Report - How Robots Change The World

HOW ROBOTS
CHANGE
THE WORLD
WHAT AUTOMATION REALLY MEANS
FOR JOBS AND PRODUCTIVITY
JUNE 2019
How Robots Change the World
TABLE OF CONTENTS
Foreword 3
Executive summary 4
Introduction 11
What drives the robot rise? 13

Three reasons for the robot surge 16
The impact of robots on manufacturing jobs 19

Global impacts 19
Regional impacts 22
The Robot Vulnerability Index 25

United States 27
Germany 28
United Kingdom 29
France 30
Japan 31
South Korea 32
Australia 33
The robotics dividend 35

Reshaping the labour market 37
Robots are coming to the service sector 40

The future impact of robots on five key service industries 43
Where service robots go from here 49
How to respond to the rise of robots 51

A framework for action 53
Appendix: econometric analysis 56
1
Fieldwork Robotics trialling
a robot raspberry harvesting
system on a British farm, 2019.
FOREWORD:
The Shape of Things to Come
The robotics revolution is on poorer local economies. In
rapidly accelerating, as fast- many places, the impact will
paced technological advances aggravate social and economic
in automation, engineering, stresses from unemployment
energy storage, artificial and income inequality in times
intelligence, and machine when increasing political
learning converge. The result polarisation is already a
will transform the capabilities worrying trend.
of robots and their ability to
take over tasks once carried At Oxford Economics our
out by humans. mission is to help our clients
better understand an ever-
The number of robots in more complex and fast-
Adrian Cooper use worldwide multiplied changing world economy, in
CEO and Chief Economist three-fold over the past two all its dimensions—and how to
Oxford Economics decades, to 2.25 million. Trends successfully operate in it. Our
suggest the global stock of clients look to us to explain the
robots will multiply even faster forces shaping their economic
in the next 20 years, reaching environment, help them
as many as 20 million by 2030, anticipate the future, and plan
with 14 million in China alone. for its uncertainties.
The implications are immense,
and the emerging challenges That is why we brought
for governments and policy- together a team of our
makers are equally daunting in economists, econometricians,
their scale. modellers and technology
experts from across our
The rise of the robots will worldwide network of over
boost productivity and 250 analysts to conduct an
economic growth. It will lead, extensive research study
too, to the creation of new to analyse the robotics
jobs in yet-to-exist industries, phenomenon. We are pleased
in a process of ‘creative to share our findings not
destruction.’ But existing only with our clients but with
business models across many all who want to understand
sectors will be seriously the implications of one of
disrupted. And tens of millions the most profound shifts
of existing jobs will be lost, the world economy will
with human workers displaced experience this century.
by robots at an increasing rate
as robots become steadily
more sophisticated.
For both people and

businesses, the effects of these
job losses will vary greatly
across countries and regions,
with a disproportionate toll
on lower-skilled workers and
3
EXECUTIVE SUMMARY
20m
Over the past decade, a robotics revolution has captured the
world’s imagination. As their capabilities expand, so does the
rate at which industries purchase and install these increasingly
intelligent machines. Since 2010, the global stock of industrial
robots has more than doubled—and innovations in engineering
Number of manufacturing and machine learning portend an accelerated adoption of robots
in service sector occupations over the next five years.
jobs that could be displaced
by industrial robots by This report sheds new light on both the current impact of robots
2030—8.5% of the global on manufacturing jobs around the world and the potential
for robots to transform the much larger (but as-yet far less
manufacturing workforce. automated) global services sector. To evaluate the implications
of this ongoing robot revolution, we have brought together
the combined expertise of Oxford Economics’ economists,
econometricians, modellers, and subject-matter experts.
The rise of robots has already had a profound effect on

industrial employment around the world: today, approximately
one of every three new manufacturing robots is being installed
in China, the world’s great workshop. Our econometric
modelling finds that on average each newly installed robot
displaces 1.6 manufacturing workers. By 2030, we estimate that 1
as many as 20 million additional manufacturing jobs worldwide

could be displaced due to robotization. 2
Lower-income regions are more at risk

This great displacement will not be evenly distributed around
the world, or within countries. Our research shows that the
negative effects of robotization are disproportionately felt in
the lower-income regions of the globe’s major economies—on
average, a new robot displaces nearly twice as many jobs in
lower-income regions compared with higher-income regions
of the same country. At a time of worldwide concern about
3
growing levels of economic inequality and political polarisation,

this finding has important social and political implications.
Given the stakes, policy-makers need an early warning system

to help them mitigate the risks of automation on employment.
As part of this study, we have developed a Robot Vulnerability
Index that ranks every region of seven developed economies in
terms of how susceptible their respective workforces are to the
installation of industrial robots (see page 25).
4 1
This finding is based on an analysis of a large, regional panel-dataset of robot stock, and other labour market indicators, over a 11 year timeframe, for 24
EU countries (minus Croatia, Cyprus, Luxembourg and Malta), along with Norway, the United States, Japan, and South Korea.
2
Countries included in this estimate account for more than 90% of industrial robot installations: EU 28, US, Japan, South Korea, Australia, China, Taiwan,
Thailand, Mexico, India, Canada, Singapore, Brazil, Turkey, Malaysia. We assume the rate of robot installations in manufacturing up to 2030 follows the
latest projections by the International Federation of Robotics, and we also account for long-term depreciation of existing robot stock.
3
Throughout this report, higher- and lower-income regions are defined as those with average household income levels above and below the national
average, respectively.
In many cases, our Index highlights that the most vulnerable

regions are somewhat removed from the wealthier districts of
their home countries—such as Cumbria in the UK, Franche-
Our research shows
Comté in France, and the high desert of Eastern Oregon in the the negative effects
US. These rural regions often include towns or cities with strong of robotization are
manufacturing heritages that play a surprisingly large part in the
regional economy. In contrast, regions that surround knowledge-
disproportionately felt in
intensive cities, such as Toulouse and Grenoble in France, or the lower-income regions
Munich and Stuttgart in Germany, typically show much lower of major economies.
levels of vulnerability to the rise of the robots. This is also true of
capital cities such as London, Paris, Seoul, and Tokyo.
Fig.1: Job losses from robots hit lower-income regions harder 4
Change in number of jobs due to one additional robot
-1.6 Average
effect
Lower-
-2.2 income
regions
Higher-
-1.3 income
regions
-2.5 -2.0 -1.5 -1.0 -0.5 0.0

Long-term impact Short-term impact
Source: Oxford Economics
4
Our modelling differentiates between a “short-term” effect, within the year of a robot installation, and a longer-term effect that builds over 10 to 15 years. 5
The $5 trillion robotics dividend

While regional impacts vary, fears about permanent global job
As the pace of robotics destruction generated by robots appear somewhat exaggerated.
adoption quickens, Our study shows that the current wave of robotization tends
policy-makers will be to boost productivity and economic growth, generating new
employment opportunities at a rate comparable to the pace
faced with a dilemma: of job destruction. We estimate that a 1% increase in the stock
while robots enable of robots per worker in the manufacturing sector leads to 0.1%
growth, they exacerbate boost to output per worker across the wider workforce.
income inequality. These increases are large enough to drive meaningful growth.
Using Oxford Economics’ Global Economic Model (GEM), we
calculated how changes in the rate of installation of industrial
robots could affect the global economy. Overall, we found
that a faster adoption of robots has a positive impact on both
short- and medium-term growth. For example, boosting robot
installations to 30% above the baseline forecast by 2030 would
lead to an estimated 5.3% boost in global GDP that year. This
equates to adding an extra $4.9 trillion per year to the global
economy by 2030 (in today’s prices)—equivalent to an economy
greater than the projected size of Germany’s.
The future of service robots

Robots are steadily gaining traction in specific segments of
the service economy, from baggage handling in airports to
loading inventory in warehouses. In this report, we assess the
likely impact (and timeframe) of service robot roll-outs in
five key sectors: healthcare, retail, hospitality, transport, and
construction and farming. For the purposes of this study we are
considering robots only as physical machines, and not including
the already-popular service-industry software like robotic
process automation (RPA) that can speak, hear, read, conduct
transactions, automate processes, and so on.
One key consideration for anticipating the pace of robot

deployment in service industries is the environment in which
these robots may be asked to operate—in particular, the extent
to which service jobs include repetitive functions. Jobs like
warehouse work are in imminent danger, while other jobs in less
structured environments will likely be carried out by humans for
decades to come.
6
It will be difficult for machines to replace humans in service

sector occupations that demand compassion, creativity, and
social intelligence. Physical therapists, dog trainers, and social It will be difficult for
workers are likely to remain secure in their jobs, for instance, machines to replace
even if truckers and warehouse workers see the future of their humans in service sector
jobs jeopardised.
occupations that demand
Policy implications compassion, creativity,
As the pace of robotics adoption quickens, policy-makers will and social intelligence.
be faced with a dilemma: while robots enable growth, they
exacerbate income inequality. Automation will continue to
drive regional polarisation in many of the world’s advanced
economies, unevenly distributing the benefits and costs
across the population. This trend will intensify as the impact of
automation on jobs spreads from manufacturing to the services
sector, making questions about how to deal with displaced
workers increasingly critical.
The challenges will be daunting. Our analysis of the job

moves of more than 35,000 US individuals over the course of
their careers shows that more than half the workers who left
production jobs in the past two decades were absorbed into
just three occupational categories: transport, construction and
maintenance, and office and administration work. Ominously,
our analysis found that these three occupational areas are
among the most vulnerable to automation over the next decade.
These findings, however, should not lead policy-makers and
other stakeholders to seek to frustrate the adoption of robot
technology. Instead the challenge should be to distribute the
robotics dividend more evenly by helping vulnerable workers
prepare for and adapt to the upheaval it will bring. Policy-
makers, business leaders, technology companies, educators, and
workers all have a role to play. We conclude the report with a
framework for action for each of these groups to navigate the
challenges and opportunities that robotization will bring.
Robots are on the rise as never before. Preparing for and

responding to the social impacts of automation will be a
defining challenge of the next decade.
7
A vision of human-free
production in Italy.
INTRODUCTION
Over the past decade, the warned the disruption
global stock of industrial caused by the automation of
robots has risen dramatically, cognitive skills could have “as This era of automation
and is projected to grow even wrenching and lengthy [an] presents significant
faster in the next 10 years, impact on the jobs market” as opportunities for
led by China’s record pace Britain’s industrial revolution. 5
of installation. The robotics He urged policy-makers to businesses to boost

industry has experienced learn the “lessons of history,” productivity. But there
exponential investment growth, with governments stepping will be winners and losers
upending decades-long up to train workers for the
trends of gradual and steady new world of work while in the labour market.
expansion. A convergence providing a welfare state
of innovations in digital to cushion the blow from
technologies (e.g., artificial technological change.
intelligence and machine
learning) along with advances To shed new light on the future This multi-disciplinary
in robotics engineering and impacts of automation, Oxford approach enables us to
energy storage, is dramatically Economics combined the construct a set of questions
transforming the capabilities expertise of its economists, for policy-makers about
of robots. New breeds of econometricians, modellers, the impact of increased
“cobots”—small, highly mobile, and other subject-matter robotization—as well as other
and dextrous machines that experts around the world. Our processes of automation—
can readily collaborate with analysis begins by modelling on economies and societies
humans—are entering the the latest and best data for around the world. Greater
manufacturing and logistics industrial robot installations understanding of these issues
arenas, and can be easily in all manufacturing sectors will be key to making the most
“trained” to work with humans around the world. These are of robot-driven gains in the
to optimise productivity. credible, longitudinal datasets future while supporting and
from which we draw fresh protecting those who stand
This era of automation presents insights regarding the impact to lose out from this era of
significant opportunities of robots on employment dramatic technological change.
for businesses to boost and productivity in different
productivity. But there will countries, and in the higher- and
be winners and losers in lower-income regions within
the labour market as these those countries.
opportunities are seized.
Millions of workers around the Building on these insights, we
world, across all sectors of the then assess the future impact
economy, will see many of the of increased robotization on
functions they were once paid global service sectors—an area
to perform handled instead where rates of robot adoption
by new technology. Millions have been much lower than
more will see the nature of in manufacturing to date,
their jobs altered significantly but which employs a much
as they are required to master greater proportion of the global
new skills to collaborate with workforce. Around three-
intelligent machines. In autumn quarters of workers across
2018, Andy Haldane, the Bank advanced economies earn their
of England’s chief economist, wages from service labour.
5
Haldane warns AI threatens lengthy widespread unemployment’ (Financial Times, 20/08/2018). 11
WHAT IS OUR DEFINITION OF

A ROBOT?
The quantitative modelling industrial robots at the end the impact of robots on
aspects of this study are of 2016, according to the employment and productivity
focused on industrial International Federation of levels. But the story will
robots used in all types of Robotics. Automation has long
6
continue to unfold as
manufacturing around the been a critical component of manufacturing itself undergoes
world. These automatically manufacturing, particularly rapid technological change. In
controlled, reprogrammable in the automotive industry, recent years, new, collaborative
machines are typically which in 2016 accounted for categories of AI- and cloud-
used for a host of physical more than 43% of the total enabled robots have emerged
activities in production, operational stock of industrial that seamlessly bridge the
such as processing materials robots in global manufacturing. gap between skilled manual
(laser cutting, mechanical The industry is at the leading assembly and automated
grinding), assembling and edge of robotic applications. production. These “cobots”
disassembling, precision create new opportunities for
welding, painting, and The quantitative analysis in this automation—even on short,
handling a wide range of report is focused on physical mixed production runs that
operations for measurement, machines for which rich, require both high levels of
inspection, packaging, longitudinal data exists. We do precision (at which robots
bending, and casting. not incorporate into this aspect excel), and sophisticated
These robots can be fixed of the analysis the growing vision, handling, and creativity
installations or mobile, role of disembodied software (where human workers
and the latest versions are applications sometimes continue to add great value).
increasingly powered by referred to as robots or bots,
artificial intelligence, so they including programmes used in
are “smart” and responsive to call centres and in RPA.
their surroundings.
Based on robust data, our
Manufacturing accounted analysis of the manufacturing
for more than 86% of the sector offers the best
world’s operational stock of perspective to date on
12 6
International Federation of Robotics (2017) “World Robotics: Industrial robots”
WHAT DRIVES THE

ROBOT RISE?
20%
Since 2010, the global stock could have as many as 14
of robots in industry has more million industrial robots in
than doubled: as many robots use, dwarfing the rest of the
were installed in the past world’s stock of industrial
four years as over the eight robots as it reinforces its Proportion of the world’s
previous. During this period, position as the world’s primary
the centre of gravity in the manufacturing hub. robot stock located in
world’s robot stock has shifted China. Approximately
towards new manufacturers, In contrast, though it has every third robot is now
mainly in China, Korea, and grown by around 370,000
Taiwan but also India, Brazil, units since 2000, the installed there.
and Poland. combined robot inventory of
the US and Europe has fallen
Approximately every third to under 40% of the global
robot worldwide is now share from its peak of close
installed in China, which to 50% in 2009. And Japan—
accounts for around one- formerly the world leader in
fifth of the world’s total automation—has reduced
stock of robots—up from its active stock of robots by
just 0.1% in 2000 (see Fig. around 100,000 units since
2). In 2017, China expanded the start of the millennium,
its lead as the world’s largest in line with a rebalancing
market for industrial robots, of its economy away from
accounting for 36% of global manufacturing and the
sales, up from 30% in 2016. If migration of many production
this trajectory of investment facilities offshore, especially
continues, by 2030 China to China.
Fig. 2: Robot installations by country, 2000 to 2016 7
New robot installations
350,000
300,000
250,000
Rest of World
200,000
China
150,000 South Korea
US
100,000
Rest of Europe
50,000 Germany
0 Japan
00 01 02 03 04 05 06 07 08 09 10 0 11 0 12 0 13 0 14 0 15 0 16
20 20 20 20 20 20 20 20 20 20 20 2 2 2 2 2 2
Source: IFR
7
Note: US data include immaterial robot installation numbers for Mexico and Canada prior to 2010 13
The automotive sector has of new robot installations in

long been the predominant high tech manufacturing grew 8
user of robots: innovations to 31% in 2016, from 21% in

in autonomous and electric 2000, reflecting rapid growth
vehicle manufacturing requires both in the sector and in the
increasingly sophisticated integration of robots into
production chains, and this production. Robots have also
has sparked demand for new, been increasingly introduced
more powerful, and intelligent into the production of rubber
machines to build them. and plastics, and are slowly
However, other manufacturing finding their way into the food
industries are now taking a and beverage manufacturing
more prominent role in robot industry (see Fig. 3).
use. For example, the share
Fig. 3: New industrial robot installations across the world by usage, 2000 vs. 2016
Rubber & Other

plastic products
0 53,000
16,000
84,000
25,000 21,000
103,000 91,000
Automotive High tech
Inner circle—2000
Outer circle—2016
Numbers refer to global robot installations in each sector for that year. Source: Oxford Economics
8
High tech manufacturing is defined as electronic devices, semiconductors, LCDs, LEDs, computer equipment, telecommunication equipment, medical 15
equipment, and electrical appliances
THREE REASONS FOR THE

ROBOT SURGE
Our analysis of the use of of a robot fell by 11% between Innovations have made today’s
industrial robots across the 2011 and 2016. 9
robots smaller, more sensitive
manufacturing sector identifies to their environments, and more
three main drivers behind this Rising labour costs in major collaborative. Thanks to AI, they
new pace of adoption: price, manufacturing economies can learn from their experiences
innovative applications, and also contribute to increasingly and make decisions informed
consumer demand. attractive pricing dynamics. In by data from a network of other
China, for example, unit labour robots. These developments
Trend #1: Robots are becoming costs in manufacturing have have helped propel robot
cheaper than humans increased by more than 65% adoption in sectors beyond the
since 2008. Wage rates have automotive industry (see Fig. 4).
The rapid expansion in robot also been rising consistently
installations is driven in part in Korea, Japan, the US, and
by the plummeting real costs Germany, in part due to the
of the machines. As with ageing of the population in
other advanced technologies, these countries.
exponential growth in
the processing power of Trend #2: Robots are rapidly
microchips, extended battery becoming more capable
lives, and the benefits of
ever-larger, smarter networks As robot technologies improve,
have all dramatically increased they are being used in ever-
the per-unit value of many more sophisticated processes,
technological components, in more varied contexts, and
while the average unit price can be installed more rapidly.
Fig. 4: Robot adoption growing faster outside the automotive sector
Percentage change in robot densification between 2011 and 2016
United States
14%
40%
China 199%
267%
-
Japan -22%
7%
South Korea 51%

83%
- -1%
Germany
27%
-50% 0% 50% 100% 150% 200% 250% 300%
Dark bar=Automotive sector

Source: IFR, Oxford Economics Light bar=Other sectors
16 9
Figures may be subject to upward bias by a trend in robot sales toward smaller installations
Trend #3: Demand for journey. Despite its rapidly and the establishment of
manufactured goods is rising, growing inventory, China high-tech manufacturing,
and China is investing in only uses 68 robots per we expect China will likely
robots to position itself as the 10,000 workers in general continue its acceleration in
global manufacturing leader manufacturing, compared robot investments for the
with 303 per 10,000 in next decade. By 2030, if the
Much of the growth in robot Japan, and 631 per 10,000 in investment in industrial robots
stock over the past decade South Korea. The imbalance continues to grow at its current
can be attributed to rising between stock and density is trajectory, China will have close
demand for manufactured shown in Fig 5. Large sections to eight million industrial robots
goods. China is at the heart of China’s workforce are still in use, as its robot density
of this change: it has become engaged in manual processes, approaches levels comparable
the world’s largest automotive meaning vast potential remains with the average across the
manufacturing site, and a for further robotization of its European Union. 10
major producer of consumer manufacturing sector—moreso

electronic devices, batteries, than in any other country.
and semi-conductors—
all highly robot-intensive With government policies
manufacturing sectors. aimed at expanding the
This trend is set to continue, use of electric vehicles
as China is still only at the (which will require large-
beginning of its automation scale battery production),
Fig. 5: Chinese scope for catch-up in robot density (2016)
Column=Robots per 10,000 workers (LHS)

Stock of robots in manufacturing (RHS)
1,000 300,000
900
250,000
800
700
200,000
600
500 150,000
400
300 631 100,000
200
309 303 50,000
100 189
0
68 0
South Korea Germany United States Japan China
Source: IFR, Oxford Economics

Robotization Potential
10
2030 projections based on short-term International Federation of Robotics forecasts, controlling for longer-term stock depreciation. 17
THE IMPACT OF ROBOTS ON

MANUFACTURING JOBS
While China leads the way in manufacturing also create
in robot investment, many employment across the wider
other major manufacturing economy. We explore this Throughout history,
economies have also rapidly positive economic impact in the geographical
expanded their use of greater detail on page 35. imbalance between the
industrial robots in recent
years. We quantified the At a regional or local level, positive and negative
impact of this global rise in however, the impact on jobs effects of automation has
industrial robot inventory on varies greatly. Since most had significant economic,
manufacturing employment manufactured goods are highly
since 2000. We also forecast tradable (because they are social and political
the number of manufacturing cheap to transport and have a implications.
jobs that could be lost to long shelf life), the households
robotization around the world that benefit from cheaper
by 2030, and the distribution goods are widely dispersed.
of potential changes across By contrast, the communities
higher- and lower-income most reliant on manufacturing as it adopts automation:
regions within countries. jobs—and thus most affected the true productivity gains
by the introduction of new can take several years to
It’s important to note that technology—are typically materialise as workers receive
despite the rising pace of much more concentrated. appropriate training, and as
robotics investment and Throughout history, this firms understand how best to
installation, popular fears geographical imbalance reorganise their production
that robots will create huge between the positive processes and business
swathes of unemployment and negative effects of models to exploit the benefits
around the world are automation has had significant of the new technology at scale.
somewhat misplaced. economic, social, and political
This is because the value implications. We developed an
created by robots across econometric model to quantify
the economy more than the impact on manufacturing
offsets their disruptive jobs in each country’s higher-
impact on employment. and lower-income regions.
Manufacturers automate their
production processes to boost GLOBAL IMPACTS
productivity.
Since 2004, each new
This creates a “displacement industrial robot installed in the
effect” on manufacturing jobs, manufacturing sector displaced
since the new technology an average of 1.6 workers from
can perform a worker’s job their jobs. The full impact takes
more cost-effectively for a time to materialise, however.
given standard of quality. It Within the first year of a robot’s
also reduces unit-production installation, roughly 1.3 workers
costs that, in a competitive are displaced, on average, from
market, translate into lower their job; this extends to 1.6
prices and effectively raises workers over subsequent years.
the real spending power of
consumers. Therefore, the This finding is consistent with
same robots that displace jobs other evidence from industry
19
A NOTE ABOUT OUR

ECONOMETRIC MODELLING
Our study presents our labour markets—these include
econometric analysis of the changes in real wages, shifts
link between robot installations in global trade patterns, and
and manufacturing job losses other unobservable regional
at both the national level and and industry-related factors.
for regions within specific
countries. Our model focuses See Appendix for a
on 29 manufacturing-intensive full explanation of this
countries using 11 years of methodology.
data, offering unprecedented
levels of detail about the
past and future impacts of
robotization on manufacturing Fig. 6: Our econometric modelling framework
jobs around the world.
Source Variable
In addition to providing
absolute figures, we have
calculated the marginal impact Momentum
of each additional robot Oxford
Manufacturing jobs per
Economic’s
installation on manufacturing Global capita in previous year
jobs across the countries Economics Data used for 29 countries
over 11 years from 2004 to
and
studied. Our modelling Global Economic 2016, disaggregated by
region and sector.
performance
establishes how this impact Cities
databanks
compares between lower- and GDP per capita
higher-income regions within Employment impact
a country (defined as regions
with average household Outsourcing
income levels above and below Share of
Comtrade manufacturing jobs
the national average). database;
Trade with China
in local economy
Oxford
Economics
Drawing on data from the calculations
Export Manufacturing jobs per
1,000 workers
International Federation of aptitude
Trade with the rest
Robotics (IFR), an industry of world
trade group, we investigated
the ways in which the
installation of additional Region-specific
industrial robots affected local factors
We isolate the average marginal
manufacturing employment Panel data techniques
impact on manufacturing jobs
at local level from each
in Japan, the European Union, additional robot intsalled.
the United States, South Korea, IFR; Robot
and Australia. By constructing
11
Oxford densification
Economics
a large, regional panel dataset calculations Robots per 1,000 workers
of robot stock alongside other
labour market indicators over
an 11-year timeframe, we were
able to isolate the impact
of robotization versus other
strong influences on local
20 11
Despite its prominence in global manufacturing, China was omitted from our econometric modelling exercise due to a lack of data in other important
modelling variables.
We also calculated the total workforce (some 400,000 We have projected the growth
amount of manufacturing jobs jobs). In China, as many as in the active robot stock
lost to robotization throughout 550,000 manufacturing across major manufacturing
the world since the turn of jobs have been displaced economies to 2030, based on
the century, considering
12
by robotization since 2000, the IFR’s three-year growth
factors such as redundancies equivalent to around 1% of projections for new robot
caused by off-shoring and the its current manufacturing installations and including
globalisation of supply chains. workforce. the need to replace some
In all, we estimate that around robots over time as they
1.7 million manufacturing jobs Assuming robot investments deteriorate. On this basis,
have been wiped out since continue at their current we expect almost 20 million
2000 due to the global rise of pace, many millions of manufacturing jobs to
industrial robots. Fig. 7 illustrates additional manufacturing disappear around the world
the impact by country: in the jobs are likely to be because of robotic automation
US, we estimate that more displaced by robots by (see Fig. 8). Put differently, if
than 260,000 jobs have been 2030. While considerable current trends hold, the global
lost to robots (around 2% uncertainties exist around manufacturing workforce
of today’s manufacturing the rate of adoption of new would be 8.5% larger by 2030
workforce), while in the technologies, it is possible if robots were not remaking
European Union, robots have to estimate the likely impact the market. 13
taken the place of 1.5% of of robotization in the

the current manufacturing coming years.
Fig. 7: Cumulative jobs losses implied by automation since 2000
Cumulative jobs lost since 2000

0
-200,000 400,000
-400,000 550,000 EU28
-600,000 China
260,000
-800,000 US
-1,000,000 340,000
South Korea
-1,200,000
Rest of world
-1,400,000 100,000
-1,600,000
-1,800,000
01 1
02 0 03 04 0 05 06 0 07 08 09 0 10 20 1 0 12 0 13 0 14 0 15 0 16
20 20 2 20 2 20 2 20 20 2 2 2 2 2 2
12
Global estimate based on more than 90% of known global industrial robot installations, according to the International Federation of Robotics. 21
13
Manufacturing employment projections from Oxford Economics’ Global Industry model.
REGIONAL IMPACTS HIT levels are either above or

HARDER IN LOWER-INCOME below the national average. It
Installing one extra AREAS also controls for regionally-
industrial robot in a specific labour market
lower-income region Our modelling also allows shocks and underlying
us to look at the impact employment trends.
leads to almost twice of automation on different
as many manufacturing regions within each country. Why do these regional
job losses as in higher- These regional differences differences occur? They are
offer important social and not driven by the relative size
income regions. political implications for of the manufacturing sector—
policy-makers. manufacturing accounts
for roughly the same share
Our analysis shows that of economic activity and
installing one extra industrial employment in both lower-
robot in a lower-income and higher-income regions
region leads to almost twice in our sample, and our model
as many manufacturing job controls for sector size. But
losses as in higher-income there are structural differences
regions (see Fig. 9). This in the composition of
finding is based on an employment in manufacturing
analysis of our 29 sample that influence the impact
countries, distinguishing robots have.
between regions whose
average household income
Fig. 8: Projected cumulative jobs losses by automation, up to 2030

14
0
EU28
-5,000,000
China
-10,000,000 US
-15,000,000 South Korea

Rest of world
-20,000,000
-25,000,000
18 0 19 20 0 21 22 23 24 25 26 27 28 29 30
20 2 20 2 20 20 20 20 20 20 20 20 20
22 14
Projections for ‘Rest of World’ include countries covering more than 99% of the estimated global total.
Fig. 9: Manufacturing job losses skew towards lower-income regions
Change in number of jobs due to one additional robot
-1.6 Average
effect
Lower-
-2.2 income
regions
Higher-
-1.3 income
regions
-2.5 -2.0 -1.5 -1.0 -0.5 0.0

Long-term impact Short-term impact
Manufacturing workers in contrast, a significantly higher carry out are—on balance—

lower-income areas tend to proportion of managers easier to automate. These
have lower skill levels and are and professionals in the efficiency gains can be
therefore more vulnerable to manufacturing industry are in realised by laying off staff,
automation. There is typically higher-income regions. or by moving the firm to a
a difference in the number This vulnerability has evolved new, more productive (and
of robots per manufacturing over time. In the past, lower- likely more automated) site.
worker between higher- income areas competed with Either way, the manufacturing
and lower-income regions, more expensive cities and workers in those regions are
indicating that those in regions for manufacturing at risk.
lower-income regions are, on investment, with the lure
average, less productive. Data of lower unit costs of Moreover, the regions of a
from the UK Labour Force production. This competitive country most likely to shed
Survey, for example, shows edge was a consequence manufacturing workers will
that manufacturing workers in of relying on a lower-paid, not benefit equally from the
lower-income regions of the less-productive workforce to “robotics dividend” —the
UK are more likely to work in carry out lower-skilled jobs. new jobs created from the
lower-skilled occupations— In the new era of automation, productivity boost that feed
elementary workers and the occupational mix in lower into the wider economy.
machine operatives account income areas means those Instead, increased industrial
for around one-third of the same manufacturers face automation will tend to
workforce in lower income the biggest opportunities exacerbate the regional
regions, compared with 22% for efficiency savings. The inequalities that already exist
in higher-income regions. In functions their employees within advanced economies.
23
THE ROBOT VULNERABILITY

INDEX
It is vital for policy-makers to economic upheaval in the
understand how an uneven years ahead. Mapping the
distribution of robotics will vulnerability to robot adoption Our Robot Vulnerability
affect different parts of their across all regions of these five Index shows that specific
country. We have developed advanced economies revealed regions that are at
a Robot Vulnerability Index some common patterns,
to help identify which regions which can be summarised in highest risk of labour
within our chosen economies three key trends. disruption—but also
(the US, Germany, UK, France, reveals some common
Japan, South Korea, and Trend #1: Existing
Australia) will be hardest hit by inequalities will patterns across regions.
the ongoing automation of the intensify
manufacturing sector.
Successful economic
Our index produces a performance at the regional
vulnerability score for level in advanced economies
each sub-national region , 15
is usually inversely correlated
comprised of three equally with robot vulnerability. In
weighted indicators: the UK, France, and Germany,
those regions that have
• Local dependence performed best in recent
on manufacturing years (in terms of overall GDP
employment—defined as growth) are the least exposed
the manufacturing share to future robot automation,
of total employment in the and vice versa.
region.
• Future readiness of local This means the regional
industry—characterised inequalities that exist within
by a region’s current countries, such as England’s
intensity of robot use north-south divide, could be
in manufacturing, exacerbated by the rise of
controlling for the type the robots. This trend has
of manufacturing activity important implications for
undertaken, and measured policy design in advanced
relative to international economies pursuing
competitors. international competitiveness
• Productivity of the local through automation.
manufacturing workforce—
measured relative to the
national average.
The index is thus designed

to highlight regions that are
economically dependent
on a less productive (or
lower-skilled) manufacturing
industry and do not currently
use many robots, since these
areas are at highest risk of
15
Sub-national regions correspond to European NUTS 2, US States, Japanese prefectures, Australian states, and South Korean districts. 25
Trend #3: Rural Country-by-country analysis

regions mask hidden
The pockets of workers vulnerabilities Over the next seven pages, we
most vulnerable to illustrate each local region’s
automation can often be The pockets of workers most relative vulnerability to future
vulnerable to automation can manufacturing automation,
found in rural areas. often be found in rural areas. according to our Robot
Despite relatively sparse Vulnerability Index. Each
populations, these regional map is colour-coded from
economies are frequently “high vulnerability” to “low
grounded to isolated towns vulnerability” regions (relative
Trend #2: Many major with more manufacturing- to the rest of that country)
cities are safe (for now) intensive industrial and includes commentary
structures on which the on some of the most striking
Our analysis shows that major wider region depends. This geographical results.
cities are often safe havens is especially problematic
for workers in the face of when manufacturing in these
robot led job displacement. towns is characterised by
Diversified economies depend traditional, labour-intensive
less on manufacturing jobs, techniques, low levels of
and higher labour costs mean productivity, and dated
manufacturers located there manufacturing processes.
are already highly productive
and tend to employ more In many countries, such
highly skilled workers. London, regions have often been left
Paris, Seoul, Sydney and Tokyo behind as metropolitan centres
are all examples. prospered, and those dynamics
have generated political
But manufacturing-intensive polarisation. This highlights the
cities (including many in importance of taking policy
South Korea) face a more action to cushion the likely
uncertain future. Cities impact of robotization in these
with large populations that vulnerable areas.
are more dependent on
the manufacturing sector
for employment but lag
their industry peers in
robot intensity and labour
productivity are vulnerable to
disruption. Fierce competition
will ultimately lead these city-
based industries to pursue
further automation or risk
losing out to more productive
competition elsewhere.
Either way, additional job
displacement of current
manufacturing workers is likely.
26 1
Xxxxxx
UNITED STATES
Oregon is the
most vulnerable state in
the US to a future acceleration
in robot installations. The state
has had success in transitioning out
of traditional sectors into the
production of high-tech components.
But high dependence on
Low vulnerability manufacturing, particularly in and
Lower-medium vulnerability around Portland, and the state’s
exposure to globally competitive
Upper-medium vulnerability sectors, mean its workers are
High vulnerability vulnerable to rapid
technological
progress.
New England
states tend to have low
While Texas and its vulnerability to the future
neighbour Louisiana are two spread of manufacturing
particularly vulnerable states in robots, as do those with a higher
the south, Indiana is equally reliance on tourism (Florida,
vulnerable in the mid-west. It is Nevada, Hawaii). The same is true
associated with steel-making (and for New York state, which,
with heavy industry more Alaska alongside a significant
generally), albeit with an Hawaii manufacturing base has a high
increasing focus on developing concentration of financial
the growth of its higher-value, and business services.
knowledge-based
industries.
Most State: Index Score: Least State: Index Score:

vulnerable Oregon 0.58 vulnerable Hawaii 0.17
states Louisiana 0.58 states District of Columbia (DC) 0.18
Texas 0.50 Nevada 0.25
Indiana 0.46 Florida 0.25
North Carolina 0.46 Vermont 0.26
27
GERMANY
Germany’s least-vulnerable
region is Hamburg. It has a low
level of dependence on
manufacturing jobs, and what
manufacturing it does have is
typically advanced and highly
productive, with cutting-edge
levels of automation.
Low vulnerability
Lower-medium vulnerability
Upper-medium vulnerability
High vulnerability
A cluster of four
eastern regions close to
the Czech
border—Chemnitz, Thüringen,
The home regions of BMW Oberfranken, and
and Mercedes—Bavaria and Oberpfalz—look to be the most
Stuttgart, respectively—are vulnerable to robotization. All have
examples of future-ready high concentrations of
production ecosystems, featuring a manufacturing employment, and
highly skilled, highly productive (typically) low levels of
workforce. productivity—particularly
Chemnitz and
Thüringen.
Most Region: Index Score: Least Region: Index Score:

vulnerable Chemnitz 0.56 vulnerable Hamburg 0.06
regions Thüringen 0.49 regions Darmstadt 0.13
Oberfranken 0.49 Oberbayern 0.17
Oberpfalz 0.47 Köln 0.19
Freiburg 0.46 Berlin 0.20
28
UNITED KINGDOM
East Yorkshire and Northern
Lincolnshire, Shropshire and
Staffordshire, Cumbria, and West
Wales and the Valleys exhibit the
highest vulnerability scores in the UK.
Low vulnerability These regions are relatively
Lower-medium vulnerability dependent on manufacturing for
employment, and have a relatively
Upper-medium vulnerability high incidence of low-skilled
High vulnerability workers.
The UK’s most Robotization will

vulnerable regions to exacerbate the
robotization can be found in north-south divide. Inner
its more rural areas. These London is perhaps the least
sparsely populated regions vulnerable part of the country to
may contain towns with the rise of robots, and the South East
concentrated manufacturing region is similarly well-placed for the
industries. Cumbria tops next phase of industrial automation.
our UK Index. Manufacturing operations in these
regions tend to be more advanced
and more automated than in
other parts of the country,
reflecting the higher cost
of labour here.
The West Midlands’

manufacturing processes are
already among the most
automated in the UK, and the
region is nearly as robot-dense as
international market leaders. However,
it is also characterised by low levels
of productivity, and with a high
dependence on manufacturing
employment, which could still
imply a challenging future.
Region: Index Score: Region: Index Score:

Cumbria 0.59 Inner London (East) 0.15
Most East Yorkshire & Least Inner London (West) 0.17
vulnerable North Lincolnshire 0.59 vulnerable Outer London
regions Shropshire & regions (West & NW) 0.20
Staffordshire 0.58 Berks, Bucks &
West Wales & Oxfordshire 0.25
the Valleys 0.56 Surrey, East &
Lincolnshire 0.54 West Sussex 0.28
29
FRANCE
The Île-de-France,
centred on Paris, is France’s
least-vulnerable region. It is least
dependent on manufacturing jobs,
and what manufacturing activity it
does have is (a) highly productive
and (b) the most robot-intensive in
Low vulnerability
the country, alongside the
Lower-medium vulnerability Midi-Pyrenees. This means it
has already undertaken
Upper-medium vulnerability significant levels of
High vulnerability automation
We find that the most

France’s most southerly vulnerable region to
regions, plus Rhône-Alpes, robotization is Franche-Comté.
are collectively the ‘runners-up’ France’s most
behind Paris in terms of their low manufacturing-intensive region
vulnerability on our Index. These is nevertheless relatively rural and
regions are home to advanced sparsely populated. Its relatively
high-tech manufacturing companies, low rate of robotization means
notably in leading cities such as there could be high levels of
Toulouse (home to Airbus, among automation coming.
others) and Grenoble, and thus
benefit from a future-ready,
highly skilled workforce.

Most Franche-Comté 0.61
Least Île de France 0.03
vulnerable Basse-Normandie 0.51 vulnerable Provence-Alpes-Côte
regions Picardie 0.51 regions d'Azur 0.26
Limousin 0.51 Languedoc-Roussillon 0.30
Auvergne 0.49 Aquitaine 0.35
Midi-Pyrénées 0.36
30
JAPAN
Hokkaido,
Japan’s northernmost
island—famous for brewing beer
Low vulnerability and as a skiing destination and
gateway to the Hokkaido
Lower-medium vulnerability mountains—is one of the least
Upper-medium vulnerability manufacturing-intensive parts of the
country. After Tokyo, it is the
High vulnerability second-least vulnerable region on
our Index.
Some of Japan’s most

mountainous prefectures
feature among the most
vulnerable to job losses. Although
sparsely populated, these large
regions are punctuated with
traditional manufacturing enclaves,
which may prove highly vulnerable to
change. The regions of Kochi, Nara,
and rural Tottori are, in this sense,
among the most vulnerable to
the trends of automation.
Japan’s
largest and most economically
important prefecture, Tokyo, is the
country’s least-exposed region to robots
displacing manufacturing jobs, according to
our Index. Companies here have already
established advanced levels of robot
intensity, and the region’s diverse economy
means workers are less dependent on the
manufacturing sector for employment. A
similar pattern is true of the regions
surrounding other important cities
such as Osaka, Yokohama, and
Kawasaki.

Most Tottori 0.54
Least Tokyo 0.09
vulnerable Kochi 0.51 vulnerable Hokkaido 0.20
regions Nara 0.49 regions Osaka 0.25
Shiga 0.49 Fukuoka 0.28
Saga 0.48 Miyagi 0.28
31
SOUTH KOREA
Workers in South Korea’s largest

city, Seoul, are the country’s least
vulnerable to the growth of
manufacturing robots. The regional
economy is diverse, meaning it has a
Low vulnerability low dependence on the
Lower-medium vulnerability manufacturing sector for work, and
the labour force is highly
Upper-medium vulnerability productive.
High vulnerability
Incheon and Daegu

are the most exposed
regions. These major
manufacturing hubs have
relatively low levels of
manufacturing
productivity, so are
ripe for change.
Korea’s second city, Busan,

and its neighbour, Ulsan, appear
vulnerable to robots on our Index.
Ulsan is home to major car plants,
shipbuilding facilities, and oil
refineries. It has very high levels
of manufacturing productivity, but
its relatively high robot vulnerability
score is driven by a remarkable
dependence on manufacturing
employment.

Most Daegu 0.38
Least Seoul 0.11
vulnerable Incheon 0.35 vulnerable Jeollanam-do 0.13
regions Ulsan 0.33 regions Gangwon 0.19
Gyeongnam 0.32 Chungcheongnam-do 0.21
Busan 0.29 Gyeongbuk 0.23
32
AUSTRALIA
Australia’s most populous

state, New South Wales, looks
rather less vulnerable than either
Victoria or South Australia. In this
state, the labour market has become
less dependent on manufacturing
Low vulnerability jobs in recent years, while
Lower-medium vulnerability manufacturing productivity has
improved. So the impact of
Upper-medium vulnerability further robot densification will
High vulnerability likely be muted.
South Australia
is the most vulnerable Victoria is less vulnerable
part of the country to future to robots than South
robot rollout, according to our Australia, and also faster
Index. The state is Australia’s growing. Melbourne and its
most manufacturing intensive surrounding area have a diversified
but has the slowest-growing manufacturing base, although one
economy and low levels of that is declining in relative
manufacturing importance as Melbourne’s service
productivity. economy strengthens. Victoria’s
manufacturing productivity is
also higher than that of
South Australia.
Regions and Region: Index Score:
territories South Australia 0.42
ranked from Victoria 0.39
most-to-least Tasmania 0.37
vulnerable Queensland 0.32
New South Wales 0.28
Western Australia 0.14
Northern Territory 0.06
Australian Capital Territory 0.06
33
A delivery robot being trialled
in London, 2017.
THE ROBOTICS DIVIDEND
1%
Despite the decline of To capture the potential
manufacturing jobs over implications of the new era
the past decade, it would of robotics on the global
be simplistic to characterise economy, we used Oxford
robotization as only a Economics’ Global Economic increase in the stock of
destroyer of jobs. While Model (GEM). The GEM
certain sets of workers lose covers 80 countries and is robots per worker in the
their jobs to robots, many the foundation of all Oxford manufacturing sector
in the wider population Economics’ country, industry, leads to a 0.1% boost to
benefit from a “robotics and city forecasts. It enables
dividend”—lower prices for us to test the sensitivity of output per worker across
manufactured goods, higher macroeconomic outcomes to the wider workforce.
real incomes, and stronger different rates of investment
tax revenues. This will be across many advanced
particularly important to economies around the world.
the lower-income regions This modelling suggests that
we have identified as being the rate of industrial robot 30% above baseline projections
most vulnerable to the robot adoption over the coming for 2030. For China’s
revolution. years will have a significant manufacturing sector, this
impact on global GDP growth. would put its robot density on
Our modelling shows that a par with the levels of robot
robots have delivered The first step in our GEM density that currently exist in
considerable productivity analysis was to establish a Japan and Germany.
gains in recent years. We baseline projection for GDP
analysed the impact of robot growth consistent with the By contrast, the low scenario
densification on productivity short-term robot investment assumes the pace of robot
growth in an international trajectories forecast by the adoption slows, leaving the
sample of countries over 11 International Federation of stock of industrial robots some
years, controlling for factors Robots (IFR) trade group. 17
30% lower than the baseline
such as skill levels and other These trajectories for the by 2030. This would put
capital investment, across US, Europe, and large Asian the robot density of China’s
29 of the world’s most economies were calibrated manufacturing sector at a
advanced economies. We 16
against historical growth levels level comparable with the
found that a 1% increase for both robot stock and current robot density of the
in the stock of robots per robot density. Our baseline US manufacturing sector—a
worker in the manufacturing projections for the growth in level significantly lower than
sector alone leads to a robot stock amounted to an Japan and Germany. (For more
0.1% boost to output per annual increase of roughly 5% information on how we used the
worker across the wider for China, 3% for the US, 2% GEM to simulate the impact of
workforce. This confirms for both South Korea and the different robot adoption rates
our hypothesis: that by Eurozone, and 0.7% for Japan. on the annual GDP performance
displacing automatable jobs of key economies around the
in manufacturing, robots Next, we explored “high” world, see box on page 37).
free up many workers to and “low” scenarios for
contribute productively robotization, relative to the
elsewhere in the economy, IFR’s short-term benchmark.
as they meet the demands The high scenario assumes
generated by lower prices that the global stock of
for manufactured goods. industrial robots will accelerate
16
The sample size for this model differs to our employment model due to data availability. 35
17
The IFR’s latest three-year growth projections for new robot installations appear in its publication World Robotics 2017: Industrial Robots.
potential of the economy. We

Results of our high and low used the GEM to simulate a
The jobs displaced by robotization scenarios corresponding increase in the
industrial robots will level of business investment
be concentrated in the Overall, we find that faster and a boost to productivity,
adoption of robots has a positive both in the manufacturing
manufacturing sector— impact on both short- and sector and the wider economy.
where their uses are most medium-term economic growth. As Fig. 10 illustrates, a speed-
well established. up of robot investment results
Specifically, the GEM suggests in significant gains in GDP
the high adoption scenario growth for the world’s largest
would boost global GDP economies. US GDP rises
by 5.3% above our baseline by 13.1% above the baseline
GDP growth forecast in projection by 2030. South
2030. This equates to adding Korea GDP rises 11.7%, and
an extra $4.9 trillion to the China experiences a 9%
global economy that year— increase. The relative gains
equivalent to an economy and losses for Japan in either
greater than the projected scenario are significantly
size of Germany’s. Under18
lower than for all other
the low adoption scenario, major economies, due in
we predict a similarly sized part to a slower rate of robot
negative impact on the global investment in the baseline
economy’s growth trajectory. forecast, since Japan’s
manufacturing sector is
The high-adoption scenario already heavily robotized.
bolsters the productive
Fig. 10: Projected impact of different scenarios on annual GDP in 2030
Percentage difference from baseline

15
13.1
10 11.7
9.0
5 7.5 Dark bar=High scenario
2.2
0
-4.3
-1.9
-5 -7.7
Light bar=Low scenario
-8.5
-9.4
-10
-15
China US EU Japan South Korea
36 18
Value expressed in 2018 prices and compared with estimated German GDP in 2030.
MODELLING THE ECONOMIC

IMPACTS OF ROBOTIZATION
To investigate the implications In practice, this meant using scenarios. This adjustment
of different rates of investment the GEM to apply three key, was calibrated with our
in and adoption of industrial robot-related economic headline econometric
robots, we simulated some “shocks” to our baseline model: result detailed on page
stylised scenarios for the 20 of this report, which
US, Europe, and large Asian • An increase/decrease in found that in our modelled
economies using the Oxford “total factor productivity” economies, each additional
Economics Global Economic (the output achieved by a industrial robot ultimately
Model (GEM). certain amount of capital displaces 1.6 manufacturing
and labour inputs) that workers, on average.
We used the GEM to establish result from adopting more/
a baseline projection for fewer robots into industrial Once these initial economic
global growth consistent practices; shocks were applied, the GEM
with the current “benchmark” used its modelled linkages
robot investment trajectories • A rise/fall in business between business, household,
established by the investment, capturing government, and international
International Federation different levels of sectors to derive the overall
of Robots (IFR). We then expenditure on industrial impacts on the different
employed the GEM’s robots; economies.
modelling capabilities to
explore two alternative • A shock to employment,
scenarios. We set these expressing the fact that,
“high” and “low” robotization to generate a given level
scenarios at 30% above of output, fewer/more
or below the IFR’s current workers may be required
benchmark rates of adoption. under the high/low
RESHAPING THE LABOUR But while this “robotics Historically, low- and medium-
MARKET dividend” will boost skilled workers displaced from
employment across many an increasingly productive
The results of our GEM sectors of the global economy, manufacturing sector have
analysis show that jobs are the jobs displaced by industrial found opportunities in the
both created and destroyed robots will be concentrated service sector. But as robotic
through the increased use in the manufacturing sector— technology converges with
of automation and industrial where their uses are most well rapid digital innovations, what
robots. Specifically, an increase established. And while some can unemployed workers do
in the rate of robot adoption new manufacturing jobs will if robots take on service jobs
would significantly affect be created by the robotics as well? Next, we explore
firms’ productivity levels, and dividend, it is unlikely they the new frontier of service
hence the size of the economy. will equal the number of jobs robotics, and how this is
This increased wealth is that could be displaced by manifesting itself across the
therefore likely to result in automation in that sector—up service sectors of the world’s
job creation that will offset to 20 million around the world largest economies.
the displacement of local by 2030.
manufacturing employment
we have identified.
37
38
ROBOTS ARE COMING TO THE

SERVICE SECTOR
While the rise of industrial robotic process automation)
robots has already reduced across service industries has
Innovations in AI, manufacturing employment been slow, for important
machine learning, and significantly in advanced reasons. The underlying return
computing power suggest economies around the world, on large capital investments
manufacturing accounts can be harder to justify
a significant acceleration for only a minor share of in environments without
in the adoption of robots total employment in these significant scale. Many aspects
across service industries. countries. Instead, the vast of service work can prove
majority of people work in difficult to automate.
the sprawling service sector—
around three-quarters of all However, innovations in
workers, according to the artificial intelligence, machine
latest Oxford Economics learning, and computing
estimates. power suggest a significant
acceleration is coming,
To date, the adoption rate particularly in the category
for physical robots (as of logistics systems—led in
opposed to software-only part by the global expansion
bots and applications which of Amazon and other
have already gained wide multinational e-commerce
acceptance in powering companies (Fig. 11).
Fig. 11: Number of professional service robots distributed, by broad category (top-five most popular)
Thousands 2015 2016 2017 2018 2019-2021 (average)
180
160
140
120
100
80
60
40
20
0
Logistics systems Public relation Defence Field robotics Powered human
robots and joy applications exoskeletons
rides
Source: IFR
40
Where will automation occur three years,” says Dr. Kai-fu

most quickly in the service Lee, an expert on artificial
sector, and which jobs will intelligence, founder of “For work that takes
remain immune to the rise Sinovation Ventures, and place in unstructured
of the robots? As industrial former senior executive at environments, I don’t
workers displaced by both Google and Microsoft
automation seek alternative “But for work that takes place anticipate robots fully
employment in the services in unstructured environments, replacing people for
sector—whether as taxi I don’t anticipate robots fully another 20 or 30 years.”
drivers, shop assistants, or replacing people. Some more
hotel porters—will robots breakthroughs will be required.”
and artificial intelligence Dr. Kai-fu-Lee, venture
start to pinch those In his current role as a venture capitalist and author
opportunities as well? capitalist and author, Dr. Lee 19
has an interest in promoting

Robots are steadily gaining AI—which makes his measured
traction in specific segments view about the subject all the
of the service economy, from more telling. He notes that
baggage handling in airports an inventory warehouse, like
to maintaining inventory a motorway, represents a
in warehouses and even relatively defined space where
bricklaying on construction programming rules and sensing
sites. Guided by sensors, technology can help delineate
cameras, and machine the precise work landscape.
intelligence, which grow “But in a flexible and highly
cheaper and more powerful changeable environment, like a
by the year, robots are room full of people at a cocktail
beginning to stake out their party, the tasks are far more
presence in the hospital ward, difficult,” he says.
on the retail sales floor, and in
hotels and restaurants. AI-powered systems will only
prove cost-effective where
But how quickly a widespread large economies of scale exist,
shift to service robots occurs Dr. Lee says. For example, a
depends on several factors. robot could likely be employed
While some service jobs may to vacuum rooms across a
be considered standardised hotel chain, because each floor
and relatively easy to contains rooms that follow
automate, others demand only two or three standardised
uniquely human qualities floorplans the robot can
such as social intelligence, recognise. Likewise, hospitals
imagination, empathy, have a series of extensive,
and other cognitive skills standardised procedures, many
not readily translated into of which could be passed off
algorithms. to a robot system. To better
calculate the potential impact
“Yes, Amazon warehouse of robotics on service sector
workers are likely to be employment, it is therefore
replaced by robots in two or imperative to differentiate
19
Kai-Fu Lee, “AI Super-Powers: China, Silicon Valley, and the New World Order” (Houghton Mifflin Harcourt, 2018). 41
Robot receptionists in Tokyo’s
Henn Na Hotel in Ginza.
between occupations and in assisted living homes, hospitals,

tasks to complete certain and clinics.
occupations even in the same According to IFR data,
sector—and to accurately Over time, robots will sales of medical robots
assess the mix of routine displace personnel in the increased by 49% in 2018
and highly cognitive skills performance of specific
embedded within a specific tasks, leaving humans with compared with 2017, to
occupation or job function. more time to focus on the total over 4,400 units—
compassion, empathy, and totalling an estimated
Another key factor is emotional intelligence that
whether scale can be used remains a major component of $1.9 billion.
to offset the added costs healthcare. Robots can readily
associated with deploying transport blood samples to a
robot systems. For example, lab, or obtain medicines from
an autonomous vehicle is far a dispensary. The potential for
more expensive to build than robots to be used in controlled
a conventional car, because of hospital environments may
all the cameras, sensors, and be especially compelling,
software needed. “That car since healthcare in many
only pays for itself when you developed markets faces an
can exclude the pay for the acute labour shortage, and
driver, and run that vehicle all few nurses can expect to
day long,” says Dr. Lee. confront unemployment if
robots take over some of their
To better unravel the impact of menial tasks. Because the cost
robots on a variety of service of developing and deploying
jobs, we have examined several robots is significant, only the
key sectors. largest institutions are likely to
introduce them rapidly into a
clinical setting, but they may
Healthcare become more commonplace
over the next decade.
The healthcare industry in the Robot-assisted surgery

developed world faces real (typically in procedures
challenges. Populations are that are minimally invasive)
ageing rapidly: the World Health already lets doctors perform
Organization has estimated that complex procedures with more
by 2020 the number of people precision and flexibility. The
aged 60 years and older will most common robotic system
outnumber children younger than includes a camera arm and
five—and by 2050 the proportion mechanical arms connected
of the world’s population aged to surgical instruments, which
over 60 years will nearly double, the surgeon controls while
from 12% to 22%. Many countries seated at a computer console
face a severe shortage of care near the operating table. The
workers to assist the elderly and console gives the surgeon a
infirm. So it is realistic to expect high-definition, magnified view
robots to have a greater impact of the surgical site. Minimally
43
invasive surgeries can reduce largely conforms with Mr.

In health and care, hospital stays and shorten Cousins’ view, noting that “in
recovery times. health and care, automation
unlike other settings, will primarily complement
machines would work According to IFR data, sales of human skills and talents
alongside human medical robots increased by by reducing the burden of
49% in 2018 compared with administrative tasks. Unlike
beings, not replace 2017, to total over 4,400 units other settings, machines would
them. at an estimated $1.9 billion. IFR work alongside human beings,
expects that more than 22,000 not replace them, so patients
units will be sold between now would benefit.” 20
and 2021, a compound annual

growth rate of 27%, making
medical robots the most Retail
valuable of all service robot
sectors.
The US retail sector employs
One robot system, the Relay, some 16 million workers that
manufactured by San Jose, include cashiers, salespeople,
California-based Savioke, safely stock clerks, and customer
and reliably transports items service representatives. Robots
throughout a hospital, freeing are already displacing human
time for pharmacists, lab techs, workers in giant warehouses
and other skilled workers to and logistics centres, where
focus on more valuable work online retailers such as Amazon
and patient care. Able to are continuously seeking
operate elevators and doors productivity improvements to
and navigate in busy public reduce costs.
corridors, the Relay delivers
medicines, blood, lab specimens, At the leading edge of
snacks, and documents safely automation, Amazon keeps
and reliably 24 hours a day. finding new ways of getting
robots to do work once
“You don’t want nurses spending handled by employees. In 2014,
time doing things that aren’t the company began rolling
about nursing,” says Steve out robots to its warehouses
Cousins, CEO of Savioke, which using machines originally
builds the Relay. “Why should a developed by Kiva Systems—a
nurse walk down the hall to take company Amazon bought
a blood sample to the lab, when for $775 million seven years
a robot can do it?” Collaborating ago and renamed Amazon
with robots not only increases Robotics. Amazon now has
worker satisfaction, he says, it more than 100,000 robots
also helps alleviate a chronic in action around the world
nursing shortage. and plans to add many more.
These robots are said to make
A report prepared by the warehouse work less tedious
Institute for Public Policy and physically taxing while
Research (IPPR) in the UK also enabling the kinds of
efficiency gains that allow a
44 20
https://www.ippr.org/news-and-media/press-releases/embrace-full-automation-to-release-time-to-care-in-the-nhs-and-social-care-says-top-surgeon-
lord-darzi
customer to order toothpaste Delhaize is placing 500 robots The job of a checkout cashier is
after breakfast and receive it armed with sophisticated already endangered: Amazon
before dinner. cameras into US grocery stores, has started opening small,
to make sure store shelves are AI-powered, checkout-free
The IFR estimates that 69,000 stocked and spills are cleaned supermarkets, where a customer
logistics systems (the kind from the floors. On the sales uses her smartphone to pay for
used in retail warehouses) were floor of Saturn appliance stores the merchandise in her basket.
installed in 2017, an increase in Germany, a life-sized robot is
of 162% over 2016. Some likely to greet you heartily and Likewise, robots can do a better
90% were installed outside of direct to you the specific model job walking down an aisle and
factories. The value of logistics of TV you’re looking for. And in tracking inventory than humans
systems sales reached about a BevMo wine and liquor store can since they are less easily
$2.4 billion. in Walnut Creek, California, distracted. “You’re not going to
the inventory is likely to be see a robot stocking shelves,
Today, robots are also moving monitored and tracked by a at least in the near term,” says
steadily to the showroom two-wheeled assistant named John Wilson, head of research
floor, even if the roles they are Norma, who can also lead you at Cornerstone Capital Group,
being designed to perform to the shelf of chardonnay (see a New York-based investment
are somewhat circumscribed. case study). advisor. “But technology will
The Dutch giant retailer Ahold bring more efficiency.”
CASE STUDY: A ROBOT IN

THE WINE SHOP
Christian Bronstein, who works body, but no arms, and provides can help store management
at a BevMo liquor store in support to other workers in identify shopping patterns to
Walnut Creek, California says the store. “It provides greater get a better understanding
that robots help him do his work efficiency and offers something of which merchandise moves
better, and do not threaten his of a novelty,” Mr. Bronstein says. more quickly.
long-term prospects. “Naturally, people love it. It’s
kind of like our soft way into
For the last six months, he’s using automation.”
been working alongside a four-
foot robot named Norma who A similar Fellows robot, called
“scans shelves for out-of-stocks, the LoweBot, was installed as a
directs customers to the right test in a few big-box hardware
aisle, and tells jokes. Norma stores operated by Lowe’s in
makes our job easier,” he says. the US. Customers could ask
the LoweBot, by speaking or
The robot, manufactured by using a touch screen, where to
Fellows Robots in northern find items inside the sprawling
California, is an experiment store. The robot also carries out
launched by the beverage chain. real-time inventory tracking as
It features a large touch-screen it cruises down the aisles and
45
One of Apple’s prototype fleet
of self-driving vehicles, 2018.
1
Xxxxxx
Hospitality as information providers at patties, but it is too early to

convention centres or in hotels say whether this trend will be
is projected to grow from economically viable. Walmart
As with hospitals, a major 15,780 units sold in 2018 to has also revealed it is testing a
part of the work carried out in about 93,350 by 2021. kitchen robot assistant (named
hotels and restaurants requires “Flippy”) at its Bentonville,
social intelligence and a human Whether these robots will Arkansas. headquarters, to see
touch. But simple, routine tasks displace a company’s existing whether it could be used in its
are increasingly being passed employees is not always clear- many in-store delis.21
on to robots. cut, however. Savioke installed

the robots in the Vdara Hotel At least three US start-ups are
In the posh Vdara Hotel in Las and after a three-month trial, also working in Arizona to see
Vegas, for example, a room the human workers there whether robot grocery delivery
service call for fresh towels or a voted to keep the robots on can be made viable. But some
pot of hot coffee is likely to be staff “because the workers early customers have noted
answered by a robot trundling made more money,” says Mr. that while the robots can get
on wheels. A pair of robots Cousins—the workers knew the grocery up the driveway,
named Fetch and Jett deliver which room service requests they cannot get the bags into
service items to rooms, deliver from hotel guests would yield the kitchen.
coffee in the restaurant, and the largest tips, and kept those
attend to other chores for hotel jobs for themselves. The robots
guests. Guests can touch a were delegated the most Transport
tablet computer on the robots mundane, low-tip chores, like
to rate their satisfaction with delivering toothpaste or fresh
their service. towels. “The workers could do Those who have seen recent
triage to make sure they kept auto advertisements on
As underlying robotic the highest-valued jobs,” Mr. television can be forgiven for
capabilities such as vision, Cousins says. thinking that autonomous
speech recognition, and vehicles (AVs) are already on
machine learning continue According to Ashleen Bhim, the road in large numbers.
to improve, and costs for a manager at the Vdara, both While the massive investment
these components shrink, workers and guests love needed to deploy AI to create
the use of service robots in interacting with the robots. truly autonomous vehicles
hospitality areas is accelerating. “We’ve gotten great reviews has captured a great deal of
Executives in hotels and other from everybody,” she says. attention, real progress has
hospitality settings will closely “I really don’t know of any been slow—and few if any
analyse the work component negative feedback.” drivers have been displaced
of each occupation and to date.
determine which jobs can be Robots are also poised to play a
more efficiently handled by an greater role in the food service This could begin to change over
AI-powered robot. industry, as restaurant owners the next five years. The Center
find the rising cost of labour for Global Policy Solutions
The IFR estimates that often outpacing the ability of estimates that more than four
demand for what it calls customers to pay for sit-down million jobs will likely be lost
“public relations robots” will food service. In some “limited in a rapid transition to AVs.
grow strongly over the next menu” restaurants, robots Occupations such as delivery
three years. The number of may do the cooking: a San and heavy-truck drivers, bus
robots used in supermarkets, Francisco burger shop is using drivers, and taxi and chauffeur
as guides in museums, or a robot to prepare and grill its drivers would be most severely
21
https://finance.yahoo.com/news/exclusive-walmart-testing-robot-fry-cook-delis-154733360.html 47
affected. But the impact on develop individually owned impact on jobs likely will not
workers will largely depend on autonomous vehicles, be felt until the latter half of
the pace of the transition. others believe the entire the 2020s.
transportation infrastructure
Anecdotal evidence suggests will be disrupted by networks
that AV deployment in large of “vehicles on demand” in
numbers is not likely to be which a rider will summon a
swift. An accident last year vehicle as needed. Developing
in which an autonomous this kind of network could
vehicle struck and killed a take years.
pedestrian in Arizona served
as an alarm bell for regulators, Moreover, a shortage of
slowing investor enthusiasm. truck drivers already plagues
Just as significantly, there American roadways. A recent
are competing views on report by the American Center
what the future structure of for Mobility led by Michigan
autonomous transportation State University concludes
might look like. While that only a modest number of
some believe incumbent truck driver jobs, if any, will be
automakers will push to displaced by AVs, and that the
CASE STUDY: ROBOTS ON

THE WATER
Autonomous vehicles aren’t waterways to the deep-sea While the project for Yara
just for roads any more. ports of Larvik and Brevik—a is a singular case because
Norway’s Kongsberg Maritime, journey of 31 nautical miles. the ship will follow a regular
a nautical technology path—always within two miles
company, and Yara, a leading Yara’s goal, Mr. Cadet explains, of shore, as opposed to sailing
mineral fertiliser company, are “is really to reduce emissions in open waters—Mr. Cadet
building the world’s first fully and to get trucks off the road.” believes more autonomous
autonomous, battery-operated The company wanted a vessel features will find their way onto
container vessel. The Yara that had zero emissions, but ocean-going container ships as
Birkeland will reduce emissions also wanted to reduce the shipping firms continue to seek
and improve road safety by operating costs of the ship as ways to reduce costs.
removing up to 40,000 truck much as possible. A driverless
journeys annually in a densely system was really deployed “to
populated area of Norway. justify the project. We really
According to Olivier Cadet, needed the operating cost
executive vice president for of the vessel to be as low as
products and services at possible, and this is when we
Kongsberg, the vessel will introduced autonomy,” he says.
transport fertiliser from Yara’s
Porsgrunn plant via inland
48
Construction In some European countries,

and Farming up to 30% of cows are
milked by robotic machine, As developing robot
while in the US the share is replacements for humans
Developers have struggled closer to 2%, Mathew Haan, is expensive and complex,
to fabricate robots that can a dairy technology expert at
be useful in the relatively Pennsylvania State University’s job substitutions are most
uncontrolled and free-form agriculture extension program, likely in institutions that
environment of a construction recently told The New York serve many people at
site. However, this reality Times. Haan noted that dairy
is beginning to change, as farmers in Europe face higher large scale.
engineers “dissect” in granular labour costs and get more
detail the tasks demanded generous EU farm subsidies
during construction projects to help pay for the milking
and create machines that use robots, which can cost
cameras, sensors and spatial $200,000. robots are still bad at moving
awareness to work alongside their fingers, even if they
humans. New kinds of robots WHERE SERVICE ROBOTS excel at reading X-rays.
have already been built to lay GO FROM HERE Because developing robot
bricks and install sheetrock. replacements for humans is
These machines are mostly The history of technological expensive and complex, job
used in well-defined work change suggests the substitutions are most likely to
environments, where robots integration of robots into a take place in institutions that
can carry out mechanical or variety of service sectors will serve many people at large
repetitive tasks more rapidly inexorably gain momentum. scale. A major chain hotel with
than humans. In the UK, the But this transition will not take 500 standard rooms is far
first bricklaying robots can place overnight—which means more likely to install robotic
place up to 3,000 bricks a there is time for employers vacuum cleaners than a chic
day. Robotics are poised to
22
and workers to anticipate bed-and-breakfast with only
increase production, improve the service occupations that 11 rooms, each with a different
worker safety, and reduce robots and automation may floor plan.
pollution. displace, and determine the
best ways for robots and
In farming, robots are helping humans to collaborate.
to combat a growing labour Expect robots to take on only
shortage. On dairy farms, the most routine aspects of
robot milking machines can service at first—delivering
tend to cows four times a blood samples for nurses, for
day on average compared example, or frying burgers
with the traditional twice-a- in a fast-food establishment.
day regimen when humans Even as the capabilities of
manage the milking. IFR data machine learning and artificial
shows that a total of 5,386 intelligence rapidly improve,
milking robots were sold in occupations that demand
2017, a 2% increase from the compassion, creativity,
previous year, but estimates and social intelligence will
more rapid uptake in the not easily be replaced by
next three years, with CAGR machines. Nor will those
of 27%. requiring touch and feel—
22
https://www.theconstructionindex.co.uk/news/view/brick-laying-robot-reaches-the-uk 49
The ‘Sophia’ robot at a
manufacturing Expo in
Bangkok, 2018.
HOW TO RESPOND TO THE

RISE OF ROBOTS
Historical evidence from the ongoing threat to workers To better understand the skills
US labour market suggests who are at highest risk to challenges many workers will
that manufacturing workers automation as the service need to overcome to adapt
typically turn to service-sector economy enters its own era to an automated future,
employment when their old of technology-driven job Oxford Economics developed
production jobs disappear. disruption. a Skills Matching Model
Analysing the job moves of in partnership with Cisco
over 35,000 US workers over Size represents the category’s (see page 52). This model
the course of their careers share of new occupations: simulates the more complex
(using longitudinal household red boxes are shrinking dynamics at play when labour
data dating back to 1995), we occupations; green boxes are markets are disrupted, as
find that more than half the growing occupations organisations fill vacancies
workers who left production with compatible workers, who
jobs were absorbed into just At the same time, we have then leave a vacancy in their
three occupational categories: demonstrated how robot previous sector. The model
transport; construction and investment drives economic captures the many small
maintenance; and office and growth and creates new jobs moves involved in the labour
administration work (see Fig. 12). across each economy. Policy- market’s evolution around
makers must anticipate the technological disruption,
But according to our research, mix of forces this level of highlights the new skills the
these three occupational automation will unleash— economy now demands, and
categories are among including job creation as well illustrates how the skills deficit
the most vulnerable to as job displacement across might be best addressed.
automation over the next the economy.
decade. This highlights the
Fig. 12: New occupations secured by workers leaving production jobs

Size represents the category’s share of new occupations: red boxes are shrinking occupations; green boxes are growing occupations
Office & administrative Sales & related Food & preparation Installation &
support services maintenance
Architecture
&
engineering
Management
Healthcare
Transport Construction & extraction Other Building Personal care Arts &
maintenance entertainment
51
UNDERSTANDING THE
RESKILLING CHALLENGE
almost 30% of the redundant
The biggest skills shortfalls appear in cohort lacked the “interactive
skills” future jobs will require.
such skills as negotiation and customer In addition, more than a
service orientation, where humans typically quarter lacked necessary
demonstrate a distinct advantage over robots. “foundational skills” such
as continuous learning,
In a 2017 study entitled “The The study revealed a range reading, and writing. Some
AI Paradox: How Robots Will of acute shortages in skills challenges, such as ICT
Make Work More Human”, technical skills among US and technical skills, require
conducted in partnership workers—gaps that must formal education and regular
with Cisco, Oxford Economics be overcome to realise the refresher courses. Others,
produced ground-breaking productivity gains that new such as the softer skills of
analysis of the skills challenges technologies can offer. But negotiation, persuasion,
facing the US economy. We paradoxically, as technology and customer service,
developed a multi-layered becomes more capable, it is typically require on-the-job
modelling framework to in “human skills” that today’s experience, and might be
simulate how the nature of workforces are most lacking. supported by more flexible,
occupations and the shape The biggest skills shortfalls virtual training options.
of the labour market might appear in such skills as
evolve in response to rapid negotiation, persuasion, and
technological change. Our customer service orientation,
scenario assumptions were which are the skills in
informed by a broad panel which humans typically
of technology experts and demonstrate a distinct
were used to explore both the advantage over robots.
displacement of workers from
their current jobs—based on We applied the same
the unique task profiles of over modelling framework to a
800 occupations—and the different context in a follow-
productivity gains this implied up study, “Technology and
for businesses. the Future of ASEAN Jobs”.
We found that 6.6 million
Our 10-year employment jobs across the ASEAN-6
projection suggested a economies could be made
rebalancing of demand for redundant by 2028 as a result
workers across sectors and of new technology adoption.
occupational groups. US Strikingly, large numbers of
employment in transport and agriculture workers were
warehousing was predicted projected to be driven into
to shrink by around 9% under the service economy as
our scenario, while new a result, leapfrogging the
demand drove a net increase traditional re-employment
in employment in ICT, finance, route through manufacturing.
healthcare, and tourism. We Furthermore, not only did
developed the Skills Matching many vulnerable workers
Model to understand how workers lack the ICT skills needed in a
might navigate these changes. reshaped labour market, but
52
A FRAMEWORK FOR ACTION

In this report, we have Business Leaders Educators
outlined both the potential
value that robots bring • Do not hesitate to seek • Be cognizant of the
and the real fear that the technological solutions to shifting demands for
displacement they cause your business challenges; skills across the economy.
could exacerbate existing the pace of innovation Technology skills are
economic inequalities. is increasing, and global critical, but the economy
Robotization poses a competition continues to is evolving in ways that
fundamental dilemma for intensify. require many non-technical
policy-makers, who must • Get buy-in from your and “soft” skills from
balance the potential gains workforce when making workers, too.
of long-term growth with technology investments. • Develop flexible
the short-term pain of Communicate clearly your approaches to delivering
social dislocation. intentions regarding robots skills training and
directly with your workers. education. In addition to
Our analysis took a Invest in the required keeping formal training
coordinated, evidence-based training and education curriculums fresh and
approach to understand programs in parallel to relevant, this means
how robots are changing the your robotics strategy, to investing more in “lifelong
world. The issues raised will allow workers to anticipate learning” and “on-the-job
affect many different sections and adapt. training” programmes.
of business and society. The • Recognise that • Invest in closer, more
repercussions of robotization technological changes cooperative relations
are interconnected and will disrupt the lives of with local industries,
complex, but the growth in many workers and take to anticipate emerging
robotics is inevitable—these responsibility collectively workforce needs in the
challenges must be embraced to help them onto a local economy based on
and addressed. Policy-makers, path towards future assessments of sectors
business leaders, technology opportunities. that are growing and
companies, educators, and shrinking as a result of new
workers all have a role to play. “Get buy-in from your technology. Anticipate
With this in mind, we offer a workforce when making which new skills will be
framework for action so that technology investments.” needed for new jobs.
different stakeholders can seek • Share data with local
to navigate the challenges and authorities to tailor
opportunities that robotization education and training
will bring. to local strategic skills
priorities and employment
programmes. Intensive
“boot camp” training is
one option to jumpstart
vulnerable or displaced
workers into new fields of
work.
53
Technology Companies Workers Government Policy-

Makers
• Put steps in place • Audit your own job
to mitigate the job to better understand • Explore and analyse the
displacement and the balance between implications of robotization
societal disruptions unique, human skills and for the economy, the
robots can create, to automatable skills, to workplace and the wider
avoid undermining the compete in the right areas society, and adapt policy
consumer market on which and make your job more programmes to the
businesses depend. “robot proof.” evolving landscape.
• Collaborate around • Adopt a “lifetime learning” • Develop collaborative
industry-wide initiatives to mindset. Unlike in previous environments such as
invest in human capital— generations, there are no science parks, living labs,
for example committing to jobs for life. Retraining and other accessible
share a portion of profits. and upskilling will become innovation ecosystems to
• Partner with governments a normal part of the foster skills development
and educational employment landscape. for small and medium-
institutions to take • Support programs that sized enterprises.
responsibility for job develop job flexibility, • Map out the existing
retraining and coaching. even in unionised work skillsets at the local level
Recognise the range settings, to help develop with labour force surveys
of skills required in the cooperative job sharing and analyse alongside
workforce to optimise the and flexitime. regional business trends
potential of technology and growth strategies,
and to grow the market. “Adopt a ‘lifetime learning’ as a basis for strategic
• Look for technological mindset. There are no jobs planning. Incentivise
solutions for the problems for life.” companies and workers
robots create. with fiscal benefits to
engage in local programs
to retrain workers with
“Collaborate around industry- locally relevant skills.
wide initiatives to invest in • Identify the areas most
human capital.” vulnerable to dislocation
from the rise of robots
and develop aggressive,
forward-thinking programs
to counteract those
effects. Explore all policy
options, from infrastructure
investments to training
initiatives and innovative
welfare programmes (such
as universal basic income).
54
APPENDIX:
ECONOMETRIC ANALYSIS
LITERATURE REVIEW DATA
We used data on robot investment (per unit)
To explore the implications of robot from the IFR in Japan, the European Union, the
densification on manufacturing jobs and United States, South Korea, and Australia. The
economic performance we developed two IFR reports robot stock and investments for 50
econometric models, grounded in academic countries over the period from 1994 to 2014. It
literature. Our methodology built most is based on yearly surveys of robot suppliers
prominently on four previous studies: and captures around 90% of the world market.
The information is broken down at the industry
1. Graetz and Michaels (2015) drew on the 23
level, but data availability differs across
data from the International Federation of countries.
Robotics (IFR) in an industry-country panel
specification of 17 countries over the period We augmented the IFR dataset with Oxford
1993-2007. The researchers found that an Economics’ data on GDP, GVA by sector,
increase in the use of robots per hour worked employment by sector, population by 5-year
to boost total factor productivity and average age band, and wages and compensation in
wage levels, and to have a negative impact on total and by sector (where available) from our
hours worked by low-skilled workers, relative datasets. We also used data on trade from the
to middle- and high-skilled workers. COMTRADE database.
2. Acemoglu and Restrepo (2017) used the 24
same dataset but restricted their analysis to We used a sub-national unit of analysis for the
the period 1990-2007 (partly due to data modelling exercise, corresponding to European
limitations and partly to avoid the post-2007 NUTS 2, US Metro areas, Japanese prefectures,
crisis period) and explored localised effects Australian states, and South Korean districts.
of robot densification in the US economy, by We built a panel dataset for 29 countries
exploiting heterogeneity in both local labour (all of EU, plus US, Japan, Australia, Korea)
distributions across industries and national over 11 years within the period (2004-2016),
change in the use of robotics. They found that disaggregated by region and sector.
one additional robot per thousand workers
reduces the US employment-to-population We used a dynamic panel approach (using the
ratio by 0.37%, on average. Generalised Method of Moments estimator,
3. Dauth et al (2017) adopted a similar localised
25
or GMM) to account for secular trends. The
model in the German context with a timelier GMM approach allowed us to use internally
dataset, from 1994-2017. They found that generated instrumental variables (i.e., using
while industrial robots had a negative impact past values as instruments) that helped us
on employment in the German manufacturing establish a causal link between the growth in
sector, there was a positive and significant robot density and employment or productivity.
spillover effect on jobs in non-manufacturing
sectors to offset it. Model specification
4. Chiacchio et al (2018) also built on the
26
Acemoglu approach in a study of six EU The following section describes the two
countries between 1995-2007. They found models used in our analysis.
that one additional robot per 1,000 workers
reduced the employment rate by 0.16-0.20
percentage points. They do not find robust
and significant results on the impact on wage
growth.
56 23
Goerg Graetz, Guy Michaels, “Robots at Work” (CEP Discussion Paper No 1335, 2015)
24
Daron Acemoglu, Pascual Restrepo, “Robots and Jobs: Evidence from US Labor Markets” (NBER, 2017)
25
Wolfgang Dauth, et al, “German Robots: The Impact of Industrial Robots on Workers” (CEPR Discussion Paper 12306, 2017)
26
Francesco Chiacchio, et al, “The impact of industrial robots on EU employment and wages: A local labour market approach” (Bruegel Working Papers, 2018)
Model 1: Manufacturing employment Model 2: Local labour productivity
The model specification, as shown below, The model specification, as shown below,
isolates the impact of robot density (robots isolates the impact of the log of robot density
per 1,000 workers) on the manufacturing on the log of GVA per worker in each region
employment-to-population ratio in each and year, having controlled for the pre-existing
region (r) and year (t), having controlled trend (lagged productivity), globalisation
for secular trends (lagged employment to (trade with China), wage levels (compensation
population ratio), economic performance (GDP per capita), region-specific factors (using
per capita), globalisation (trade with China, panel data), and other trends/year-specific
trade with the rest of the world), wage levels events (year dummies).
(compensation per capita), region-specific
factors (using panel data), and other trends/
year-specific events (year dummies).
ANALYSING REGIONAL IMPACTS
To test the relative impact of robot

Our approach embodies one key difference densification in higher- and lower-income
with the existing literature: we use a dynamic regions, we categorised regions by average
panel method exploring year-on-year compensation per worker. Those that were
variations, while previous studies focused higher than the national average over the
on the cumulative change in employment sample period were labelled higher-income
between two distinct years (e.g., Acemoglu regions; those lower than average were
between 1990 and 2007). Acemoglu and labelled lower-income regions. We used
Restrepo justified this approach as a way of interaction terms (also known as partitioned
avoiding the potentially confounding effects variables) in the same model specifications as
of the recession post-2007. While this is a above to identify the relative differences. As a
valid concern, we mitigated this risk by using sensitivity check, we tested alternative control
more recent years of data (up to and including variables to establish that the coefficients on
2014) and dynamic panel methods. In addition, the control variables were significant and of
the cumulative change method would have the right sign.
been inadequate for the period 2004-2014
due to the changes in employment over that ANALYSING SHORT-TERM AND LONG-TERM
period (which declined in the recession years IMPACTS
and increased in the subsequent years) and
the change in robot stock (which increased To control for the fact that the impact of
consistently over time). By using appropriate robotization may take time to impart its full
control variables, including time dummies, in effect on employment and productivity, we
a dynamic panel setting, our model accounts used lag variables. This involved investigating
for the effects of the recession and isolates the the relationship between robot installations in
impact due to robots. one period, and employment and productivity
in subsequent periods, with various degrees of
time lag.
57
ESTIMATOR SELECTION ECONOMETRIC RESULTS
In order to produce robust estimates, we used Model 1: Manufacturing employment

various tests to select the most appropriate
estimator. We ran the Wooldridge test for The regression output for our three preferred
autocorrelation to determine the need for a
28
model specifications are presented in Table A1.
dynamic panel approach. The data was found
to be autocorrelated, meaning employment 1.1 Our preferred model for the average
and productivity in current periods might be employment impact across all regions.
affected by past trends in the same variables. 1.2 Our preferred model to estimate the
variable impacts in higher- and lower-
We used the Arellano Bond/Blundell Bond income regions.
estimator (“System GMM”) to account for the 1.3 Our preferred model to estimate the
presence of such “dynamic effects” in the data. productivity impact across all regions.
This method augments the Arellano–Bond
estimation (“difference GMM”) by allowing In all three models, the lag dependent variable
the introduction of more instruments and can is low but significant at the 5% level. The
dramatically improve the efficiency of the coefficients on the dependent variables are
model. Using a dynamic panel model also
29
significant and of the expected sign in all
enables us to identify overall coefficients for three models. All three models include year
explanatory variables, corresponding to the dummies. We have also tried estimating other
long-term effects as well as contemporaneous specifications across all three models but were
ones corresponding to the short-term. unable to obtain a satisfactory model.
Even where a dynamic model is the preferred Table A1: Change in the number of jobs due
specification, it still comes with potential risks. to one additional robot
The most prominent risk is omitted variable
bias. For example, the growth in ICT services Short-term impact Long-term impact
corresponds broadly with the rise in industrial 1.1 Average across -1.3 -1.6
robots but because our unit of analysis is at all regions
the regional level, there is insufficient data to 1.2 High-skilled -1.0 -1.3
control for ICT spend at local levels across the regions
29-country sample. We attempted to mitigate 1.3 Low-skilled -1.7 -2.2

regions
this risk by using a proxy for local productivity
growth (wage growth) and past values of the
dependent variable as instruments, which The employment model implies that a 1 unit
reduces the need to identify all alternative increase in robots per 1,000 workers reduces
instrumental variables. Instruments in the average employment ratio by 1.5 in the
dynamic models also account for serial short run and 2.0 in the long run. The impact is
correlation between past and current values of lower for high skilled regions, at 1.2 and higher
employment and productivity. for low skilled regions, at 2.0.
58 27
The Wooldridge test was implemented using the xtserial command in Stata using a specification comprising of dependent and independent variables.
We tested for serial correlation in the data using multiple specifications with different independent variables.
28
The “system GMM” builds a system of two equations—the original equation and the transformed one. The assumption of no correlation between the
first differences of instrument variables and fixed effects in system GMM allows for the inclusion of time-invariant regressors, which would disappear in
“difference GMM.”
We converted this into the average impact of COMPARISONS

one additional robot on the number of jobs
using the average robots per capita ratio Table A3 presents a comparison of our results
across regions. Using these results, we find with previous academic literature. All results
that a 1-unit increase in robot stock reduces regarding the employment effect have been
the number of jobs by 1.3 in the short run converted into the change in number of jobs
and 1.6 in the long run across all regions. The due to one additional robot. Our results are
impact is lower in high-skilled regions and the same sign but slightly lower than other
higher in low-skilled regions. (Table A1) studies, which find the impact of one robot
displaces between 3 to 6 jobs depending on
Model 2: Local labour productivity the sector, time period and country covered.
Graetz and Michaels (2015) and Dauth et al
Our preferred specification for estimating the (2017) also find that an increase in robots
productivity impact of robotization implies increases productivity.
that a 1% increase in robot stock leads to a
0.1% increase in productivity in the short-
term and a 0.3% increase in productivity in
the long-term. Due to the range of factors
influencing the productivity impact in a long-
term timeframe, we use the short-term impact
as a basis for our analysis.
Table A2: Change in productivity (GVA per

worker) due to a 1% increase in robots
Short-term impact Long-term impact

Average across 0.1% 0.3%
all regions
Table A3: Change in the number of jobs due

to one additional robot
Study Geography Time frame Impact on jobs (of one robot) Impact on labour productivity
Graetz and 17 countries (US, four- 1993-2007 No effect on total hours Increase in total factor
Michaels (2015) teen European coun- worked, but a reduction productivity
tries, South Korea, and in hours of low-skilled and
Australia) middle-skilled workers.
Acemoglu and Restrepo US 1990-2007 Loss of 3 to 6 jobs. -
(2017)
Dauth et al (2017) Germany 1994-2014 Loss of 2 manufacturing jobs Increases productivity
offset by a gain of 2 additional
jobs in the service sector.
Chiacchio et al (2018) 6 EU countries 1995-2007 Loss of 3 jobs. -
59
About Oxford Economics
Oxford Economics was founded in 1981 as a June 2019

commercial venture with Oxford University’s All data shown in tables and charts are
business college to provide economic Oxford Economics’ own data, except where
forecasting and modelling to UK companies otherwise stated and cited in footnotes, and
and financial institutions expanding abroad. are copyright © Oxford Economics Ltd.
Since then, we have become one of the world’s
foremost independent global advisory firms, This report is confidential and may not be
providing reports, forecasts and analytical published or distributed without the prior
tools on more than 200 countries, 250 written permission of Oxford Economics Ltd.
industrial sectors, and 7,000 cities and regions.
Our best-of-class global economic and The modelling and results presented here
industry models and analytical tools give us an are based on information provided by third
unparalleled ability to forecast external market parties, upon which Oxford Economics has
trends and assess their economic, social and relied in producing its report and forecasts in
business impact. good faith. Any subsequent revision or update
of those data will affect the assessments and
Headquartered in Oxford, England, with projections shown.
regional centres in London, New York,
and Singapore, Oxford Economics has To discuss the report further, please contact:
offices across the globe in Belfast, Boston,
Cape Town, Chicago, Dubai, Frankfurt, James Lambert:
Hong Kong, Houston, Johannesburg, Los jlambert@oxfordeconomics.com
Angeles, Melbourne, Mexico City, Milan, Paris, Edward Cone:
Philadelphia, Sydney, Tokyo, and Toronto. edcone@oxfordeconomics.com
We employ 400 full-time staff, including
more than 250 professional economists, Oxford Economics,
industry experts and business editors—one Broadwall House,
of the largest teams of macroeconomists 21 Broadwall,
and thought leadership specialists. Our London,
global team is highly skilled in a full range of SE1 9PL
research techniques and thought leadership UK
capabilities, from econometric modelling,
scenario framing, and economic impact Tel: +44 207 803 1400
analysis to market surveys, case studies, expert
panels, and web analytics. www.oxfordeconomics.com
Oxford Economics is a key adviser to The image on page 2 is by Field Robotics,

corporate, financial and government and all other images in this report are by
decision-makers and thought leaders. Our Shutterstock (www.shutterstock.com).
worldwide client base now comprises over
1,500 international organisations, including
leading multinational companies and financial
institutions, key government bodies and
trade associations, and top universities,
consultancies, and think tanks.
60
The ‘Pepper’ robot assistant
in Italy, 2018.
Global headquarters Europe, Middle East
Oxford Economics Ltd and Africa
Abbey House
121 St Aldates Oxford
Oxford, OX1 1HB London
UK Belfast
Tel: +44 (0)1865 268900 Frankfurt
Paris
London Milan
Broadwall House Cape Town
21 Broadwall Dubai
London, SE1 9PL
UK Americas
Tel: +44 (0)203 910 8000
New York
New York Philadelphia
5 Hanover Square, 8th Floor Mexico City
New York, NY 10004 Boston
USA Chicago
Tel: +1 (646) 786 1879 Los Angeles
Toronto
Singapore San Francisco
6 Battery Road Houston
#38-05
Singapore 049909 Asia Pacific
Tel: +65 6850 0110
Singapore
Sydney
Hong Kong
Tokyo
Email:
mailbox@oxfordeconomics.com
Website:
www.oxfordeconomics.com

Report - How Robots Change The World

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Report - How Robots Change The World

Загружено:

Авторское право:

Доступные форматы

HOW ROBOTS

What drives the robot rise? 13

The impact of robots on manufacturing jobs 19

The Robot Vulnerability Index 25

The robotics dividend 35

Robots are coming to the service sector 40

How to respond to the rise of robots 51

Appendix: econometric analysis 56

For both people and

The rise of robots has already had a profound effect on

as many as 20 million additional manufacturing jobs worldwide

Lower-income regions are more at risk

growing levels of economic inequality and political polarisation,

Given the stakes, policy-makers need an early warning system

In many cases, our Index highlights that the most vulnerable

Fig.1: Job losses from robots hit lower-income regions harder 4

Change in number of jobs due to one additional robot

-2.5 -2.0 -1.5 -1.0 -0.5 0.0

Source: Oxford Economics

The $5 trillion robotics dividend

The future of service robots

One key consideration for anticipating the pace of robot

It will be difficult for machines to replace humans in service

The challenges will be daunting. Our analysis of the job

Robots are on the rise as never before. Preparing for and

of installation. The robotics He urged policy-makers to businesses to boost

WHAT IS OUR DEFINITION OF

WHAT DRIVES THE

Fig. 2: Robot installations by country, 2000 to 2016 7

New robot installations

The automotive sector has of new robot installations in

user of robots: innovations to 31% in 2016, from 21% in

Rubber & Other

Automotive High tech

THREE REASONS FOR THE

Fig. 4: Robot adoption growing faster outside the automotive sector

Percentage change in robot densification between 2011 and 2016

South Korea 51%

Dark bar=Automotive sector

major producer of consumer manufacturing sector—moreso

Fig. 5: Chinese scope for catch-up in robot density (2016)

Column=Robots per 10,000 workers (LHS)

Source: IFR, Oxford Economics

THE IMPACT OF ROBOTS ON

A NOTE ABOUT OUR

taken the place of 1.5% of of robotization in the

Fig. 7: Cumulative jobs losses implied by automation since 2000

Cumulative jobs lost since 2000

Source: Oxford Economics

REGIONAL IMPACTS HIT levels are either above or

Fig. 8: Projected cumulative jobs losses by automation, up to 2030

-15,000,000 South Korea

Source: Oxford Economics

Fig. 9: Manufacturing job losses skew towards lower-income regions

Change in number of jobs due to one additional robot

-2.5 -2.0 -1.5 -1.0 -0.5 0.0

Source: Oxford Economics

Manufacturing workers in contrast, a significantly higher carry out are—on balance—

THE ROBOT VULNERABILITY

The index is thus designed

Trend #3: Rural Country-by-country analysis

Most State: Index Score: Least State: Index Score:

Most Region: Index Score: Least Region: Index Score: