an analysis
designed to tease out the most important and interesting facts
and insights. Finally, it weaves all this information into a coherent
narrative that the company claims measures up to the efforts of the
best human analysts. Once it’s configured, the Quill system can
generate business reports nearly instantaneously and deliver them
continuously—all without human intervention.3
One of Narrative
Science’s earliest backers was In-Q-Tel, the venture capital arm of
the Central Intelligence Agency, and the company’s tools will likely
be used to automatically transform the torrents of raw data collected
by the US intelligence community into an easily understandable narrative
format.
The Quill technology showcases the extent to which tasks that
were once the exclusive province of skilled, college-educated professionals
are vulnerable to automation. Knowledge-based work, of
course, typically calls upon a wide range of capabilities. Among other
things, an analyst may need to know how to retrieve information
from a variety of systems, perform statistical or financial modeling,
and then write understandable reports and presentations. Writing—
which, after all, is at least as much art as it is science—might seem
like one of the least likely tasks to be automated. Nevertheless, it
has been, and the algorithms are improving rapidly. Indeed, because
9780465059997-text.indd 85 2/12/15 12:46 PM
RISE OF THE ROBOTS86
knowledge-based jobs can be automated using only software, these
positions may, in many cases, prove to be more vulnerable than lowerskill
jobs that involve physical manipulation.
Writing also happens to be an area in which employers consistently
complain that college graduates are deficient. One recent
survey of employers found that about half of newly hired two-year
college graduates and over a quarter of those with four-year degrees
were found to have poor writing—and in some cases even reading—
skills.4
If intelligent software can, as Narrative Science claims, begin
to rival the most capable human analysts, the future growth of
knowledge-based employment is in doubt for all college graduates,
especially the least prepared.
Big Data and Machine Learning
The Quill narrative-writing engine is just one of many new software
applications being developed to leverage the enormous amounts of
data now being collected and stored within businesses, organizations,
and governments across the global economy. By one estimate,
the total amount of data stored globally is now measured in thousands
of exabytes (an exabyte is equal to a billion gigabytes), and that
figure is subject to its own Moore’s Law–like acceleration, doubling
roughly every three years.5
Nearly all of that data is now stored in
digital format and is therefore accessible to direct manipulation by
computers. Google’s servers alone handle about 24 petabytes (equal
to a million gigabytes)—primarily information about what its millions
of users are searching for—each and every day.6
All this data arrives from a multitude of different sources. On the
Internet alone, there are website visits, search queries, emails, social
media interactions, and advertising clicks, to name just a few examples.
Within businesses, there are transactions, customer contacts,
internal communications, and data captured in financial, accounting,
and marketing systems. Out in the real world, sensors continuously
9780465059997-text.indd 86 2/12/15 12:46 PM
White-Collar Jobs at Risk 87
capture real-time operational data in factories, hospitals, automobiles,
aircraft, and countless other consumer devices and industrial
machines.
The vast majority of this data is what a computer scientist would
call “unstructured.” In other words, it is captured in a variety of
formats that can often be difficult to match up or compare. This
is very different from traditional relational database systems where
information is arranged neatly in consistent rows and columns that
make search and retrieval fast, reliable, and precise. The unstructured
nature of big data has led to the development of new tools
specifically geared toward making sense of information that is collected
from a variety of sources. Rapid improvement in this area is
just one more example of the way in which computers are, at least
in a limited sense, beginning to encroach on capabilities that were
once exclusive to human beings. The ability to continuously process
a stream of unstructured information from sources throughout our
environment is, after all, one of the things for which humans are
uniquely adapted. The difference, of course, is that in the realm of
big data, computers are able to do this on a scale that, for a person,
would be impossible. Big data is having a revolutionary impact in a
wide range of areas including business, politics, medicine, and nearly
every field of natural and social science.
Major retailers are relying on big data to get an unprecedented
level of insight into the buying preferences of individual shoppers,
allowing them to make precisely targeted offers that increase revenue
while helping to build customer loyalty. Police departments across
the globe are turning to algorithmic analysis to predict the times
and locations where crimes are most likely to occur and then deploying
their forces accordingly. The City of Chicago’s data portal
allows residents to see both historical trends and real-time data in a
range of areas that capture the ebb and flow of life in a major city—
including energy usage, crime, performance metrics for transportation,
schools and health care, and even the number of potholes
9780465059997-text.indd 87 2/12/15 12:46 PM
RISE OF THE ROBOTS88
patched in a given period of time. Tools that provide new ways to
visualize data collected from social media interactions as well as sensors
built into doors, turnstiles, and escalators offer urban planners
and city managers graphic representations of the way people move,
work, and interact in urban environments, a development that may
lead directly to more efficient and livable cities.
There is a potential dark side, however. Target, Inc., provided a
far more controversial example of the ways in which vast quantities
of extraordinarily detailed customer data can be leveraged. A data
scientist working for the company found a complex set of correlations
involving the purchase of about twenty-five different health and
cosmetic products that were a powerful early predictor of pregnancy.
The company’s analysis could even estimate a woman’s due date
with a high degree of accuracy. Target began bombarding women
with offers for pregnancy-related products at such an early stage that,
in some cases, the women had often not yet shared the news with
their immediate families. In an article published in early 2012, the
New York Times reported one case in which the father of a teenage
girl actually complained to store management about mail sent to
the family’s home—only to find out later that Target, in fact, knew
more than he did.7
Some critics fear that this rather creepy story is
only the beginning and that big data will increasingly be used to
generate predictions that potentially violate privacy and perhaps even
freedom.
The insights gleaned from big data typically arise entirely from
correlation and say nothing about the causes of the phenomenon
being studied. An algorithm may find that if A is true, B is likely
also true. But it cannot say whether A causes B or vice versa—or if
perhaps both A and B are caused by some external factor. In many
cases, however, and especially in the realm of business where the
ultimate measure of success is profitability and efficiency rather than
deep understanding, correlation alone can have extraordinary value.
Big data can offer management an unprecedented level of insight
9780465059997-text.indd 88 2/12/15 12:46 PM
White-Collar Jobs at Risk 89
into a wide range of areas: everything from the operation of a single
machine to the overall performance of a multinational corporation
can potentially be analyzed at a level of detail that would have been
impossible previously.
The ever-growing mountain of data is increasingly viewed as a
resource that can be mined for value—both now and in the future.
Just as extractive industries like oil and gas continuously benefit from
technical advances, it’s a good bet that accelerating computer power
and improved software and analysis techniques will enable corporations
to unearth new insights that lead directly to increased profitability.
Indeed, that expectation on the part of investors is probably
what gives data-intensive companies like Facebook such enormous
valuations.
Machine learning—a technique in which a computer churns
through data and, in effect, writes its own program based on the
statistical relationships it discovers—is one of the most effective
means of extracting all that value. Machine learning generally involves
two steps: an algorithm is first trained on known data and
is then unleashed to solve similar problems with new information.
One ubiquitous use of machine learning is in email spam filters. The
algorithm might be trained by processing millions of emails that have
been pre-categorized as either spam or not. No one sits down and
directly programs the system to recognize every conceivable typographic
butchery of the word “Viagra.” Instead, the software figures
this out by itself. The result is an application that can automatically
identify the vast majority of junk email and can also continuously
improve and adapt over time as more examples become available.
Machine learning algorithms based on the same basic principles recommend
books at Amazon.com, movies at Netflix, and potential
dates at Match.com.
One of the most dramatic demonstrations of the power of machine
learning came when Google introduced its online language
translation tool. Its algorithms used what might be called a “Rosetta
9780465059997-text.indd 89 2/12/15 12:46 PM
RISE OF THE ROBOTS90
Stone” approach to the problem by analyzing and comparing millions
of pages of text that had already been translated into multiple
languages. Google’s development team began by focusing on official
documents prepared by the United Nations and then extended their
effort to the Web, where the company’s search engine was able to
locate a multitude of examples that became fodder for their voracious
self-learning algorithms. The sheer number of documents used to
train the system dwarfed anything that had come before. Franz Och,
the computer scientist who led the effort, noted that the team had
built “very, very large language models, much larger than anyone has
ever built in the history of mankind.”8
In 2005, Google entered its system in the annual machine translation
competition held by the National Bureau of Standards and
Technology, an agency within the US Commerce department that
publishes measurement standards. Google’s machine learning algorithms
were able to easily outperform the competition—which
typically employed language and linguistic experts who attempted
to actively program their translation systems to wade through the
mire of conflicting and inconsistent grammatical rules that characterize
languages. The essential lesson here is that, when datasets are
large enough, the knowledge encapsulated in all that data will often
trump the efforts of even the best programmers. Google’s system is
not yet competitive with the efforts of skilled human translators, but
it offers bidirectional translation between more than five hundred
language pairs. That represents a genuinely disruptive advance in
communication capability: for the first time in human history, nearly
anyone can freely and instantly obtain a rough translation of virtually
any document in any language.
While there are a number of different approaches to machine
learning, one of the most powerful, and fascinating, techniques involves
the use of artificial neural networks—or systems that are designed
using the same fundamental operating principles as the human
brain. The brain contains as many as 100 billion neuron cells—and
9780465059997-text.indd 90 2/12/15 12:46 PM
White-Collar Jobs at Risk 91
many trillions of connections between them—but it’s possible to
build powerful learning systems using far more rudimentary configurations
of simulated neurons.
An individual neuron operates somewhat like the plastic pop-up
toys that are popular with very young children. When the child
pushes the button, a colorful figure pops up—perhaps a cartoon
character or an animal. Press the button gently and nothing happens.
Press it a bit harder and still nothing. But exceed a certain force
threshold, and up pops the figure. A neuron works in essentially the
same fashion, except that the activation button can be pressed by a
combination of multiple inputs.
To visualize a neural network, imagine a Rube Goldberg–like
machine in which a number of these pop-up toys are arranged on
the floor in rows. Three mechanical fingers are poised over each
toy’s activation button. Rather than having a figure pop up, the toys
are configured so that when a toy is activated it causes several of the
mechanical fingers in the next row of toys to press down on their own
buttons. The key to the neural network’s ability to learn is that the
force with which each finger presses down on its respective button
can be adjusted.
To train the neural network, you feed known data into the first
row of neurons. For example, imagine inputting visual images of
handwritten letters. The input data causes some of the mechanical
fingers to press down with varying force depending on their calibration.
That, in turn, causes some of the neurons to activate and
press down on buttons in the next row. The output—or answer—is
gathered from the last row of neurons. In this case, the output will
be a binary code identifying the letter of the alphabet that corresponds
to the input image. Initially, the answer will be wrong, but
our machine also includes a comparison and feedback mechanism.
The output is compared to the known correct answer, and this automatically
results in adjustments to the mechanical fingers in each
row, and that, in turn, alters the sequence of activating neurons. As
9780465059997-text.indd 91 2/12/15 12:46 PM
RISE OF THE ROBOTS92
the network is trained with thousands of known images, and then the
force with which the fingers press down is continuously recalibrated,
the network will get better and better at producing the correct answer.
When things reach the point where the answers are no longer
improving, the network has effectively been trained.
This is, in essence, the way that neural networks can be used to
recognize images or spoken words, translate languages, or perform a
variety of other tasks. The result is a program—essentially a list of all
the final calibrations for the mechanical fingers poised over the neuron
activation buttons—that can be used to configure new neural networks,
all capable of automatically generating answers from new data.
Artificial neural networks were first conceived and experimented
with in the late 1940s and have long been used to recognize patterns.
However, the last few years have seen a number of dramatic breakthroughs
that have resulted in significant advances in performance,
especially when multiple layers of neurons are employed—a technology
that has come to be called “deep learning.” Deep learning systems
already power the speech recognition capability in Apple’s Siri
and are poised to accelerate progress in a broad range of applications
that rely on pattern analysis and recognition. A deep learning neural
network designed in 2011 by scientists at the University of Lugano
in Switzerland, for example, was able to correctly identify more than
99 percent of the images in a large database of traffic signs—a level
of accuracy that exceeded that of human experts who competed
against the system. Researchers at Facebook have likewise developed
an experimental system—consisting of nine levels of artificial
neurons—that can correctly determine whether two photographs are
of the same person 97.25 percent of the time, even if lighting conditions
and orientation of the faces vary. That compares with 97.53
percent accuracy for human observers.9
Geoffrey Hinton of the University of Toronto, one of the leading
researchers in the field, notes that deep learning technology “scales
beautifully. Basically you just need to keep making it bigger and
9780465059997-text.indd 92 2/12/15 12:46 PM
White-Collar Jobs at Risk 93
faster, and it will get better.”10
In other words, even without accounting
for likely future improvements in their design, machine learning
systems powered by deep learning networks are virtually certain to
see continued dramatic progress simply as a result of Moore’s Law.
Big data and the smart algorithms that accompany it are having
an immediate impact on workplaces and careers as employers, particularly
large corporations, increasingly track a myriad of metrics and
statistics regarding the work and social interactions of their employees.
Companies are relying ever more on so-called people analytics
as a way to hire, fire, evaluate, and promote workers. The amount
of data being collected on individuals and the work they engage in is
staggering. Some companies capture every keystroke typed by every
employee. Emails, phone records, web searches, database queries and
accesses to files, entry and exit from facilities, and untold numbers
of other types of data may also be collected—with or without the
knowledge of workers.11
While the initial purpose of all this data
collection and analysis is typically more effective management and
assessment of employee performance, it could eventually be put to
other uses—including the development of software to automate much
of the work being performed.
The big data revolution is likely to have two especially important
implications for knowledge-based occupations. First, the data captured
may, in many cases, lead to direct automation of specific tasks
and jobs. Just as a person might study the historical record and then
practice completing specific tasks in order to learn a new job, smart
algorithms will often succeed using essentially the same approach.
Consider, for example, that in November 2013 Google applied for a
patent on a system designed to automatically generate personalized
email and social media responses.12
The system works by first analyzing
a person’s past emails and social media interactions. Based
on what it learned, it would then automatically write responses to
future emails, Tweets, or blog posts, and it would do so employing
the person’s usual writing style and tone. It’s easy to imagine such
9780465059997-text.indd 93 2/12/15 12:46 PM
RISE OF THE ROBOTS94
a system eventually being used to automate a great deal of routine
communication.
Google’s automated cars, which it first demonstrated in 2011,
likewise provide important insight into the path that data-driven
automation is likely to follow. Google didn’t set out to replicate the
way a person drives—in fact, that would have been beyond the current
capabilities of artificial intelligence. Rather, it simplified the
challenge by designing a powerful data processing system and then
putting it on wheels. Google’s cars navigate by relying on precision
location awareness via GPS together with vast amounts of extremely
detailed mapping data. The cars also, of course, have radars, laser
range finders, and other systems that provide a continuous stream of
real-time information and allow the car to adapt to new situations,
such as a pedestrian stepping off the curb. Driving may not be a
white-collar profession, but the general strategy used by Google can
be extended into a great many other areas: First, employ massive
amounts of historical data in order to create a general “map” that
will allow algorithms to navigate their way through routine tasks.
Next, incorporate self-learning systems that can adapt to variations
or unpredictable situations. The result is likely to be smart software
that can perform many knowledge-based jobs with a high degree of
reliability.
The second, and probably more significant, impact on knowledge
jobs will occur as a result of the way big data changes organizations
and the methods by which they are managed. Big data and predictive
algorithms have the potential to transform the nature and number
of knowledge-based jobs in organizations and industries across the
board. The predictions that can be extracted from data will increasingly
be used to substitute for human qualities such as experience
and judgment. As top managers increasingly employ data-driven
decision making powered by automated tools, there will be an evershrinking
need for an extensive human analytic and management
infrastructure. Whereas today there is a team of knowledge workers
9780465059997-text.indd 94 2/12/15 12:46 PM
White-Collar Jobs at Risk 95
who collect information and present analysis to multiple levels
of management, eventually there may be a single manager and a
powerful algorithm. Organizations are likely to flatten. Layers of
middle management will evaporate, and many of the jobs now performed
by both clerical workers and skilled analysts will simply
disappear.
WorkFusion, a start-up company based in the New York City
area, offers an especially vivid example of the dramatic impact that
white-collar automation is likely to have on organizations. The company
offers large corporations an intelligent software platform that
almost completely manages the execution of projects that were once
highly labor-intensive through a combination of crowd sourcing and
automation.
The WorkFusion software initially analyzes the project to determine
which tasks can be directly automated, which can be crowd
sourced, and which must be performed by in-house professionals.
It can then automatically post job listings to websites like Elance
or Craigslist and manage the recruitment and selection of qualified
freelance workers. Once the workers are on board, the software allocates
tasks and evaluates performance. It does this in part by asking
freelancers to answer questions to which it already knows the answer
as an ongoing test of the workers’ accuracy. It tracks productivity
metrics like typing speed, and automatically matches tasks with the
capabilities of individuals. If a particular person is unable to complete
a given assignment, the system will automatically escalate that
task to someone with the necessary skills.
While the software almost completely automates management of
the project and dramatically reduces the need for in-house employees,
the approach does, of course, create new opportunities for freelance
workers. The story doesn’t end there, however. As the workers complete
their assigned tasks, WorkFusion’s machine learning algorithms
continuously look for opportunities to further automate the process.
In other words, even as the freelancers work under the direction of
9780465059997-text.indd 95 2/12/15 12:46 PM
RISE OF THE ROBOTS96
the system, they are simultaneously generating the training data that
will gradually lead to their replacement with full automation.
One of the company’s initial projects involved retrieving the information
necessary to update a collection of about 40,000 records.
Previously, the corporate client had performed this process annually
using an in-house staff at a cost of nearly $4 per record. After
switching to the WorkFusion platform, the client was able to update
the records monthly at a cost of just 20 cents each. WorkFusion has
found that, as the system’s machine learning algorithms incrementally
automate the process further, costs typically drop by about 50
percent after one year and still another 25 percent after a second year
of operation.13
Cognitive Computing and IBM Watson
In the fall of 2004, IBM executive Charles Lickel had dinner with a
small team of researchers at a steakhouse near Poughkeepsie, New
York. Members of the group were taken aback when, at precisely
seven o’clock, people suddenly began standing up from their tables
and crowding around a television in the bar area. It turned out that
Ken Jennings, who had already won more than fifty straight matches
on the TV game show Jeopardy!, was once again attempting to extend
his historic winning streak. Lickel noticed that the restaurant’s
patrons were so engaged that they abandoned their dinners, returning
to finish their steaks only after the match concluded.14
That incident, at least according to many recollections, marked
the genesis of the idea to build a computer capable of playing—and
beating the very best human champions at—Jeopardy!*
IBM had a long
history of investing in high-profile projects called “grand challenges”
* Stephen Baker’s 2011 book, Final Jeopardy: Man vs. Machine and the Quest
to Know Everything, offers a detailed account of the fascinating story that ultimately
led to IBM’s Watson.
9780465059997-text.indd 96 2/12/15 12:46 PM
White-Collar Jobs at Risk 97
that have showcased the company’s technology while delivering the
kind of organic marketing buzz that just can’t be purchased at any
price. In a previous grand challenge, more than seven years earlier,
IBM’s Deep Blue computer had defeated world chess champion Garry
Kasparov in a six-game match—an event that forever anchored the
IBM brand to the historic moment when a machine first achieved
dominance in the game of chess. IBM executives wanted a new grand
challenge that would captivate the public and position the company
as a clear technology leader—and, in particular, combat any perception
that the information technology innovation baton had passed
from Big Blue to Google or to start-up companies emerging out of
Silicon Valley.
As the idea for a Jeopardy!-based grand challenge that would
culminate in a televised match between the best human competitors
and an IBM computer began to gain traction with the company’s top
managers, the computer scientists who would have to actually build
such a system initially pushed back aggressively. A Jeopardy! computer
would require capabilities far beyond anything that had been
demonstrated previously. Many researchers feared that the company
risked failure or, even worse, embarrassment on national television.
Indeed, there was little reason to believe that Deep Blue’s triumph
at chess would be extensible to Jeopardy! Chess is a game
with precise rules that operate within a strictly limited domain; it is
almost ideally suited to a computational approach. To a significant
extent, IBM succeeded simply by throwing powerful, customized
hardware at the problem. Deep Blue was a refrigerator-sized system
packed with processors that were designed specifically for playing
chess. “Brute force” algorithms leveraged all that computing power
by considering every conceivable move given the current state of the
game. Then for each of those possibilities, the software looked many
moves ahead, weighing potential actions by both players and iterating
through countless permutations—a laborious process that ultimately
nearly always produced the optimal course of action. Deep Blue was
9780465059997-text.indd 97 2/12/15 12:46 PM
RISE OF THE ROBOTS98
fundamentally an exercise in pure mathematical calculation; all the
information the computer needed to play the game was provided in
a machine-friendly format it could process directly. There was no
requirement for the machine to engage with its environment like a
human chess player.
Jeopardy! presented a dramatically different scenario. Unlike
chess, it is essentially open-ended. Nearly any subject that would
be accessible to an educated person—science, history, film, literature,
geography, and popular culture, to name just a few—is fair game.
A computer would also face an entire range of daunting technical
challenges. Foremost among these was the need to comprehend natural
language: the computer would have to receive information and
provide its responses in the same format as its human competitors.
The hurdle for succeeding at Jeopardy! is especially high because the
show has to be not just a fair contest but also an engaging form of
entertainment for its millions of television viewers. The show’s writers
often intentionally weave humor, irony, and subtle plays on words
into the clues—in other words, the kind of inputs that seem almost
purposely designed to elicit ridiculous responses from a computer.
As an IBM document describing the Watson technology points
out: “We have noses that run, and feet that smell. How can a slim
chance and a fat chance be the same, but a wise man and a wise guy
are opposites? How can a house burn up as it burns down? Why do
we fill in a form by filling it out?”15
A Jeopardy! computer would
have to successfully navigate routine language ambiguities of that
type while also exhibiting a level of general understanding far beyond
what you’d typically find in computer algorithms designed to
delve into mountains of text and retrieve relevant answers. As an
example, consider the clue “Sink it & you’ve scratched.” That clue
was presented in a show televised in July 2000 and appeared on
the top row of the game board—meaning that it was considered to
be very easy. Try searching for that phrase using Google, and you’ll
get page after page of links to web pages about removing scratches
9780465059997-text.indd 98 2/12/15 12:46 PM
White-Collar Jobs at Risk 99
from stainless-steel kitchen sinks. (That’s assuming you exclude the
exact match on a website about past Jeopardy! matches.) The correct
response—“What is the cue ball?”—completely eludes Google’s
keyword-based search algorithm.*
All these challenges were well understood by David Ferrucci,
the artificial intelligence expert who eventually assumed leadership
of the team that built Watson. Ferrucci had previously managed a
small group of IBM researchers focused on building a system that
could answer questions provided in natural language format. The
team entered their system, which they named “Piquant,” in a contest
run by the National Bureau of Standards and Technology—the same
government agency that sponsored the machine language contest in
which Google prevailed. In the contest, the competing systems had to
churn through a defined set of about a million documents and come
up with the answers to questions, and they were subject to no time
limit at all. In some cases, the algorithms would grind away for several
minutes before returning an answer.16
This was a dramatically
easier challenge than playing Jeopardy!, where the clues could draw
on a seemingly limitless body of knowledge and where the machine
would have to generate consistently correct responses within a few
seconds in order to have any chance against top human players.
Piquant (as well as its competitors) was not only slow; it was
inaccurate. The system was able to answer questions correctly only
about 35 percent of the time—not an appreciably better success rate
than you could get by simply typing the question into Google’s search
engine.17
When Ferrucci’s team tried to build a prototype Jeopardy!playing
system based on the Piquant project, the results were uniformly
dismal. The idea that Piquant might someday take on a top
Jeopardy! competitor like Ken Jennings seemed laughable. Ferrucci
* In Jeopardy! the clues are considered to be answers and the response must be
phrased as a question for which the provided answer would be correct.
9780465059997-text.indd 99 2/12/15 12:46 PM
RISE OF THE ROBOTS100
recognized that he would have to start from scratch—and that the
project would be a major undertaking spanning as much as half a
decade. He received the green light from IBM management in 2007
and set out to build, in his words, “the most sophisticated intelligence
architecture the world has ever seen.”18
To do this, he drew
on resources from throughout the company and put together a team
consisting of artificial intelligence experts from within IBM as well
as at top universities, including MIT and Carnegie Mellon.19
Ferrucci’s team, which eventually grew to include about twenty
researchers, began by building a massive collection of reference information
that would form the basis for Watson’s responses. This
amounted to about 200 million pages of information, including
dictionaries and reference books, works of literature, newspaper archives,
web pages, and nearly the entire content of Wikipedia. Next
they collected historical data for the Jeopardy! quiz show. Over
180,000 clues from previously televised matches became fodder for
Watson’s machine learning algorithms, while performance metrics
from the best human competitors were used to refine the computer’s
betting strategy.20
Watson’s development required thousands of separate
algorithms, each geared toward a specific task—such as searching
within text; comparing dates, times, and locations; analyzing
the grammar in clues; and translating raw information into properly
formatted candidate responses.
Watson begins by pulling apart the clue, analyzing the words,
and attempting to understand what exactly it should look for. This
seemingly simple step can, in itself, be a tremendous challenge for a
computer. Consider, for example, a clue that appeared in a category
entitled “Lincoln Blogs” and was used in training Watson: “Secretary
Chase just submitted this to me for the third time; guess what, pal.
This time I’m accepting it.” In order to have any chance at responding
correctly, the machine would first need to understand that the initial
instance of the word “this” acts as a placeholder for the answer it
should seek.21
9780465059997-text.indd 100 2/12/15 12:46 PM
White-Collar Jobs at Risk 101
Once it has a basic understanding of the clue, Watson simultaneously
launches hundreds of algorithms, each of which takes a different
approach as it attempts to extract a possible answer from the
massive corpus of reference material stored in the computer’s memory.
In the example above, Watson would know from the category
that “Lincoln” is important, but the word “blogs” would likely be a
distraction: unlike a human, the machine wouldn’t comprehend that
the show’s writers were imagining Abraham Lincoln as a blogger.
As the competing search algorithms reel in hundreds of possible
answers, Watson begins to rank and compare them. One technique
used by the machine is to plug the potential answer into the original
clue so that it forms a statement, and then go back out to the reference
material and look for corroborating text. So if one of the search
algorithms manages to come up with the correct response “resignation,”
Watson might then search its dataset for a statement something
like “Secretary Chase just submitted resignation to Lincoln for the
third time.” It would find plenty of close matches, and the computer’s
confidence in that particular answer would rise. In ranking its
candidate responses, Watson also relies on reams of historical data;
it knows precisely which algorithms have the best track records for
various types of questions, and it listens far more attentively to the
top performers. Watson’s ability to rank correctly worded natural
language answers and then determine whether or not it has sufficient
confidence to press the Jeopardy! buzzer is one of the system’s
defining characteristics, and a quality that places it on the frontier
of artificial intelligence. IBM’s machine “knows what it knows”—
something that comes easily to humans but eludes nearly all computers
when they delve into masses of unstructured information intended
for people rather than machines.
Watson prevailed over Jeopardy! champions Ken Jennings and
Brad Rutter in two matches televised in February 2011, giving IBM
the massive publicity surge it hoped for. Well before the media frenzy
surrounding that remarkable accomplishment began to fade, a far
9780465059997-text.indd 101 2/12/15 12:46 PM
RISE OF THE ROBOTS102
more consequential story began to unfold: IBM launched its campaign
to leverage Watson’s capabilities in the real world. One of the
most promising areas is in medicine. Repurposed as a diagnostic tool,
Watson offers the ability to extract precise answers from a staggering
amount of medical information that might include textbooks, scientific
journals, clinical studies, and even physicians’ and nurses’ notes
for individual patients. No single doctor could possibly approach
Watson’s ability to delve into vast collections of data and discover
relationships that might not be obvious—especially if the information
is drawn from sources that cross boundaries between medical
specialties.*
By 2013, Watson was helping to diagnose problems and
refine patient treatment plans at major medical facilities, including
the Cleveland Clinic and the University of Texas’s MD Anderson
Cancer Center.
As a part of their effort to turn Watson into a practical tool,
IBM researchers confronted one of the primary tenets of the big data
revolution: the idea that prediction based on correlation is sufficient,
and that a deep understanding of causation is usually both unachievable
and unnecessary. A new feature they named “WatsonPaths”
goes beyond simply providing an answer and lets researchers see the
specific sources Watson consulted, the logic it used in its evaluation,
and the inferences it made on its way to generating an answer. In
other words, Watson is gradually progressing toward offering more
insight into why something is true. WatsonPaths is also being used
as a tool to help train medial students in diagnostic techniques. Less
* According to Stephen Baker’s 2011 book Final Jeopardy, the Watson project
leader, David Ferrucci, struggled with intense pain in one of his teeth for months.
After multiple visits to dentists and what ultimately proved to be a completely
unnecessary root canal, Ferrucci was finally—largely by happenstance—referred
to a doctor in a medical specialty unrelated to dentistry, and the problem was
solved. The specific condition was also described in a relatively obscure medical
journal article. It was not lost on Ferrucci that a machine like Watson might have
produced the correct diagnosis almost instantly.
9780465059997-text.indd 102 2/12/15 12:46 PM
White-Collar Jobs at Risk 103
than three years after a team of humans succeeded in building and
training Watson, the tables have—at least to a limited extent—been
turned, and people are now learning from the way the system reasons
when presented with a complex problem.22
Other obvious applications for the Watson system are in areas
like customer service and technical support. In 2013, IBM announced
that it would work with Fluid, Inc., a major provider of online shopping
services and consulting. The project aims to let online shopping
sites replicate the kind of personalized, natural language assistance
you would get from a knowledgeable sales clerk in a retail store. If
you’re going camping and need a tent, you’d be able to say something
like “I am taking my family camping in upstate NY in October and
I need a tent. What should I consider?” You’d then get specific tent
recommendations, as well as pointers to other items that you might
not have considered.23
As I suggested in Chapter 1, it is only a matter
of time before capability of that type becomes available via smart
phones and shoppers are able to access conversational, natural language
assistance while in brick and mortar stores.
MD Buyline, Inc., a company that specializes in providing information
and research about the latest health care technology to hospitals,
likewise plans to use Watson to answer the far more technical questions
that come up when hospitals need to purchase new equipment. The system
would draw on product specifications, prices, and clinical studies
and research to make specific and instant recommendations to doctors
and procurement managers.24
Watson is also looking for a role in the
financial industry, where the system may be poised to provide personalized
financial advice by delving into a wealth of information about
specific customers as well as general market and economic conditions.
The deployment of Watson in customer service call centers is perhaps
the area with the most disruptive near-term potential, and it is likely
no coincidence that within a year of Watson’s triumph on Jeopardy!,
IBM was already working with Citigroup to explore applications for
the system in the company’s massive retail banking operation.25
9780465059997-text.indd 103 2/12/15 12:46 PM
RISE OF THE ROBOTS104
IBM’s new technology is still in its infancy. Watson—as well as
the competing systems that are certain to eventually appear—have
the potential to revolutionize the way questions are asked and answered,
as well as the way information analysis is approached, both
internal to organizations and in engagements with customers. There
is no escaping the reality, however, that a great deal of the analysis
performed by systems of this type would otherwise have been done
by human knowledge workers.
Building Blocks in the Cloud
In November 2013, IBM announced that its Watson system would
move from the specialized computers that hosted the system for the
Jeopardy! matches to the cloud. In other words, Watson would now
reside in massive collections of servers connected to the Internet. Developers
would be able to link directly to the system and incorporate
IBM’s revolutionary cognitive computing technology into custom
software applications and mobile apps. This latest version of Watson
was also more than twice as fast as its Jeopardy!-playing predecessor.
IBM envisions the rapid emergence of an entire ecosystem of
smart, natural language applications—all carrying the “Powered by
Watson” label.26
The migration of leading-edge artificial intelligence capability
into the cloud is almost certain to be a powerful driver of white-collar
automation. Cloud computing has become the focus of intense competition
among major information technology companies, including
Amazon, Google, and Microsoft. Google, for example, offers developers
a cloud-based machine learning application as well as a largescale
compute engine that lets developers solve huge, computationally
intensive problems by running programs on massive supercomputerlike
networks of servers. Amazon is the industry leader in providing
cloud computing services. Cycle Computing, a small company that
specializes in large-scale computing, was able to solve a complex
9780465059997-text.indd 104 2/12/15 12:46 PM
White-Collar Jobs at Risk 105
problem that would have taken over 260 years on a single computer
in just 18 hours by utilizing tens of thousands of the computers that
power Amazon’s cloud service. The company estimates that prior to
the advent of cloud computing, it would have cost as much as $68
million to build a supercomputer capable of taking on the problem.
In contrast, it’s possible to rent 10,000 servers in the Amazon cloud
for about $90 per hour.27
Just as the field of robotics is poised for explosive growth as
the hardware and software components used in designing the machines
become cheaper and more capable, a similar phenomenon is
unfolding for the technology that powers the automation of knowledge
work. When technologies like Watson, deep learning neural
networks, or narrative-writing engines are hosted in the cloud, they
effectively become building blocks that can be leveraged in countless
new ways. Just as hackers quickly figured out that Microsoft’s Kinect
could be used as an inexpensive way to give robots three-dimensional
machine vision, developers will likewise find unforeseen—and perhaps
revolutionary—applications for cloud-based software building
blocks. Each of these building blocks is in effect a “black box”—
meaning that the component can be used by programmers who have
no detailed understanding of how it works. The ultimate result is
sure to be that groundbreaking AI technologies created by teams
of specialists will rapidly become ubiquitous and accessible even to
amateur coders.
While innovations in robotics produce tangible machines that are
often easily associated with particular jobs (a hamburger-making robot
or a precision assembly robot, for example), progress in software
automation will likely be far less visible to the public; it will often
take place deep within corporate walls, and it will have more holistic
impacts on organizations and the people they employ. White-collar
automation will very often be the story of information technology
consultants descending on large organizations and building completely
custom systems that have the potential to revolutionize the
9780465059997-text.indd 105 2/12/15 12:46 PM
RISE OF THE ROBOTS106
way the business operates, while at the same time eliminating the
need for potentially hundreds or even thousands of skilled workers.
Indeed, one of IBM’s stated motivations for creating the Watson
technology was to offer its consulting division—which, together with
software sales, now accounts for the vast majority of the company’s
revenues—a competitive advantage. At the same time, entrepreneurs
are already finding ways to use the same cloud-based building
blocks to create affordable automation products geared toward small
or medium-sized businesses.
Cloud computing has already had a significant impact on information
technology jobs. During the 1990’s tech boom, huge numbers
of well-paying jobs were created as businesses and organizations of
all sizes needed IT professionals to administer and install personal
computers, networks, and software. By the first decade of the twentyfirst
century, however, the trend began to shift as companies were
increasingly outsourcing many of their information technology functions
to huge, centralized computing hubs.
The massive facilities that host cloud computing services benefit
from enormous economies of scale, and the administrative functions
that once kept armies of skilled IT workers busy are now highly
automated. Facebook, for example, employs a smart software application
called “Cyborg” that continuously monitors tens of thousands
of servers, detects problems, and in many cases can perform repairs
completely autonomously. A Facebook executive noted in November
2013 that the Cyborg system routinely solves thousands of problems
that would otherwise have to be addressed manually, and that the
technology allows a single technician to manage as many as 20,000
computers.28
Cloud computing data centers are often built in relatively rural
areas where land and, especially, electric power are plentiful and
cheap. States and local governments compete intensively for the facilities,
offering companies like Google, Facebook, and Apple generous
tax breaks and other financial incentives. Their primary objective, of
9780465059997-text.indd 106 2/12/15 12:46 PM
White-Collar Jobs at Risk 107
course, is to create lots of jobs for local residents—but such hopes are
rarely realized. In 2011, the Washington Post’s Michael Rosenwald
reported that a colossal, billion-dollar data center built by Apple,
Inc., in the town of Maiden, North Carolina, had created only fifty
full-time positions. Disappointed residents couldn’t “comprehend
how expensive facilities stretching across hundreds of acres can create
so few jobs.”29
The explanation, of course, is that algorithms like
Cyborg are doing the heavy lifting.
The impact on employment extends beyond the data centers
themselves to the companies that leverage cloud computing services.
In 2012, Roman Stanek, the CEO of Good Data, a San Francisco
company that uses Amazon’s cloud services to perform data analysis
for about 6,000 clients, noted that “[b]efore, each [client] company
needed at least five people to do this work. That is 30,000 people.
I do it with 180. I don’t know what all those other people will do
now, but this isn’t work they can do anymore. It’s a winner-takes-all
consolidation.”30
The evaporation of thousands of skilled information technology
jobs is likely a precursor for a much more wide-ranging impact on
knowledge-based employment. As Netscape co-founder and venture
capitalist Marc Andreessen famously said, “Software is eating
the world.” More often than not, that software will be hosted in the
cloud. From that vantage point it will eventually be poised to invade
virtually every workplace and swallow up nearly any white-collar job
that involves sitting in front of a computer manipulating information.
Algorithms on the Frontier
If there is one myth regarding computer technology that ought to be
swept into the dustbin it is the pervasive believe that computers can
do only what they are specifically programmed to do. As we’ve seen,
machine learning algorithms routinely churn through data, revealing
statistical relationships and, in essence, writing their own programs
9780465059997-text.indd 107 2/12/15 12:46 PM
RISE OF THE ROBOTS108
on the basis of what they discover. In some cases, however, computers
are pushing even further and beginning to encroach into areas that
nearly everyone assumes are the exclusive province of the human
mind: machines are starting to demonstrate curiosity and creativity.
In 2009, Hod Lipson, the director of the Creative Machines Lab
at Cornell University, and PhD student Michael Schmidt built a system
that has proved capable of independently discovering fundamental
natural laws. Lipson and Schmidt started by setting up a double
pendulum—a contraption that consists of one pendulum attached
to, and dangling below, another. When both pendulums are swinging,
the motion is extremely complex and seemingly chaotic. Next
they used sensors and cameras to capture the pendulum’s motion and
produce a stream of data. Finally, they gave their software the ability
to control the starting position of the pendulum; in other words,
they created an artificial scientist with the ability to conduct its own
experiments.
They turned their software loose to repeatedly release the pendulum
and then sift through the resulting motion data and try to figure
out the mathematical equations that describe the pendulum’s behavior.
The algorithm had complete control over the experiment; for
each repetition, it decided how to position the pendulum for release,
and it did not do this randomly—it performed an analysis and then
chose the specific starting point that would likely provide the most
insight into the laws underlying the pendulum’s motion. Lipson notes
that the system “is not a passive algorithm that sits back, watching.
It asks questions. That’s curiosity.”31
The program, which they later
named “Eureqa,” took only a few hours to come up with a number of
physical laws describing the movement of the pendulum—including
Newton’s Second Law—and it was able to do this without being
given any prior information or programming about physics or the
laws of motion.
Eureqa uses genetic programming, a technique inspired by biological
evolution. The algorithm begins by randomly combining
9780465059997-text.indd 108 2/12/15 12:46 PM
White-Collar Jobs at Risk 109
various mathematical building blocks into equations and then testing
to see how well the equations fit the data.*
Equations that fail the
test are discarded, while those that show promise are retained and
recombined in new ways so that the system ultimately converges on
an accurate mathematical model.32
The process of finding an equation
that describes the behavior of a natural system is by no means
a trivial exercise. As Lipson says, “[P]reviously, coming up with a
predictive model could take a [scientist’s] whole career.”33
Schmidt
adds that “[p]hysicists like Newton and Kepler could have used a
computer running this algorithm to figure out the laws that explain
a falling apple or the motion of the planets with just a few hours of
computation.”34
When Schmidt and Lipson published a paper describing their
algorithm, they were deluged with requests for access to the software
from other scientists, and they decided to make Eureqa available over
the Internet in late 2009. The program has since produced a number
of useful results in a range of scientific fields, including a simplified
equation describing the biochemistry of bacteria that scientists are
still struggling to understand.35
In 2011, Schmidt founded Nutonian,
Inc., a Boston-area start-up company focused on commercializing
Eureqa as a big data analysis tool for both business and academic
applications. One result is that Eureqa—like IBM’s Watson—is now
hosted in the cloud and is available as an application building block
to other software developers.
Most of us quite naturally tend to associate the concept of creativity
exclusively with the human brain, but it’s worth remembering that the
* This is significantly more advanced than the commonly used statistical technique
known as “regression.” With regression (either linear or nonlinear), the
form of the equation is set in advance, and the equation’s parameters are optimized
so as to fit the data. The Eureqa program, in contrast, is able to independently
determine equations of any form using a variety of mathematical
components including arithmetic operators, trigonometric and logarithmic
functions, constants, etc.
9780465059997-text.indd 109 2/12/15 12:46 PM
RISE OF THE ROBOTS110
brain itself—by far the most sophisticated invention in existence—
is the product of evolution. Given this, perhaps it should come as
no surprise that attempts to build creative machines very often incorporate
genetic programming techniques. Genetic programming
essentially allows computer algorithms to design themselves through
a process of Darwinian natural selection. Computer code is initially
generated randomly and then repeatedly shuffled using techniques
that emulate sexual reproduction. Every so often, a random mutation
is thrown in to help drive the process in entirely new directions. As
new algorithms evolve, they are subjected to a fitness test that leads
to either their survival, or—far more often—their demise. Computer
scientist and consulting Stanford professor John Koza is one of the
leading researchers in the field and has done extensive work using
genetic algorithms as “automated invention machines.”*
Koza has
isolated at least seventy-six cases where genetic algorithms have produced
designs that are competitive with the work of human engineers
and scientists in a variety of fields, including electric circuit design,
mechanical systems, optics, software repair, and civil engineering.
In most of these cases, the algorithms have replicated existing designs,
but there are at least two instances where genetic programs
have created new, patentable inventions.36
Koza argues that genetic
algorithms may have an important advantage over human designers
because they are not constrained by preconceptions; in other words,
they may be more likely to result in an “outside-the-box” approach
to the problem.37
Lipson’s suggestion that Eureqa exhibits curiosity and Koza’s
argument about computers acting without preconceptions suggest
that creativity may be something that is within reach of a computer’s
* In addition to his work in genetic programming, Koza is the inventor of the
scratch-off lottery ticket and the originator of the “constitutional workaround”
idea to elect US presidents by popular vote by having the states agree to award
electoral-college votes based on the country’s overall popular-vote outcome.
9780465059997-text.indd 110 2/12/15 12:46 PM
White-Collar Jobs at Risk 111
capabilities. The ultimate test of such an idea might be to see if a
computer could create something that humans would accept as a work
of art. Genuine artistic creativity—perhaps more so than any other
intellectual endeavor—is something we associate exclusively with the
human mind. As Time’s Lev Grossman says, “Creating a work of art
is one of those activities we reserve for humans and humans only. It’s
an act of self-expression; you’re not supposed to be able to do it if you
don’t have a self.”38
Embracing the possibility that a computer could
be a legitimate artist would require a fundamental reevaluation of
our assumptions about the nature of machines.
In the 2004 film I, Robot, the protagonist, played by Will Smith,
asks a robot, “Can a robot write a symphony? Can a robot turn a
canvas into a beautiful masterpiece?” The robot’s reply “Can you?” is
meant to suggest that, well, the vast majority of people can’t do those
things either. In the real world of 2015, however, Smith’s question
would elicit a more forceful answer: “Yes.”
In July 2012, the London Symphony Orchestra performed a composition
entitled Transits—Into an Abyss. One reviewer called it
“artistic and delightful.”39
The event marked the first time that an
elite orchestra had played music composed entirely by a machine. The
composition was created by Iamus, a cluster of computers running
a musically inclined artificial intelligence algorithm. Iamus, which
is named after a character from Greek mythology who was said to
understand the language of birds, was designed by researchers at
the University of Malaga in Spain. The system begins with minimal
information, such as the type of instruments that will play the music,
and then, with no further human intervention, creates a highly complex
composition—which can often evoke an emotional response in
audiences—within minutes. Iamus has already produced millions of
unique compositions in the modernist classical style, and is likely to
be adapted to other musical genres in the future. Like Eureqa, Iamus
has resulted in a start-up company to commercialize the technology.
Melomics Media, Inc., has been set up to sell the music from an
9780465059997-text.indd 111 2/12/15 12:46 PM
RISE OF THE ROBOTS112
iTunes-like online store. The difference is that compositions created
by Iamus are offered on a royalty-free basis, allowing purchasers to
use the music in any way they wish.
Music is not the only art form being created by computers. Simon
Colton, a professor of creative computing at the University of London,
has built an artificial intelligence program called “The Painting
Fool” that he hopes will someday be taken seriously as a painter (see
Figure 4.1). “The goal of the project is not to produce software that
can make photos look like they’ve been painted; Photoshop has done
that for years,” Colton says. “The goal is to see whether software can
be accepted as creative in its own right.”40
Colton has built a set of capabilities he calls “appreciative and
imaginative behaviors” into the system. The Painting Fool software
can identify emotions in photographs of people and then paint an
abstract portrait that attempts to convey their emotional state. It
can also generate imaginary objects using techniques based on genetic
programming. Colton’s software even has the ability to be selfcritical.
It does this by incorporating another software application
called “Darci” that was built by researchers at Brigham Young
Figure 4.1. An Original Work of Art Created by Software
© ThePaintingFool.com
9780465059997-text.indd 112 2/12/15 12:46 PM
White-Collar Jobs at Risk 113
University. The Darci developers started with a database of paintings
that had been labeled by humans with adjectives like “dark,” “sad,”
or “inspiring.” They then trained a neural network to make the associations
and turned it loose to label new paintings. The Painting
Fool is able to use feedback from Darci to decide whether or not it is
achieving its objectives as it paints.41
My point here is not to suggest that large numbers of artists
or musical composers will soon be out of a job. Rather, it is that
the techniques used to build creative software—many of which, as
we have seen, rely on genetic programming—can be repurposed in
countless new ways. If computers can create musical compositions or
design electronic components, then it seems likely that they will soon
be able to formulate a new legal strategy or perhaps come up with
a new way to approach a management problem. For the time being,
the white-collar jobs at highest risk will continue to be those that are
the most routine or formulaic—but the frontier is advancing quickly.
Nowhere is the rapid pace of that advance more evident than
on Wall Street. Where once financial trading was highly dependent
on direct communication between people, either in bustling trading
pits or via telephone, it has now come to be largely dominated by
machines communicating over fiber-optic links. By some estimates,
automated trading algorithms are now responsible for at least half,
and perhaps as much as 70 percent, of stock market transactions.
These sophisticated robotic traders—many of which are powered by
techniques on the frontier of artificial intelligence research—go far
beyond simply executing routine trades. They attempt to profit by detecting
and then snapping up shares in front of huge transactions initiated
by mutual funds and pension managers. They seek to deceive
other algorithms by inundating the system with decoy bids that are
then withdrawn within tiny fractions of a second. Both Bloomberg
and Dow News Service offer special machine-readable products designed
to feed the algorithms’ voracious appetites for financial news
that they can—perhaps within milliseconds—turn into profitable
9780465059997-text.indd 113 2/12/15 12:46 PM
RISE OF THE ROBOTS114
trades. The news services also provide real-time metrics that let the
machines see which items are attracting the most attention.42
Twitter,
Facebook, and the blogosphere are likewise all fodder for these competing
algorithms. In a 2013 paper published in the scientific journal
Nature, a group of physicists studied global financial markets and
identified “an emerging ecology of competitive machines featuring
‘crowds’ of predatory algorithms,” and suggested that robotic trading
had progressed beyond the control—and even comprehension—of
the humans who designed the systems.43
In the realm inhabited by these continuously battling algorithms,
the action unfolds at a pace that would be incomprehensible to the
fastest human trader. Indeed, speed—in some cases measured in millionths
or even billionths of a second—is so critical to algorithmic
trading success that Wall Street firms have collectively invested billions
of dollars to build computing facilities and communications
paths designed to produce tiny speed advantages. In 2009, for example,
a company called Spread Networks spent as much as $200
million to lay down a new fiber-optic cable link stretching 825 miles
in a straight line from Chicago to New York. The company operated
in stealth mode so as not to alert the competition even as it blasted
its way through the Allegheny Mountains. When the new fiberoptic
path came online, it offered a speed advantage of perhaps three
or four thousandths of a second compared with existing communications
routes. That was enough to allow any algorithmic trading
systems employing the new route to effectively dominate their competition.
Wall Street firms, faced with algorithmic decimation, lined
up to lease bandwidth—reportedly at a cost as much as ten times
that of the original, slower cable. A similar cable stretching across the
Atlantic between London and New York is currently in progress, and
is expected to shave about five thousandths of a second off current
execution times.44
The impact of all this automation is clear: even as the stock market
continued on its upward trajectory in 2012 and 2013, large Wall
9780465059997-text.indd 114 2/12/15 12:46 PM
White-Collar Jobs at Risk 115
Street banks announced massive layoffs, often resulting in the elimination
of tens of thousands of jobs. At the turn of the twenty-first
century, Wall Street firms employed nearly 150,000 financial workers
in New York City; by 2013, the number was barely more than
100,000—even as both the volume of transactions and the industry’s
profits soared.45
Against the backdrop of that overall collapse in employment,
Wall Street did create at least one very high-profile job: in
late 2012, David Ferrucci, the computer scientist who led the effort
to build Watson, left IBM for a new gig at a Wall Street hedge fund,
where he’ll be applying the latest advances in artificial intelligence to
modeling the economy—and, presumably, trying to gain a competitive
advantage for his firm’s trading algorithms.46
Offshoring and High-Skill Jobs
While the trend toward increased automation of white-collar jobs
is clear, the most dramatic onslaught—especially for truly skilled
professions—still lies in the future. The same cannot necessarily be
said for the practice of offshoring, where knowledge jobs are moved
electronically to lower-wage countries. Highly educated and skilled
professionals such as lawyers, radiologists, and especially computer
programmers and information technology workers have already felt a
significant impact. In India, for example, there are armies of call center
workers and IT professionals, as well as tax preparers versed in the
US tax code and attorneys specifically trained not in their own country’s
legal system but in American law, and standing ready to perform
low-cost legal research for US firms engaged in domestic litigation.
While the offshoring phenomenon may seem completely unrelated
to the jobs lost to computers and algorithms, the precise opposite is
true: offshoring is very often a precursor of automation, and the jobs it
creates in low-wage nations may prove to be short-lived as technology
advances. What’s more, advances in artificial intelligence may make it
even easier to offshore jobs that can’t yet be fully automated.
9780465059997-text.indd 115 2/12/15 12:46 PM
RISE OF THE ROBOTS116
Most economists view the practice of offshoring as just another
example of global trade and argue that it invariably makes both
parties to the transaction better off. Harvard professor N. Gregory
Mankiw, for example, while serving as George W. Bush’s chairman
of the White House Council of Economic Advisers, said in 2004
that offshoring is “the latest manifestation of the gains from trade
that economists have talked about at least since Adam Smith.”47
Abundant evidence argues to the contrary. Trade in tangible goods
creates a great many peripheral jobs in areas like shipping, distribution,
and retail. There are also natural forces that tend to mitigate the
impact of globalization to some degree; for example, a company that
chooses to move a factory to China incurs both shipping costs and a
significant delay before completed products reach consumer markets.
Electronic offshoring, in contrast, is almost completely frictionless
and subject to none of these penalties. Jobs are moved to low-wage
locations instantly and at minimal cost. If peripheral jobs are created,
it is much more likely to be in the country where the workers reside.
I would argue that “free trade” is the wrong lens through which
to view offshoring. Instead, it is much more akin to virtual immigration.
Suppose, for example, that a huge customer service call center
were to be built south of San Diego, just across the border from
Mexico. Thousands of low-wage workers are issued “day worker”
passes and are bused across the border to staff the call center every
morning. At the end of the workday, the buses travel in the opposite
direction. What is the difference between this situation (which would
certainly be viewed as an immigration issue) and moving the jobs
electronically to India or the Philippines? In both cases, workers are,
in effect, “entering” the United States to offer services that are clearly
directed at the domestic US economy. The biggest difference is that
the Mexican day worker plan would probably be significantly better
for the California economy. There might be jobs for bus drivers, and
there would certainly be jobs for people to maintain the huge facility
located on the US side of the border. Some of the workers might
9780465059997-text.indd 116 2/12/15 12:46 PM
White-Collar Jobs at Risk 117
purchase lunch or even a cup of coffee while at work, thus injecting
consumer demand into the local economy. The company that owned
the California facility would pay property tax. When the jobs are offshored,
and the workers enter the United States virtually, the domestic
economy receives none of these benefits. I find it somewhat ironic
that many conservatives in the United States are adamant about securing
the border against immigrants who will likely take jobs that
few Americans want, while at the same time expressing little concern
that the virtual border is left completely open to higher-skill workers
who take jobs that Americans definitely do want.
The argument put forth by economists like Mankiw, of course,
measures in the aggregate and glosses over the highly disproportionate
impact that offshoring has on the groups of people who either
suffer or benefit from the practice. On the one hand, a relatively
small but still significant group of people—potentially measured in
the millions—may be subjected to a substantial downgrade in their
income, quality of life, and future prospects. Many of these people
may have made substantial investments in education and training.
Some workers may lose their income entirely. Mankiw would likely
argue that the aggregate benefit to consumers makes up for these
losses. Unfortunately, although consumers may benefit from lower
prices as a result of the offshoring, this savings may be spread across a
population of tens or even hundreds of millions of people, perhaps
resulting in a cost reduction that amounts to mere pennies and has a
negligible effect on any one individual’s well-being. And, needless to
say, not all the gains will flow to consumers; a significant fraction
will end up in the pockets of a few already-wealthy executives, investors,
and business owners. This asymmetric impact is, perhaps not
surprisingly, intuitively grasped by most average workers but seemingly
lost on many economists.
One of the few economists to recognize offshoring’s disruptive
potential is the former vice chairman of the Federal Reserve’s
Board of Governors, Alan Blinder, who wrote a 2007 op-ed in the
9780465059997-text.indd 117 2/12/15 12:46 PM
RISE OF THE ROBOTS118
Washington Post entitled “Free Trade’s Great, but Offshoring Rattles
Me.”48
Blinder has conducted a number of surveys aimed at assessing
the future impact of offshoring and has estimated that 30–40 million
US jobs—positions employing roughly a quarter of the workforce—
are potentially offshorable. As he says, “We have so far barely seen
the tip of the offshoring iceberg, the eventual dimensions of which
may be staggering.”49
Virtually any occupation that primarily involves manipulating
information and is not in some way anchored locally—for example,
with a requirement for face-to-face interaction with customers—is
potentially at risk from offshoring in the relatively near future and
then from full automation somewhat further out. Full automation is
simply the logical next step. As technology advances, we can expect
that more and more of the routine tasks now performed by offshore
workers will eventually be handled entirely by machines. This has
already occurred with respect to some call center workers who have
been replaced by voice automation technology. As truly powerful
natural language systems like IBM’s Watson move into the customer
service arena, huge numbers of offshore call center jobs are poised
to be vaporized.
As this process unfolds, it seems likely that those companies—
and nations—that have invested heavily in offshoring as a route to
profitability and prosperity will have little choice but to move up
the value chain. As more routine jobs are automated, higher-skill,
professional jobs will be increasingly in the sights of the offshorers.
One factor that is, I think, underappreciated is the extent to which
advances in artificial intelligence as well as the big data revolution
may act as a kind of catalyst, making a much broader range of highskill
jobs potentially offshorable. As we’ve seen, one of the tenets of
the big data approach to management is that insights gleaned from
algorithmic analysis can increasingly substitute for human judgment
and experience. Even before advancing artificial intelligence applications
reach the stage where full automation is possible, they will
9780465059997-text.indd 118 2/12/15 12:46 PM
White-Collar Jobs at Risk 119
become powerful tools that encapsulate ever more of the analytic intelligence
and institutional knowledge that give a business its competitive
advantage. A smart young offshore worker wielding such tools
might soon be competitive with far more experienced professionals
in developed countries who command very high salaries.
When offshoring is viewed in combination with automation, the
potential aggregate impact on employment is staggering. In 2013,
researchers at the University of Oxford’s Martin School conducted
a detailed study of over seven hundred US job types and came to the
conclusion that nearly 50 percent of jobs will ultimately be susceptible
to full machine automation.50
Alan Blinder and Alan Krueger
of Princeton University conducted a similar analysis with respect to
offshoring and found that about 25 percent of US jobs are at risk of
eventually being moved to low-wage countries.51
Let’s hope there’s
significant overlap between those two estimates! Indeed, in all likelihood
there is plenty of overlap when the estimates are viewed in
terms of job titles or descriptions. The story is different along the
time dimension, however. Offshoring will often arrive first; to a significant
degree, it will accelerate the impact of automation even as it
drags higher-skill jobs into the threat zone.
As powerful AI-based tools make it easier for offshore workers to
compete with their higher-paid counterparts in developed countries,
advancing technology is also likely to upend many of our most basic
assumptions about which types of jobs are potentially offshorable.
Nearly everyone believes, for example, that occupations that require
physical manipulation of the environment will always be safe. Yet,
military pilots located in the western United States routinely operate
drone aircraft in Afghanistan. By the same token, it is easy to
envision remote-controlled machinery being operated by offshore
workers who provide the visual perception and dexterity that, for the
time being, continues to elude autonomous robots. A need for
face-to-face interaction is another factor that is assumed to anchor a
job locally. However, telepresence robots are pushing the frontier in
9780465059997-text.indd 119 2/12/15 12:46 PM
RISE OF THE ROBOTS120
this area and have already been used to offshore English language
instruction from Korean schools to the Philippines. In the not too
distant future, advanced virtual reality environments will likewise
make it even easier for workers to move seamlessly across national
borders and engage directly with customers or clients.
As offshoring accelerates, college graduates in the United States
and other advanced countries may face daunting competition based
not just on wages but also on cognitive capability. The combined population
of India and China amounts to roughly 2.6 billion people—or
over eight times the population of the United States. The top 5 percent
in terms of cognitive ability amounts to about 130 million people—
or over 40 percent of the entire US population. In other words, the
inescapable reality of the bell-curve distribution stipulates that there
are far more very smart people in India and China than in the United
States. That will not necessarily be a cause for concern, of course,
as long as the domestic economies in those countries are capable of
creating opportunities for all those smart workers. The evidence so
far, however, suggests otherwise. India has built a major, nationally
strategic industry specifically geared toward the electronic capture
of American and European jobs. And China, even as the growth
rate of its economy continues to be the envy of the world, struggles
year after year to create sufficient white-collar jobs for its soaring
population of new college graduates. In mid-2013, Chinese authorities
acknowledged that only about half of the country’s current crop
of college graduates had been able to find jobs, while more than 20
percent of the previous year’s graduates remained unemployed—and
those figures are inflated when temporary and freelance work, as
well as enrollment in graduate school and government-mandated
“make work” positions, are regarded as full employment.52
Thus far, a lack of proficiency in English and other European
languages has largely prevented skilled workers in China from competing
aggressively in the offshoring industry. Once again, however,
technology seems likely to eventually demolish this barrier.
9780465059997-text.indd 120 2/12/15 12:46 PM
White-Collar Jobs at Risk 121
Technologies like deep learning neural networks are poised to transport
instantaneous machine voice translation from the realm of science
fiction into the real world—and this could happen within the
next few years. In June 2013, Hugo Barra, Google’s top Android
executive, indicated that he expects a workable “universal translator”
that could be used either in person or over the phone to be
available within several years. Barra also noted that Google already
has “near perfect” real-time voice translation between English and
Portuguese.53
As more and more routine white-collar jobs fall to
automation in countries throughout the world, it seems inevitable
that competition will intensify to land one of the dwindling number
of positions that remain beyond the reach of the machines. The very
smartest people will have a significant advantage, and they won’t
hesitate to look beyond national borders. In the absence of barriers
to virtual immigration, the employment prospects for nonelite
college-educated workers in developed economies could turn out to
be pretty grim.
Education and Collaboration with the Machines
As technology advances and more jobs become susceptible to automation,
the conventional solution has always been to offer workers
more education and training so that they can step into to new, higherskill
roles. As we saw in Chapter 1, millions of lower-skill jobs in
areas like fast food and retail are at risk as robots and self-service
technologies begin to encroach aggressively in these areas. We can
be sure that more education and training will be the primary proffered
solution for these workers. Yet, the message of this chapter has
been that the ongoing race between technology and education may
well be approaching the endgame: the machines are coming for the
higher-skill jobs as well.
Among economists who are tuned in to this trend, a new flavor
of conventional wisdom is arising: the jobs of the future will involve
9780465059997-text.indd 121 2/12/15 12:46 PM
RISE OF THE ROBOTS122
collaborating with the machines. Erik Brynjolfsson and Andrew
McAfee of the Massachusetts Institute of Technology have been especially
strong proponents of this idea, advising workers that they
should learn to “race with the machines”—rather than against them.
While that may well be sage advice, it is nothing especially new.
Learning to work with the prevailing technology has always been a
good career strategy. We used to call it “learning computer skills.”
Nevertheless, we should be very skeptical that this latest iteration will
prove to be an adequate solution as information technology continues
on its relentless exponential path.
The poster child for the machine-human symbiosis idea has come
to be the relatively obscure game of freestyle chess. More than a decade
after IBM’s Deep Blue computer defeated world chess champion
Garry Kasparov, it is generally accepted that, in one-on-one contests
between computers and humans, the machines now dominate absolutely.
Freestyle chess, however, is a team sport. Groups of people,
who are not necessarily world-class chess players individually, compete
against each other and are allowed to freely consult with computer
chess programs as they evaluate each move. As things stand in
2014, human teams with access to multiple chess algorithms are able
to outmatch any single chess-playing computer.
There are a number of obvious problems with the idea that humanmachine
collaboration, rather than full automation, will come to
dominate the workplaces of the future. The first is that the continued
dominance of human-machine teams in freestyle chess is by no
means assured. To me, the process that these teams use—evaluating
and comparing the results from different chess algorithms before
deciding on the best move—seems uncomfortably close to what IBM
Watson does when it fires off hundreds of information-seeking algorithms
and then succeeds in ranking the results. I don’t think it is
much of a stretch to suggest that a “meta” chess-playing computer
with access to multiple algorithms may ultimately defeat the human
teams—especially if speed is an important factor.
9780465059997-text.indd 122 2/12/15 12:46 PM
White-Collar Jobs at Risk 123
Secondly, even if the human-machine team approach does offer
an incremental advantage going forward, there is an important question
as to whether employers will be willing to make the investment
necessary to leverage that advantage. In spite of the mottos and slogans
that corporations direct at their employees, the reality is that
most businesses are not prepared to pay a significant premium for
“world-class” performance when it comes to the bulk of the more
routine work required in their operations. If you have any doubts about
this, I’d suggest trying to call your cable company. Businesses will make
the investment in areas that are critical to their core competency—in
other words, the activities that give the business a competitive advantage.
Again, this scenario is nothing new. And, more importantly, it
doesn’t really involve any new people. The individuals that businesses
are likely to hire and then couple with the best available technology
are the same people who are largely immune to unemployment today.
It is a small population of elite workers. Economist Tyler Cowen’s
2013 book Average Is Over quotes one freestyle chess insider who
says that the very best players are “genetic freaks.”54
That hardly
makes the machine collaboration idea sound like a systemic solution
for masses of people pushed out of routine jobs. And, as we have just
seen, there is also the problem of offshoring. A great many of those
2.6 billion people in India and China are going to be pretty eager to
grab one of those elite jobs.
There are also good reasons to expect that many machine collaboration
jobs will be relatively short-lived. Recall the example of
WorkFusion and how the company’s machine learning algorithms incrementally
automate the work performed by freelancers. The bottom
line is that if you find yourself working with, or under the direction
of, a smart software system, it’s probably a pretty good bet that—
whether you’re aware of it or not—you are also training the software
to ultimately replace you.
Yet another observation is that, in many cases, those workers
who seek a machine collaboration job may well be in for a “be careful
9780465059997-text.indd 123 2/12/15 12:46 PM
RISE OF THE ROBOTS124
what you wish for” epiphany. As one example, consider the current
trends in legal discovery. When corporations engage in litigation, it
becomes necessary to sift through enormous numbers of internal
documents and decide which ones are potentially relevant to the case
at hand. The rules require these to be provided to the opposing side,
and there can be substantial legal penalties for failing to produce anything
that might be pertinent. One of the paradoxes of the paperless
office is that the sheer number of such documents, especially in the
form of emails, has grown dramatically since the days of typewriters
and paper. To deal with this overwhelming volume, law firms are
employing new techniques.
The first approach involves full automation. So-called e-Discovery
software is based on powerful algorithms that can analyze millions of
electronic documents and automatically tease out the relevant ones.
These algorithms go far beyond simple key-word searches and often
incorporate machine learning techniques that can isolate relevant
concepts even when specific phrases are not present.55
One direct
result has been the evaporation of large numbers of jobs for lawyers
and paralegals who once would have sorted laboriously through
cardboard boxes full of paper documents.
There is also a second approach in common use: law firms may
outsource this discovery work to specialists who hire legions of recent
law school graduates. These graduates are typically victims of the
bursting law school enrollment bubble. Unable to find employment
as full-fledged lawyers—and often burdened with enormous student
loans—they instead work as document reviewers. Each attorney sits
in front of a monitor where a continuous stream of documents is
displayed. Along with the document, there are two buttons: “Relevant”
and “Not Relevant.” The law school graduates scan the document
on the screen and click the proper button. A new document
then appears.56
They may be expected to categorize up to eighty
documents per hour.57
For these young attorneys, there are no courtrooms,
no opportunity to learn or to grow in their profession, and no
9780465059997-text.indd 124 2/12/15 12:46 PM
White-Collar Jobs at Risk 125
opportunity for advancement. Instead, there are—hour after hour—
the “Relevant” and “Not Relevant” buttons.*
One obvious question regarding these two competing approaches
is whether the collaboration model is sustainable. Even at the relatively
low wages (for attorneys) commanded by these workers, the
automated approach seems far more cost-effective. As to the low
quality of these jobs, you might assume that I’ve simply cherry-picked
a rather dystopian example. After all, won’t most jobs that involve
collaboration with machines put people in control—so that workers
supervise the machines and engage in rewarding work, rather than
simply acting as gears and cogs in a mechanized process?
The problem with this rather wishful assumption is that the data
does not support it. In his 2007 book Super Crunchers, Yale University
professor Ian Ayres cites study after study showing that algorithmic
approaches routinely outperform human experts. When people,
rather than computers, are given overall control of the process, the
results almost invariably suffer. Even when human experts are given
access to the algorithmic results in advance, they still produce outcomes
that are inferior to the machines acting autonomously. To the
extent that people add value to the process, it is better to have them
provide specific inputs to the system instead of giving them overall
control. As Ayres says, “Evidence is mounting in favor of a different
and much more demeaning, dehumanizing mechanism for combining
expert and [algorithmic] expertise.”58
* If you find this type of work appealing but lack the requisite legal training,
be sure to check out Amazon’s “Mechanical Turk” service, which offers many
similar opportunities. “BinCam,” for example, places cameras in your garbage
bin, tracks everything you throw away, and then automatically posts the record
to social media. The idea is, apparently, to shame yourself into not wasting food
and not forgetting to recycle. As we’ve seen, visual recognition (of types of garbage,
in this case) remains a daunting challenge for computers, so people are
employed to perform this task. The very fact that this service is economically
viable should give you some idea of the wage level for this kind of work.
9780465059997-text.indd 125 2/12/15 12:46 PM
RISE OF THE ROBOTS126
My point here is that while human-machine collaboration jobs
will certainly exist, they seem likely to be relatively few in number*
and often short-lived. In a great many cases, they may also be unrewarding
or even dehumanizing. Given this, it seems difficult to
justify suggesting that we ought to make a major effort to specifically
educate people in ways that will help them land one of these
jobs—even if it were possible to pin down exactly what such training
might entail. For the most part, this argument strikes me as a way to
patch the tires on a very conventional idea (give workers still more
vocational training) and keep it rolling for a bit longer. We are ultimately
headed for a disruption that will demand a far more dramatic
policy response.
SOME OF THE FIRST JOBS to fall to white-collar automation are sure
to be the entry-level positions taken by new college graduates. As
we saw in Chapter 2, there is already evidence to suggest that this
process is well under way. Between 2003 and 2012, the median income
of US college graduates with bachelor’s degrees fell from nearly
$52,000 to just over $46,000, measured in 2012 dollars. During the
same period, total student loan debt tripled from about $300 billion
to $900 billion.59
Underemployment among recent graduates is rampant, and
nearly every college student seemingly knows someone whose degree
has led to a career working at a coffee shop. In March 2013,
* In Average Is Over, Tyler Cowen estimates that perhaps 10–15 percent of the
American workforce will be well equipped for machine collaboration jobs. I
think that in the long run, even that estimate might be optimistic, especially
when you consider the impact of offshoring. How many machine collaboration
jobs will also be anchored locally? (One exception to my skepticism about
machine collaboration jobs may be in health care. As discussed in Chapter 6,
I think it might eventually be possible to create a new type of medical professional
with far less training than a doctor who would work together with an AIbased
diagnostic and treatment system. Health care is a special case, however,
because doctors require an extraordinary amount of training and there is likely
to be a significant shortage of physicians in the future.)
9780465059997-text.indd 126 2/12/15 12:46 PM
White-Collar Jobs at Risk 127
Canadian economists Paul Beaudry, David A. Green, and Benjamin
M. Sand published an academic paper entitled “The Great Reversal
in the Demand for Skill and Cognitive Tasks.”60
That title essentially
says it all: the economists found that around the year 2000, overall
demand for skilled labor in the United States peaked and then went
into precipitous decline. The result is that new college graduates have
increasingly been forced into relatively unskilled jobs—often displacing
nongraduates in the process.
Even those graduates with degrees in scientific and technical
fields have been significantly impacted. As we’ve seen, the information
technology job market, in particular, has been transformed by
the increased automation associated with the trend toward cloud
computing as well as by offshoring. The widely held belief that a degree
in engineering or computer science guarantees a job is largely a
myth. An April 2013 analysis by the Economic Policy Institute found
that at colleges in the United States, the number of new graduates
with engineering and computer science degrees exceeds the number
of graduates who actually find jobs in these fields by 50 percent. The
study concludes that “the supply of graduates is substantially larger
than the demand for them in industry.”61
It is becoming increasingly
clear that a great many people will do all the right things in terms
of pursuing an advanced education, but nonetheless fail to find a
foothold in the economy of the future.
While some of the economists who focus their efforts on sifting
through reams of historical data are finally beginning to discern
the impact that advancing technology is having on higher-skill jobs,
they are typically quite cautious about attempting to project that
trend into the future. Researchers working in the field of artificial
intelligence are often far less reticent. Noriko Arai, a mathematician
with Japan’s National Institute of Informatics, is leading a project to
develop a system capable of passing the Tokyo University entrance
examination. Arai believes that if a computer can demonstrate the
combination of natural language aptitude and analytic skill necessary
9780465059997-text.indd 127 2/12/15 12:46 PM
RISE OF THE ROBOTS128
to gain entrance to Japan’s highest-ranked university, then it will
very likely also be able to eventually perform many of the jobs taken
by college graduates. She foresees the possibility of massive job displacement
within the next ten to twenty years. One of the primary
motivations for her project is to try to quantify the potential impact
of artificial intelligence on the job market. Arai worries that 10 to
20 percent of skilled workers replaced by automation would be a
“catastrophe” and says she “can’t begin to think what 50 percent
would mean.” She then adds that it would be “way beyond a catastrophe
and such numbers can’t be ruled out if AI performs well
in the future.”62
The higher-education industry itself has historically been one of
the primary employment sectors for highly skilled workers. Especially
for those who aspire to a doctoral degree, a typical career path
has been to arrive on campus as a college freshman—and then never
really leave. In the next chapter we’ll look at how that industry, and
a great many careers, may also be on the verge of a massive technological
disruption.
9780465059997-text.indd 128 2/12/15 12:46 PM