C Impact of technology on

labour market outcomes

This section considers the effects of technology on the level and
composition of employment and wages. Technological progress,
by increasing the productivity of factors of production, expands an
economy’s production possibility frontier, so that the same amount
of output can be produced with fewer resources, or more output
can be produced with the same amount of resources.
Contents
1. Introduction 76

2. Overall net employment and wage effects of technology 78

3. T he impact of technology on skills and work tasks 83

4. T
 echnology and the future of work 90

5. Conclusions 99

Some key facts and findings

• Technological progress can assist workers, through labour-augmenting
technology, or replace them, via automation. In both cases, the overall effects
on the market’s demand for labour are ambiguous.

• Current technological progress has led to a higher relative demand for skilled
workers and a lower relative demand for workers performing routine activities.

• The use of computers in the workplace has been the central force driving
changes in the wages of skilled workers relative to the wages of unskilled
workers.

• Various methodologies have been developed to estimate the share of jobs at

risk from automation and computerization. According to these methodologies,
the estimated share of jobs at risk tends to be larger in developing countries
than in developed countries because of their larger share of employment in
routine occupations.

• Automation does not necessarily equate with future unemployment because

the development, adoption and widespread use of future technologies will
hinge on a number of factors, including feasibility and affordability, as well
as the legal and regulatory framework in particular countries and public
acceptance of new technology. However, future automation is likely to be
disruptive for workers whose skills will become obsolete and who face the risk
of job losses and having to switch tasks and jobs.
 WORLD TRADE REPORT 2017

1. Introduction good examples of labour-augmenting technologies.

Automation technologies, in turn, complete cognitive
Technology can be broadly defined as "the state or manual tasks without human intervention.1
of knowledge concerning ways of converting Repetitions (such as executing loops in a programme
resources into outputs" (OECD, 2011a) or as the code or corking wine bottles in a winery) are good
"machinery and equipment developed from the examples of automation.
application of scientific knowledge" (Oxford English
Dictionary). There are two types of processes Several recent studies show the positive effects of
involved in producing a new technology: invention new technology on labour productivity. In the valve-
and innovation. Invention involves the formulation of manufacturing industries of the United Kingdom
scientific principles or processes. Innovation entails and the United States, the adoption of computer-
the direct application of this knowledge to a useful controlled technology resulted in a substantial
purpose in response to presumed profit opportunities. increase in productivity by reducing setup time,
As argued in Section B of this report, innovation can production time and inspection time (Bartel et al.,
take the form of new products or a new quality of a 2007). Collard-Wexler and De Loecker (2015),
product (product innovation) or of new production who study the US steel industry, show that the
techniques (process innovation). New technologies (partial) displacement of older technology (vertically
stemming from innovation have effects on the integrated producers) with a new production process
economy, and on society more generally, that are (the minimill) was responsible for over one-third of
proportional to how widely they are adopted. General the increase in the industry’s total factor productivity,
purpose technologies (GPTs) – technologies that or 38 per cent in the period 1963 to 2002. Shifts to
transform both household life and the ways in which energy-efficient technology may increase workers’
firms conduct business (Jovanovic and Rousseau, productivity, such as the move from standard
2005) – have more widespread effects across firms fluorescent lighting to LED lighting in factories, which
and sectors than technologies destined for particular improves working conditions in hot humid climates
production processes or purposes. in Bangalore (India) due to the lower heat emissions
produced by LED lighting (Adhvaryu et al., 2016). 2 In
Technology can complement workers (so-called the services sector, a travel agency in China which
labour-augmenting technology) or substitute for them employs 16,000 workers saw a 13 per cent rise in
(so-called labour-saving technology, or automation). labour productivity for home-based workers (Bloom et
If technology complements workers, this implies that al., 2015). Box C.1 shows how technological change
it increases labour productivity. Autopilot technology that raises labour productivity can be conceptualized
on planes or statistical software for data analysis are in a production possibility frontier (PPF) framework.

Box C.1: Technological change in a production possibility frontier (PPF) framework

The production possibility frontier (PPF) of an economy describes the amount of output that can be produced
for given amount of inputs, measured in efficiency units.

To see how technological change affects an economy’s PPF, consider the simplest possible case of a two-
sector economy (х and у), with production in both sectors being subject to diminishing returns to a single
factor of production: labour. Diminishing returns imply that, in each sector, the marginal productivity of labour
diminishes in the amount of labour employed in that sector. With diminishing returns, the PPF is concave
(as plotted in Figure C.1). Another characteristic of production functions giving rise to concave PPFs are
different factor intensities across the two sectors (even under constant returns) (see Snyder and Nicholson,
2010, pages 416-7).

Production functions for the two goods are х =  and у =  , where ,i = х,у represent the efficiency
units of labour allocated to each sector (physical units of labour multiplied by a sector-specific technology
parameter Α i ):

 = Α х L х and  = Αу Lу

76
 TRADE , TECHNOLOGY AND JOBS

Box C.1: Technological change in a production possibility frontier (PPF) framework (continued)

The PPF of this economy is represented by the following quarter circle:

у 2 = Αу L −   х 2 ,
where
L = L х +Lу

is the total amount of labour in the economy. This is represented by the solid line in panels (a), (b) and (c)
of Figure C.1 (the solid line in panels (a)-(c) of Figure C.1 is drawn assuming Α х = Αу = 1 and L = 100).

Technological progress – an increase in Α – causes an outward shift in the PPF. This is due to the fact that
an increase in Α increases the marginal productivity of labour in the sector(s) where it occurs, so that the
economy can produce more with the same amount of physical units of input (in this case, with the same
number of workers). Marginal productivity of labour in sector i, i = х,у is equal to ( . This increases (
in Αi and, due to diminishing returns, decreases in Li .

Panels (a)-(c) of Figure C.1 represent three types of technological progress. In panel (a) such progress is
neutral, since it increases labour productivity equally in the two sectors. This is referred to as Hicks-neutral
or balanced productivity change. This type of technological progress could, for instance, be due to the
introduction of a general-purpose technology (GPT), adopted in all sectors. This is represented by a parallel

C. I MPACT OF TECHNOLOGY
OUTCOMES
ON LABOUR MARKET
outward shift in the PPF from the solid to the dashed line – the dashed line is drawn assuming a doubling of
productivity in each sector, so that Α х = Αу = 2, and keeping L =100. In panel (b), technological progress is
biased in favour of sector х: it is assumed that technology improves labour productivity only in х, and not in у.
Panel (c) presents the opposite case of technological change biased in favour of sector у.

The dashed line in panel (b) is drawn assuming a doubling of productivity only in sector х, Α х = 2, and keeping
Αу  = 2 and L = 100. Conversely, the dashed line in panel (c) is drawn assuming a doubling of productivity only
in sector у, Αу  = 2, and keeping Α х  = 2 and L = 100. Clearly, even if technological change occurred in both
sectors, it would still be biased in favour of one sector if the increase in labour productivity was larger in that
sector than in the other.

In all cases, technological change causes an outward shift in the PPF, allowing the economy to produce
(and consume) more for a given amount of inputs. Note that the equilibrium of the economy (not shown in the
figures) will be at the point of tangency between the PPF and the highest indifference curve representing
consumers' preferences. The new equilibrium could also be reached with international trade. Trade, rather than
shifting the PPF, changes relative prices and allows for a separation between production and consumption
decisions. In this sense, trade and technology could have the same effects in general equilibrium.

Figure C.1: Technological change in a production possibility frontier framework

Original production possibility frontier New production possibility frontier following technological change

14 14 14

12 12 12

10 10 10

8 8 8
Sector y

Sector y

Sector y

6 6 6

4 4 4

2 2 2

0 0 0
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
Sector x Sector x Sector x
(a) Neutral technological change (b) х - biased technological change (c) у - biased technological change
Source: WTO Secretariat.

77
 WORLD TRADE REPORT 2017

The use of computers, information technology unprecedented pace. Graetz and Michaels (2015)
(IT) and the internet has effects that go far beyond report that from 1993 to 2007, mean robot density
labour productivity. Digital trade and international increased by more than 150 per cent in 17 industrial
e-commerce reduce transaction costs and boost the countries. Boston Consulting Group (2017) reports
transparency of markets (Lippoldt and House, 2017). that the number of industrial robots in operation could
They allow consumers a more convenient and efficient increase from the current figure of between 1.5 and
shopping experience, raising living standards beyond 1.75 million to between 4 and 6 million by 2025.
real GDP growth (Hulten and Nakamura, 2017). More These significant increases in automation, and the
generally, the use of information and communications potentially even wider use of robots in non-industrial
technology (ICT) increases the availability of market sectors, have sparked a debate on the future of work,
information, leading to a better and more stable in particular on whether the demand for human labour
functioning of markets (consider improved job might decrease permanently, leading to a “jobless
matches in the labour market due to more readily future” characterized by artificial intelligence and
available information on wages, job vacancies, skill robotics at a massive scale. Section C.4 reviews the
requirements, and labour market conditions). evidence on the pace of technology adoption and the
arguments of technology optimists and pessimists
The effects of labour-augmenting and labour- regarding the future of work. The section also
replacing technologies on labour demand are discusses the implications for skills development.
ambiguous. An example is the introduction of
technologies in agriculture. The related increases
in agricultural labour productivity can be correlated 2. Overall net employment and
with a reduction in agricultural employment if, as a wage effects of technology
result of falling relative prices of agricultural goods,
economy-wide prosperity increases and household Throughout history, technological change has often
demand for agricultural produce grows less than been a source of anxiety for many workers. In England
demand for other goods. Automation, in turn, is between 1811 and 1816, a group of workers who
intrinsically labour-saving, as it reduces labour called themselves “Luddites” destroyed machinery
requirements per unit of output produced. However, which they believed was threatening their jobs,
even labour-saving technology can be associated especially in cotton and woollen mills. Nineteenth-
with rising labour demand due to lower production century economists like Karl Marx and David Ricardo
costs. The first part of this section reviews the predicted that the mechanization of the economy
mechanisms that give rise to the ambiguous effects would worsen conditions for workers, ultimately
of technology on employment, and discusses their condemning them to live on a subsistence wage. In
empirical relevance. the last century, too, prominent economists like John
M. Keynes (in the 1930s) and Wassily Leontief (in
By making some products or production processes the 1950s) expressed the fear that more and more
obsolete, and by creating new products or expanding workers would be replaced by machines, and that
demand for products that are continuously innovated, this would lead to technological unemployment. 3
technological change is necessarily associated with More recently, Brynjolfsson and McAfee (2014) have
the reallocation of labour across and within sectors claimed that such disruptive technologies reduce the
and firms. Such technology-induced reallocations demand for labour and put workers at a permanent
affect workers differently, depending on their skills or disadvantage.
on the tasks they perform. ICTs tend to be used more
intensively and more productively by skilled workers This section discusses the mechanisms behind the
than by unskilled workers. Automation tends to affect relationship between technological change and
routine activities more than non-routine activities, overall employment, and the empirical evidence
because machines still do not perform as well as related to those mechanisms.
humans when it comes to dexterity or communication
skills. In Section C.3, evidence is presented in favour (a) Theoretical mechanisms
of the hypothesis that the labour market effects of
technology are relatively more favourable to skilled As famously shown by Baumol (1967), technologically
workers and to workers performing tasks that are advancing sectors – that is, those experiencing more
harder to automate. rapid productivity growth – tend to contract as a share
of employment, while technologically lagging sectors
Advances in smart technology, artificial intelligence, – that is, those with slow productivity growth – tend
robotics and algorithms, often referred to as the to expand. This is because technological progress
fourth industrial revolution, are taking place at reduces labour requirements per unit of output
78
 TRADE , TECHNOLOGY AND JOBS

produced. Figure C.2 plots indexes of manufacturing The deployment of labour-saving technologies is not
employment (as a share of total employment) and a recent phenomenon. Without going back as far as
labour productivity (output per employed person the Industrial Revolution of the late eighteenth and
in manufacturing) between 1970 and 2011 for two early nineteenth centuries, one could mention the
major industrial countries (Germany and the United spectacular decrease in the share of agricultural
States). During the six decades covered by the employment in developed countries during the last
data, manufacturing employment as a share of total century. Autor (2015), for instance, reports that in
employment fell substantially, but manufacturing 1900, 41 per cent of the US workforce was employed
labour productivity increased. 4 in agriculture; by 2000, that share had fallen to 2 per
cent, mostly due to a wide range of technologies
These trends can be correlated with the evolution including automated machinery, such as field
of one particular type of automation, namely the machinery and irrigation systems.7
use of industrial robots, since the mid-1990s. 5
Between 1993 and 1997, robot density (defined Labour-saving technologies, however, are not only
as the number of robots per million hours worked) deployable in the primary and secondary sectors. The
increased by 160 per cent in Germany and by 236 introduction of earth-moving equipment and powered
per cent in the United States.6 Therefore, increasing tools displaced manual labour from the construction
automation is broadly correlated with lower labour sector, for instance (Autor, 2015). Occupations such
requirements per unit of output in manufacturing as telegraph or elevator operators, which figured in the
adopting such labour-saving technologies. Graetz 1950 US Census, have been eliminated altogether,
and Michaels (2015) estimate that, in their sample of due to technological obsolescence in the case of

C. I MPACT OF TECHNOLOGY
14 industries in 17 countries from 1993-2007, robot the former and to automation of the latter (Bessen,

OUTCOMES
ON LABOUR MARKET
densification increased labour productivity by about 2017). 8 Thanks to advances in ICT, the automation of
0.37 percentage points. According to the authors, logistics and processing and of self-service (e.g. in
this figure is fairly comparable to the estimated total document creation and management, which no longer
contribution of steam technology to British annual require clerical support, or in retail self-checkout)
labour productivity growth, which was around 0.35 and digitization (e.g. of data entry and of publishing/
percentage points, but was sustained over a period printing) are all services sector activities where
that was about four times longer, from 1850 to 1910 labour-saving technologies can be deployed and can
(Crafts, 2004, cited in Graetz and Michaels, 2015). substitute for workers (see World Bank, 2016).

Figure C.2: Evolution of employment and output per worker in manufacturing of selected
industrial countries (1970 to 2011)

Germany United States

140 45 180 30

40 160
120
25
35 140
Employment share (percentage)

Employment share (percentage)

100
Labour productivity index

Labour productivity index

30 120 20

80 25 100
15
60 20 80

15 60 10
40
10 40
5
20
5 20

0 0 0 0
1970

1974

1978

1982

1970
1986

1974
1990

1978
1994

1982
1998

1986
2002

1990
2006

2011

1994

1998

2002

2006

2011

Employment share Labour productivity index

Source: US Department of Labor Bureau of Labor Statistics (BLS) International Labor Comparisons (ILC) data.
Notes: Labour productivity measured as output per employed person (index, 2002 = 100 and percentage).

79
 WORLD TRADE REPORT 2017

A simple conceptual framework to understand the saving technology in the form of routine-biased
effects of the deployment of new technologies on technological change (see Section C.3), argue that
overall labour demand is based on the balance technological change creates high-tech jobs which
between substitution and compensation mechanisms generate additional demand in non-tradable sectors.10
(Vivarelli, 2015). It is argued in Section B that if One could cite as an example the ICT sector, which
technological change takes the form of a new product includes manufacturing sectors, e.g. office machinery,
that is substituted for an older one, technological and services sectors, e.g. telecommunications.11
innovation lowers the demand for the old product In terms of employment, the ICT sector is small,
while it raises the demand for the new product. with ICT occupations accounting for 1 per cent of
This translates into an upward shift in the demand employment in developing countries, and 2 to 5 per
for labour used to produce the old product and a cent in Organisation for Economic Co-operation
downward shift in the demand for the workers who and Development (OECD) countries (World Bank,
produce the new products. In other words, the 2016).12 Moreover, the ICT sector only accounts for a
substitution mechanism at work operates via product minimal proportion of employment creation, because
displacement. While the resulting adjustment (with it is by definition capital-intensive.13 For each job
some jobs being destroyed and others being created) created by the high-tech industry, however, around
may not be without frictions, in this context it is worth five additional, complementary jobs are created in the
noting that higher labour demand in the growing local economy, mostly in the non-tradable services
sector can partially or fully offset lower labour demand sector (Moretti, 2010; Moretti and Thulin, 2013; Goos
in the declining sector, a compensation mechanism et al., 2015).
that can produce ambiguous effects on overall labour
demand. 9 Fourth, and most importantly, one should consider that
technology adoption by firms is a decision affected by
In the case of labour-replacing automation (analysed various factors, including changes in relative prices of
in Section B as a reduction in the price of capital), production factors.14 In the theoretical framework of
technological change induces firms to adopt more Acemoglu and Restrepo (2016), as a factor becomes
capital-intensive technologies and to substitute labour cheaper, the range of tasks allocated to it expands
for capital, lowering labour demand at any given wage and also generates incentives for direct technologies
rate (substitution effect). There are, however, several that utilize this factor more intensively.15 This implies
compensation mechanisms that can counterbalance that by reducing the effective cost of producing with
the initial labour-saving impact of automation, and of labour, automation discourages further automation
process innovation in general (Vivarelli, 2015). First, and generates a self-correcting force towards
while workers are displaced in those industries that stability in the long run. Thus, it is possible that rapid
introduce the technology incorporated in the new automation need not disrupt labour, but might simply
machinery, additional workers are needed in the be a transitioning phase towards new technologies
industries that produce the new machinery. benefiting labour.16

Second, automation (and process innovation more The extent to which the compensation mechanisms
generally) reduces average costs. Acemoglu and described above can counterbalance the labour-
Restrepo (2017) show that this leads: i) to a price- saving impacts of technological change depends on
productivity effect (as the cost of production goes several underlying assumptions and conditions.17
down, the industry can expand and increase its In this context, it is sufficient to point out that the
labour demand); and ii) to a scale-productivity effect question of whether technological change increases
(the reduction in costs due to automation results in or decreases overall employment and wages is,
an expansion of total output, raising the demand for ultimately, an empirical one, which will be analysed in
labour in all industries). Similarly, Vivarelli (2015) the next subsection.
argues that lower average costs can either translate
into lower prices (if the industry market structure is (b) Empirical evidence
perfectly competitive), stimulating product demand,
or into extra profits (if the industry structure is not So far, the concerns expressed by prominent
perfectly competitive). If these extra profits are nineteenth- and twentieth-century economists
re-invested in the firm, this investment can create new like Marx, Ricardo, Keyes and Leontief, that the
jobs. replacement of workers by machines would lead to
technological unemployment, have not materialized.
A third compensating effect potentially leading Although some individuals may have lost their jobs
to higher labour demand relates to local demand permanently, the past two centuries of technological
spillovers. Gregory et al. (2016), who study labour- progress have not made human labour obsolete.
80
 TRADE , TECHNOLOGY AND JOBS

The employment-to-population ratio rose during the did not fall because ATMs allowed banks to operate
20 th century, and there is no apparent long-run branch offices at lower costs. This prompted banks to
increase in the unemployment rate (Autor, 2015). open many more branches, offsetting the loss in teller
jobs (Bessen, 2015).
Case study evidence focusing on particular
sectors and occupations shows that, even after There is abundant econometric evidence on the
the introduction of labour-replacing technologies, overall employment effects of technological change.
employment increased when those technological The studies in this field can be classified according
changes led to significant scale effects. to the type of technological change considered (i.e.
product innovation, process innovation, routine-biased
Bessen (2015) reports the telling examples of technological change,18 computerization or exposure
19 th -century cloth weaving and 20 th -century cash- to industrial robots), the income level of an economy
handling. During the 19 th century, 98 per cent of (developed or developing) and the unit of analysis
the labour required to weave a yard of cloth was (firm, industry or local labour markets). The general
automated. However, the number of weaving jobs conclusion from this literature is that technology has
actually increased. Automation drove the price of affected the structure of employment, but has had
cloth down, increasing the (highly elastic) demand for small (and mostly positive) effects on the overall level
cloth, resulting in net job growth despite the labour- of employment (Vivarelli, 2014; Arntz et al., 2016b).
saving technology (Bessen, 2015).
A positive link between technology and employment
In the United States, ATMs (i.e. automatic teller is especially evident when research and development

C. I MPACT OF TECHNOLOGY
machines) were introduced in the 1970s, and their (R&D) and/or product innovation are adopted as

OUTCOMES
ON LABOUR MARKET
number rose fourfold (from 100,000 to 400,000) proxies of technological change, as well as when the
between 1995 and 2010. ATMs took over cash- focus turns to high-tech sectors (Bogliacino et al.,
handling tasks, yet since 2000 the number of 2012). There are, however, a few relevant exceptions,
full-time equivalent bank tellers has increased by with studies showing negative labour-demand effects
2 per cent per year, substantially faster than the of technological change.
overall US labour force (see Figure C.3). Employment

Figure C.3: ATMs and full-time equivalent bank tellers in the United States (1970 to 2010)

500

400
Tellers, ATMs (1,000s) in the United States

300

200

100

0
1970 1980 1990 2000 2010

Full-time equivalent workers ATMs

Source: Bessen (2015).

81
 WORLD TRADE REPORT 2017

At the country and industry levels, in a sample of of 5.6 workers in their favourite specification) and
32 industries in 19 developed economies between wages by 0.25-0.5.19
1970 and 2007, Autor and Salomons (2017) find
that productivity growth has been employment- Conversely, in a study focusing on 238 regions
augmenting rather than employment-reducing. In across 27 European countries over the 1999-2010
particular, the fall in industry-level employment as period, Gregory et al. (2016) find that routine-
industry productivity rises (in line with Baumol, replacing technological change led to overall positive
1967) is more than offset by the rise in country-level labour demand effects. In terms of the mechanisms
employment as aggregate productivity rises. This they propose and that were discussed above, this
indicates that productivity growth in each sector suggests that the labour demand and the local
generates employment growth spillovers elsewhere demand spillover effects dominated the substitution
in the economy. These spillovers are sufficiently large effect. The authors argue that the net effect of routine-
to more than offset employment losses in industries replacing technological change on labour demand
making rapid productivity gains. was an increase of between 1.9 and 11.6 million jobs
across Europe, depending on whether non-wage
In a similar vein, Bessen (2017) finds that between income (returns from technology investments) fed
1984 and 2007, computer use had a significant back into the local economy in terms of consumption
negative effect on manufacturing employment in or whether it was spent abroad. 20
the United States, but a mild positive employment
effect on other industries. Ebenstein et al. (2015), At the firm level, several studies contrast the effects
conversely, argue that greater use of computers of product innovation and of process innovation,
and capital equipment is associated with lower finding negative employment effects of process
employment, higher unemployment and lower labour innovation, which tend to be compensated by positive
force participation across all US occupations. Graetz employment effects of product innovation. Using
and Michaels (2015), using International Federation data on firms in the manufacturing and services
of Robotics (2012) data, estimate that across 17 sectors in France, Germany, Spain and the United
countries over the period 1993-2007, while increased Kingdom, Harrison et al. (2014) find that product
utilization of robots (robot densification) in a range innovation has a positive impact on employment, but
of different industries – particularly in transport that process innovation has a displacing effect on
equipment, chemicals and metal industries – affected employment. However, the positive impact of product
the composition of employment and wages across innovation generating employment is larger than
skill groups (see Section C.3), there were no adverse the displacement effect of process innovation, and
aggregate employment effects (i.e. no reduction therefore the net effect of innovation on employment
in aggregate hours worked) of robot densification. tends to be positive. Similarly, Hall et al. (2008) find
Moreover, they estimate the positive and statistically a low but positive effect of product innovation on
significant effects of robot densification on mean employment in Italy, and no displacement effect from
wages. This implies that some of the productivity process innovation.
gains from robot densification were shared with
workers (Graetz and Michaels, 2015). Concerning developing countries, Ugur and Mitra
(2017), who conduct a review of 43 qualitative
Some recent studies consider the effects of studies and 12 empirical studies, report that the
technological change on local labour markets. In a effect of technology adoption on employment is
study using commuting zones in the United States as more likely to be positive when the evidence is
units of analysis, Autor et al. (2015) find that exposure related to skilled labour employment and product
to routine task specialization had largely neutral innovation. The qualitative studies included in the
overall employment effects between 1980 and 2007, review by Ugur and Mitra (2017) further suggest that
only affecting the occupational composition within the employment effects of technology adoption are
sectors. Acemoglu and Restrepo (2017) consider how more likely to be positive in the presence of strong
exposure to industrial robots affected employment linkages between innovative firms/farms/industries
and wages in local labour markets between 1990 and and the rest of the economy, and in the presence of
2007, and they estimate large and robust negative governance institutions that encourage and facilitate
effects of robots on employment and wages across technology adaptation instead of relying on off-the-
commuting zones. They suggest that an additional shelf technology only.
robot per thousand workers reduces employment
to population ratio by about 0.18-0.34 percentage Most recent empirical work on the overall
points (one more robot being associated with a employment effects of technological change in
reduction in relative commuting zone employment developing countries uses firm-level data. The most
82
 TRADE , TECHNOLOGY AND JOBS

comprehensive study is that of Cirera and Sabetti employment. These effects are analysed in the next
(2016), who use a sample of over 15,000 firms in subsection.
Africa, Central Asia, Eastern Europe, the Middle East
and North Africa (MENA), and South Asia. They show
that new sales associated with product innovations 3. The impact of technology
tend to be produced with just as much or higher levels on skills and work tasks
of labour intensity. This positive employment effect
of product innovation is largest in least-developed The previous subsection considered the overall
countries (LDCs) and in the African region, where employment level effects of technology. Due to
firms are less advanced in terms of technological various mechanisms, including productivity effects
development. and product demand spillovers, it was argued
that labour-saving technologies need not reduce
Cirera and Sabetti (2016) also show that process overall employment. This subsection considers the
innovations that involve the automation of production heterogeneous effects of technology on workers,
do not have a short-term negative impact on firm depending on their skills and on the tasks they
employment. 21 However, there is some evidence of perform at work.
a negative effect of automation on employment that
is manifested in increases in efficiency that reduce The basic consideration motivating the analysis is that
the elasticity of new sales to employment (Cirera and technology can be biased in favour of certain groups
Sabetti, 2016). of workers depending on their skills or on the tasks
they perform. In particular, technology is skill-biased

C. I MPACT OF TECHNOLOGY
In developing countries, and especially in LDCs, most if it tends to complement skilled workers, increasing

OUTCOMES
ON LABOUR MARKET
technological change occurs because of technology their productivity when using technology at work,
transfer. Trade and foreign direct investment (FDI) and therefore increasing the relative demand for
are important vectors of technological upgrading their labour services for given wages, with little or no
because developing countries can import technology direct effect on unskilled workers. Typical examples of
embodied in capital goods, in particular machinery skill-biased technical change (SBTC) are information
(Vivarelli, 2014). The qualitative studies surveyed by technologies, which are used more intensively by
Ugur and Mitra (2017) suggest that the employment skilled workers than by unskilled workers. 22
effects are more likely to be small or negative when
technology adoption is dependent on imported Workers of all skill levels perform a variety of tasks
technology. Vivarelli (2014) also suggests that at work. These tasks can be classified along two
technology transfers can reduce the domestic main dimensions: i) their degree of routinization;
demand for labour in developing countries if they ii) whether they are manual or cognitive in nature.
involve labour-saving process innovation. An example of a routine manual task is driving an
underground train in a city. An example of a routine
While the existing empirical literature does not non-manual task is computing the average of a set of
offer conclusive evidence on the overall labour variables. An example of non-routine manual task is
demand effects of technology transfer, there is babysitting. An example of a non-routine non-manual
significant evidence that imports of capital-intensity task is organizing a wedding. 23 The substitution
technologies in developing countries are skill-biased effects of labour-saving technologies discussed in
(see the next subsection). Since technology transfer Section C.2 mainly concern routine tasks. Therefore,
mainly occurs via trade, this is a case in which it is technological change is routine-biased, in the sense
virtually impossible to distinguish between the effects that it decreases the demand for routine tasks (so
of technology and trade in determining labour market called “routine biased technical change”, RBTC).
outcomes.
The rest of this subsection presents the theoretical
To sum up the results of this subsection, the empirical mechanisms behind SBTC and RBTC, and discusses
literature has overwhelmingly found small and their empirical evidence.
possibly even positive effects of technological change
on aggregate labour demand and employment. (a) Skill-biased technical change
There are, however, a few relevant exceptions, with
some studies showing the negative labour demand In a set of developed and developing countries, one
effects of technological change. A common theme the most important labour market developments in
in the literature is that in developed and developing the 1980s and the 1990s was an increase in the
countries alike, the most relevant effects are skill premium. 24 Autor et al. (2008) show that, in
found in the composition, rather than in the level of the United States, the skill premium, while declining
83
 WORLD TRADE REPORT 2017

during the 1970s, rapidly increased in the 1980s 1990s in the United States. Acemoglu (1998; 2002)
and (less rapidly) in the 1990s. The increase in skill introduced the idea that the development and use of
premium since the 1980s also occurred in many other new technology may be directed or endogenous. An
high-income countries, such as Australia, Canada, increase in the relative supply of skilled workers will
Germany and Japan, although at substantially slower make the development and adoption of technologies
rates than in the United States (Pavcnik, 2011). that complement skilled workers more profitable.
Goldberg and Pavcnik (2007) show that increases in In other words, technology will become more skill-
skill premium since the 1980s were not confined to biased following an exogenous increase in the supply
developed countries. They also occurred, at different of high-skilled workers.
paces, in Argentina, Brazil, Chile, Colombia, Hong
Kong (China), India and Mexico during the 1980s and According to this framework, technical change has
1990s. In economies such as the United States, in been skill-biased in the twentieth century mainly
which the increase in the skill premium in the 1980s due to the development of skill-complementary
and 1990s occurred at the same time as an increase technologies in response to the rapid increase in
in the relative supply of college-educated workers, the supply of skilled workers. In contrast, the early
concurrent with the increase in the supply of skills, nineteenth century was mostly characterized by
there has been an increase in the (relative) demand skill-replacing technological development because
for skills (Acemoglu and Autor, 2011). the increased supply of low-skilled workers in the
cities made the introduction of technologies that
The rapid diffusion of ICTs in the work place is complement unskilled labour profitable. Hence, the
consistent with an increase in the (relative) demand accelerating skill-biased technical change in the
for skills because of complementarity between ICTs 1980s in several developed economies is also likely
and skills. Violante (2008) discusses three alternative due to the rapid increase in the supply of skilled
formulations of the ICT-skill complementarity workers in the late 1960s and early 1970s.
hypothesis.
(i) Empirical evidence
First, a decline in the constant quality relative price of
equipment investment, in particular in ICTs, leads to There seems to be a consensus that technological
an increased use of equipment capital in production. change has been skill-biased over the past few
Since skilled labour is relatively more complementary decades. 26 In their analysis of 450 US manufacturing
to equipment capital than is unskilled labour, growth sectors during the 1980s, Berman et al. (1994) find
in the equipment stock increases the relative demand a positive relationship between employment shift
for skilled labour and, in turn, the skill premium. 25 toward skilled workers and investment in computers
and research and development (R&D). Autor et al.
Second, skilled workers are less adversely affected (1998) extend the study to include non-manufacturing
by the turmoil created by major technological sectors. They also find that between 1979 and
transformations, since it is less costly for them to 1993, skill upgrading was larger in US industries
acquire the additional knowledge needed to adopt with greater computer utilization, with a consequent
a new technology. Therefore, rapid technological steady increase in the skill premium.
transitions are skill-biased, as more able workers
adapt better to change. Firm- or plant-level studies confirm these findings.
In a specific plant-level study of the US valve-
Third, ICTs induce an organizational shift which is manufacturing industry, Bartel et al. (2007) show
skill-biased, because it leads to flatter hierarchical that between 1997 and 2002, the adoption of new
structures where workers perform a wide range of computer-based IT equipment increased demand for
tasks within teams. Adaptable workers who have more skilled workers, particularly those with technical
general skills and who are more versed at multi- skills. Bresnahan et al. (2002) also provide firm-
tasking activities benefit from this transformation. level evidence that information technology, together
with IT-enabled workplace organizational change, is
Throughout history, technological change has not key to skill-biased technological change in the US
always been biased towards skilled workers. Goldin manufacturing and services industries.
and Katz (1998) provide evidence that manufacturing
technologies were skill-complementary in the Empirical studies for other OECD countries also
early twentieth century, but may have been skill- produce results that are consistent with the SBTC
substituting prior to that time. Autor et al. (1998) hypothesis. For instance, Falk and Seim (1999)
suggest that there was an acceleration in the skill investigate the link between skill intensity and IT
bias of technological change in the 1980s and in the service sector of Germany over the period
84
 TRADE , TECHNOLOGY AND JOBS

1994 to 1996. They show that firms with a higher IT level heterogeneity and find evidence of a robust
investment-output ratio employ a larger fraction of association between skilled labour demand and
high-skilled workers. In another study, Falk (2001) technology adoption as measured by patent usage
also shows that ICT penetration in German firms is and other technology indicators.
positively related to the employment share of skilled
workers and negatively related to the share of both In the case of developing countries, the adoption
medium- and unskilled-workers. A study by Spitz- of new technologies occurs mainly through import
Oener (2006) further confirms that, in Germany, flows and FDI inflows, which generate technological
skill demand has increased greatly in occupations spillovers. In a sample of 28 manufacturing sectors
adopting technology more intensively. Skill-biased for 23 LDCs and middle-income countries over the
effects of technological change have been assessed period 1980–91, Conte and Vivarelli (2011) discuss
in Canada (Gera et al., 2001), France (Greenan et al., the occurrence of skill-enhancing technology imports,
2001), Italy (Piva et al., 2005), Spain (Aguirregabiria namely, the relationship between imports of embodied
and Alonso-Borrego, 2001) and the United technology and widening skill-based employment
Kingdom (Machin, 1995; Gregory et al., 2001). differentials. They show evidence of capital–skill
These studies document a positive relationship complementarity as a possible source of skill bias, and
between employment of skilled labour and various of imported skill-enhancing technology as an additional
measures of technological innovation, such as the driver of increasing demand for the skilled workers.
use of computers, R&D intensity, and the number of
innovations and patents. A number of country-level studies provide similar
evidence of relative skill bias emerging through

C. I MPACT OF TECHNOLOGY
Cross-country studies also confirm the empirical embodied technological change in several developing

OUTCOMES
ON LABOUR MARKET
validity of the SBTC hypothesis in advanced countries, such as Brazil (Fajnzylber and Fernandes,
economies. Machin and Van Reenen (1998) show 2009), Costa Rica (Robbins and Gindling, 1999),
that for seven OECD countries (Denmark, France, Ghana (Görg and Strobl, 2002), Mexico (Hanson
Germany, Japan, Sweden, the United Kingdom and Harrison, 1999; Meza, 1999; Fajnzylber and
and the United States), the relative demand for Fernandes, 2009), Turkey (Srour et al., 2013; Meschi
skilled workers in the manufacturing sector was et al., 2016) and the Middle East and North Africa
positively linked to R&D expenditure between 1973 region (Mnif, 2016).
and 1989. Berman et al. (1998) provide evidence
that, for 12 developed countries in the 1980s, three Evidence that is common across both industrialized
manufacturing industries – machinery and computers, and developing countries is often considered as
electrical machinery, and printing and publishing – suggestive of common technological change across
where skill-biased technological changes are most the world. Berman et al. (1998) find that in the 1970s
pervasive, together accounted for 40 per cent of the and 1980s, across industrialized countries, most
within-industry increase in the relative demand for industries increased the proportion of the high-
skills in manufacturing sector. skilled (non-production) wage bill to the low-skilled
(production) wage bill, despite rising or stable relative
It is argued above that ICTs induce organizational wages for high-skilled workers. Berman and Machin
changes – such as the flattening of hierarchies, (2000) document that relative wage bills of high-
the decentralization of authority, and increased skilled workers jointly increased in the manufacturing
multitasking – which are skill-biased. Caroli and Van industries of 37 high-, middle-, and low-income
Reenen (2001) provide evidence for a panel of British countries during the 1980s. They find that industry-
and French plants that organizational change and level skill upgrading in all countries was positively
skills are complementary, as they reduce the demand correlated with US computer usage and OECD R&D
for unskilled workers and lead to greater productivity intensity. In summary, changes in skill intensity were
increases at plants with larger initial skill intensities. similar and widespread across countries at different
income levels, and they were closely related to
Evidence of skill-biased technological change exists technology usage in industrialized countries. This is
also for developing countries. Using plant-level data consistent with simultaneous global SBTC.
for Chile, Pavcnik (2003) finds evidence of capital-
skill complementarity. This might contribute to the (ii) Quantification of the effects of skill-
increased within-industry relative demand for skilled biased technical change
workers during the 1980s, although there might
not be a causal relationship. Fuentes and Gilchrist Some studies quantify the contribution of technology
(2005) extend the analysis over an additional nine to observed changes in relative employment or
years (1979-95) to control for unobserved plant- relative wages of skilled versus unskilled workers.
85
 WORLD TRADE REPORT 2017

These quantification exercises, however, are subject for cognitive tasks at the technology’s maturity
to the important caveats that the results are highly still exceeds demand before the introduction of the
dependent on the definition of technology and vary general-purpose technology, but it is no longer as
significantly across different studies. high as during the initial adoption phase.

For the United States, Japan and nine European The link between de-skilling and automation is even
countries, from 1980 to 2004, Michaels et al. (2014) more general. One of the most salient characteristics
argue that ICTs can account for up to a quarter of of automation is the breaking down of complex
the cross-country variation in demand growth for operations into simple tasks. That was what Ford
highly skilled workers. In their study of relative wage did in the first car factories at the beginning of the
and employment outcomes among US workers twentieth century. Instead of employing skilled
in the 1980s, Feenstra and Hanson (1999) show artisanal workers to build cars, as Daimler Benz did
that the share of the wage differential paid to non- in Germany, Ford hired unskilled workers, which were
production workers during the 1980s attributable in abundant supply due to large inflows to the United
to technology is equal to around 30 per cent when States of immigrants from overseas, to perform
using high-tech equipment measured with an ex ante simple tasks. In analysing the impact of technology on
rental price. When the authors alter their measure of the future of work, Section C.4 will discuss artificial
high-tech equipment to place more weight on more intelligence (AI). By first recognising patterns in
recently installed (and thus arguably more advanced) information before the human eye and analytics can,
equipment, the contribution of technology to wage and then breaking complex cognitive tasks into simple
outcomes increases substantially, more than tripling. tasks that require little or no skills, AI makes the link
Using US data for the period 1984 to 2003 in a between automation and de-skilling clear
rich structural model, Burstein et al. (2015) find that
computerization is the central force driving changes (b) Routine-biased technical change
in the skill premium, accounting for 60 per cent of its
rise. 27 While fully consistent with labour market
developments in the 1970s and in the 1980s, models
(iii) Can technological maturity lead to of SBTC are less successful in explaining more recent
“de-skilling”? developments. The evolution of the skill premium has
been very heterogeneous across countries since
It has recently been documented that, in about the the mid-1990s. As detailed in Section B.2, while in
year 2000, the demand for skills (more specifically, some countries there has been a reduction in the skill
the demand for cognitive tasks that are often premium over the last 15 years, in other countries the
associated with high educational skills) underwent a opposite occurred. An important recent labour market
reversal in the United States (Beaudry et al., 2016; development in many developed countries, such as
Charles et al., 2016). According to Beaudry et al. Germany and the United States, during the last two or
(2016), in response to this demand reversal, high- three decades has been the hollowing out of middling
skilled workers moved down the occupational ladder occupations (employment polarization), as discussed
and began to perform jobs traditionally performed in Section B.2. Several developing countries have also
by lower-skilled workers. This de-skilling process, experienced such polarization in the last two decades
in turn, resulted in high-skilled workers pushing low- (World Bank, 2016; de Vries, 2017).
skilled workers even further down the occupational
ladder and, to some degree, out of the labour force Drawing on the seminal contribution by Simon (1960),
altogether. Autor et al. (2003) present a theoretical framework
linking employment polarization to technology. In this
Charles et al. (2016) link these developments to framework, technology affects specific tasks, rather
technology. They argue that during the initial adoption than specific skills. 28 Autor et al. (2003) classify work
phase of a general-purpose technology such as an tasks along two main dimensions: i) their degree of
ICT, demand for cognitive tasks grows fast because routinization; ii) whether they are manual or cognitive
the associated machinery and equipment need to in nature. Technological progress tends to replace
be built and installed. However, once the general- routine tasks and to complement cognitive skills. This
purpose technology has been widely adopted, is represented in Table C.1. The ease of automation
demand for cognitive tasks partially drops because is determined by the routine versus non-routine
at the maturity of the technology, those activities nature of work tasks. The skill complementarity is
are still needed for the maintenance and occasional determined by the cognitive versus manual nature
replacement of the technology, but are no longer of work tasks. Technology is predicted to improve
necessary for its adoption. In absolute levels, demand relative employment prospects for workers in the
86
 TRADE , TECHNOLOGY AND JOBS

Table C.1: How technology and skills at work interact

Skill complementarity

High Low
automation
Ease of

High Routine cognitive Routine manual

Low Non-routine cognitive Non-routine manual

Source: World Bank (2016).

Table C.2: Expected effects of technology on employment and earnings by types of occupation
Type of occupation Expected impact Expected impact
(by skill intensity) on employment on earnings

Non-routine cognitive Positive Positive

Routine cognitive
Negative Negative
and manual

Non-routine manual Positive Negative

C. I MPACT OF TECHNOLOGY
OUTCOMES
ON LABOUR MARKET
Source: World Bank (2016).

bottom left quadrant, because they perform non- extent to which workers in the lower quadrants of
routine tasks (which are not easily automated)29 and Table C.1 (those performing tasks less susceptible
involving cognitive skills, where ICT technologies to automation) see their relative earnings increase or
make them more productive. Workers in the upper decrease depends on the elasticity of labour supply
quadrants perform tasks subject to automation, and (Autor, 2015). 30 If the elasticity of labour supply is
their relative employment prospects are reduced high enough, new entry of labour can partially or fully
by labour-saving automation technologies. Finally, offset average wage gains that would have occurred.
workers in the bottom right quadrant are the least As argued by the World Bank (2016), workers in non-
affected by technology, because they perform routine cognitive occupations are likely to see their
non-routine manual tasks, which are neither easily higher productivity rewarded as higher earnings
automated nor subject to ICT-skill complementarity. because entry barriers are high (therefore the
elasticity of labour supply is low). Conversely, low-
The framework proposed in Table C.1 helps to skilled workers in non-routine manual occupations
explain why technology can lead to employment are likely to see their earnings fall over time, as
polarization (Autor et al., 2003; 2006; 2008). middle-skilled workers in routine occupations are
Non-routine cognitive tasks which – given current displaced by automation and start competing for the
technological feasibilities – are difficult to automate available jobs in low-paying occupations, where entry
and are complementary to ICTs, are typical of skilled costs are low and the elasticity of the labour supply is
professional and managerial jobs, which tend to be high. 31 These insights are summarized in Table C.2.
assigned to skilled workers. The non-routine manual
tasks which are not directly affected by technology (i) Empirical evidence
are typically unskilled jobs such as housecleaning
and tend to be assigned to unskilled workers; routine Recent shifts in the nature of work include a strong
cognitive or manual tasks, in which technology has decline of occupations that are intensive in routine
the potential to substitute for human labour, are work steps. For the United States, Cortes et al. (2016)
typical of jobs performed by middle-skilled workers. document a decrease in routine employment from 40
per cent of the population aged 20-64 in 1979 to 31
The discussion so far has focused on relative demand per cent of the population aged 20-64 in 2014 (see
for workers depending on their skills, or on the Figure C.4). 32 At the same time, non-routine manual
nature of their work tasks. How changes in relative employment expanded by 3.9 percentage points and
demand translate into changes in relative earnings non-routine cognitive employment expanded by 6.7
crucially depends on labour supply. In particular, the percentage points (Cortes et al., 2016, Table 2).
87
 WORLD TRADE REPORT 2017

Figure C.4: Evolution of employment shares of routine occupations in the United States
(1979 to 2014)

45
Employment share of routine occupations (percentage)

40

35

30

25

20

15

10

0
1979 1989 1999 2009 2014

Routine cognitive Routine manual

Source: Cortes et al. (2016).

Notes: Employment shares based on individuals aged 20-64 from the monthly Current Population Survey (US Bureau of Labor Statistics),
excluding those employed in agriculture and resource occupations.

With few exceptions, the empirical literature confirms countries over the period 1993 to 2010. Focusing on
the idea that technological change in developed automation in the form of industrial robots, however,
economies is a major driver in the decline in routine Graetz and Michaels (2015) do not find that this
occupations, and in the consequent employment type of technology is biased against middle-skilled
polarization. Conversely, for developing economies, workers. On the contrary, they find that robot density
there is limited empirical evidence consistent with shifts demand from the low-skilled towards the high-
the RBTC hypothesis. As already emphasized above skilled. This result could depend on the definition of
when discussing the overall employment effects of skills, or on the different ways in which routine jobs
technology, comparisons across studies are only are affected by general purpose technologies, like
meaningful for studies that use the same definition of ICTs, as opposed to industrial automation.
technological change.
In the case of developing countries, there is limited
At the level of local labour markets in the United empirical evidence consistent with RBTC. Job
States, Autor and Dorn (2013) show that between polarization in the labour markets of Colombia and
1980 and 2005, local labour markets that specialized Mexico in the 2000s was shown to occur due to
in routine tasks differentially adopted information reductions in the cost and increased adoption of
technology, reallocated low-skill labour into service computer technology (Medina and Posso, 2010).
occupations (employment polarization), experienced The same, however, does not apply to several other
earnings growth at the tails of the distribution (wage developing countries (Brazil, China, India and the
polarization), and received inflows of skilled labour. Russian Federation), nor to LDCs (Medina and
Similarly, Autor et al. (2015) show that local US labour Posso, 2010; Gimpelson and Kapeliushnikov, 2016;
markets more specialized in routine occupations Maloney and Molina, 2016).
have experienced employment losses in routine task-
intensive occupations. These losses are, however, (ii) Quantification of the effects of routine-
largely offset by local employment growth in abstract biased technical change
and manual task-intensive occupations. 33
Quantification of the contribution of technology to
At the industry level, Goos et al. (2014) estimate the decline in middle-skilled employment can be
that routine-biased technological change is mostly found in Goos et al. (2014) for 16 Western European
responsible for observed patterns of employment countries over the period 1993-2010, and in Cortes
polarization in a sample of 16 Western European et al. (2016) for the United States over the period
88
 TRADE , TECHNOLOGY AND JOBS

1979-2014 (or just 1989 to 2014 in an alternative side, increasing educational attainment and population
specification). Goos et al. (2014) estimate a model ageing in the United States have reduced the fraction
where the routine task intensity index (the standard of workers with these demographic attributes.
proxy for RBTC used in the literature) explains most of
the observed employment polarization. In particular, However, labour supply alone cannot account for
for the group of the eight highest-paid occupations, the labour market experience of these demographic
the model predicts an increase in employment shares groups. Within each group, the propensity to work in
(hours worked as a share of total hours) of 79 per routine occupations has decreased dramatically. For
cent of the increase actually observed (with the instance, while more than 60 per cent of low-educated
estimated increase equal to 4.45 and the increase young men worked in routine manual occupations
actually observed equal to 5.62). For the group of nine in 1979, this figure dropped to one-third in 2014
middling occupations, the model predicts a decrease (Cortes et al., 2016). The decline in the probability of
of 74 per cent of the total observed employment routine employment (equal to 8.1 percentage points
share decrease (with the estimated decrease equal between 1979 and 2014 for manual employment, and
to 6.86 per cent and the increase actually observed to 1.2 percentage points for cognitive employment in
equal to 9.27). Lastly, the model predicts an increase the same period, as shown in Figure C.4) was offset
for the group of the four lowest-paid occupations by increases in non-employment and in non-routine
of 66 per cent of the observed increase (with the manual employment.
estimated increase equal to 2.41 and the increase
actually observed equal to 3.65). The results of Cortes et al. (2016) suggest that, on
average, it has been very difficult for US workers

C. I MPACT OF TECHNOLOGY
Cortes et al. (2016) calibrate a structural model of the employed in routine occupations to find employment

OUTCOMES
ON LABOUR MARKET
US economy which matches observed occupational in high-paying non-routine cognitive occupations.
reallocations. They find that automation shocks Cortes (2016) shows that the outcomes vary across
(measured as the deviation of ICT capital from a workers, depending on their abilities. In particular, he
balanced growth trend) explain at most one-third of provides theoretical and empirical evidence showing
the decline in middle-skilled employment. It should be that low-ability routine workers are more likely to
noted, however, that these figures do not provide a switch to non-routine manual jobs, while high-ability
quantification of the contribution of technology to the routine workers are more likely to switch to non-
decline of manufacturing employment, because not routine cognitive jobs.
all manufacturing employment is middle-skilled, and
not all middle-skilled employment is in manufacturing. (iv) Can RBTC explain “jobless
recoveries”?
(iii) The nature of adjustment to RBTC
Routine-biased technical change has also been linked
Recent empirical work on labour market adjustments to so-called “jobless recoveries” (periods following
related to RBTC is available for the United States. recessions in which rebounds in aggregate output are
Cortes et al. (2014) show that during the 30 years accompanied by much slower recoveries in aggregate
preceding their study, the decline in middle-skilled employment). In particular, the argument has been
jobs in the United States was driven mainly by the made that routine-biased automation might be
paucity of transitions from out of the labour force and responsible not only for job losses during downturns,
from unemployment into routine employment, rather but also for sluggish employment growth during
than by job losses. In other words, it was very difficult economic recoveries. In this connection, Brynjolfsson
to find employment in routine jobs. and McAfee (2011; 2014) refer to a “great uncoupling”,
in which economic growth has become detached from
Further insights on low entry rates into routine jobs employment growth for the first time in the modern era.
are provided by Cortes et al. (2016). They document
that the decline of middle-skilled occupations in the There is, however, no consensus on this issue, since
United States between 1979 and 2014 was primarily the empirical evidence is mixed. In the case of the
driven by the disappearance of routine jobs among United States, the disappearance of employment
workers in specific demographic groups: male high in routine occupations has been concentrated in
school dropouts of all ages and male high school economic downturns (Jaimovich and Siu, 2014).
graduates under the age of 50 in the case of manual Specifically, 88 per cent of job losses in routine
employment, and young (20-29) and prime-aged occupations since the mid-1980s have occurred
(30-49) females with either high school diplomas or within a 12-month window of recessions (all of which
with some degree of post-secondary education in the have been characterized by jobless recoveries). The
case of cognitive employment. On the labour supply displaced workers have then been forced into time-
89
 WORLD TRADE REPORT 2017

consuming transitions to different occupations and the meaning and enjoyment of working attributed
sectors, resulting in slow job growth during the by workers and the availability and attractiveness of
recovery. Jaimovich and Siu (2014) and Graetz and alternatives to working. Similarly, the future of labour
Michaels (2017) provide empirical evidence of a link demand depends, among other things, on the relative
between the hollowing out of middle-skill routine cost of investment goods and financing conditions,
occupations and jobless recoveries for the United product demand, and the existence as well as the
States. affordability of specific technologies.

However, Graetz and Michaels (2017), using data The ongoing and upcoming wave of advances in
on 71 recessions which took place in 17 developed smart technology, artificial intelligence, robotics and
countries other than the United States from 1970 algorithms, often referred to as the fourth industrial
to 2011, do not find evidence that industries that revolution, is gathering increasing expert and media
make more intensive use of routine jobs, and that are attention. In this context, a debate has emerged about
therefore more susceptible to technological change, the impact that these emerging technologies will have
have had particularly slow employment growth during on the future of jobs. Some experts argue that history
periods of economic recovery. The same result holds will repeat itself and the transformative processes
for industries in which labour was more exposed to brought by the next wave of technological advances
automation by industrial robots. They also find that will replace many existing jobs but eventually create
middle-skill employment grew similarly in routine- new jobs and opportunities. Other experts argue
intensive industries and other industries during recent that the impact of the new wave of labour-saving
recoveries. Graetz and Michaels (2017) therefore technological change and innovations on jobs will
conclude that technology is not causing jobless be different this time around, with the replacement
recoveries in developed countries outside the United of human jobs by artificial intelligence and robotics
States. on a massive scale, leading to a “jobless future”. This
subsection will review the main arguments put forward
Summing up the results of Section C.2, it can be by both sides regarding the impact of technology on
argued that technological change impacts workers future jobs, and discuss the implications for skills
differently, depending, among other conditions, on development.
their skills and on the work tasks they perform. Current
technological change tends to be skill-biased, in the (a) Moving with or against technological
sense that it increases the relative demand for skills, advance?
and routine-biased, in the sense that it decreases
demand for routine tasks. Therefore, relatively skilled The view that the new technological advances in
workers performing non-routine tasks tend to benefit artificial intelligence and robotics will not lead to a
from technological change, which can be disruptive “jobless future” is based on historical experience.
for relatively unskilled workers employed in routine Although each wave of technological change has
tasks. generated technological anxiety and led to temporary
disruptions with the disappearance of some tasks
The next subsection will provide insights into whether and jobs, other jobs have been modified, and new
these conclusions might still apply in the near future, and often better jobs have eventually been developed
or whether, with the upcoming wave of advances in and filled through three interrelated mechanisms
smart technology, artificial intelligence, robotics and (Autor and Handel, 2013; Autor, 2015; Bessen, 2015;
algorithms, technological disruption might affect Mokyr et al., 2015).
ever-increasing numbers of workers at all levels of
skills and work tasks. First, new technological innovations still require
a workforce to produce and provide the goods,
services and equipment necessary to implement
4. Technology and the future the new technologies. Recent empirical evidence
of work suggests that employment growth in the United
States between 1980 and 2007 was significantly
As discussed in Section B, the level and structure greater in occupations encompassing more new job
of employment depends on the supply and demand titles (Berger and Frey, 2017).
of labour. The future of jobs is no exception and
hinges on the future of both the labour supply and Second, the new wave of technologies may
demand. The future of labour supply depends, among enhance the competitiveness of firms adopting
other things, on demographic developments, the these technologies by increasing their productivity.
future level and distribution of wealth, as well as These firms may experience a higher demand for
90
 TRADE , TECHNOLOGY AND JOBS

the goods or services they produce, which could pace, providing individuals with more time to adjust
imply an increase in their labour demand. 34 Several (Comin and Hobijn, 2010). For instance, the United
empirical studies reviewed in Section C.2 above find States took 30 years to achieve a 10 per cent
that the adoption of labour-saving technologies did adoption rate of electricity, while it took less than five
not reduce the overall labour demand in European years for tablet devices to reach the same level of
countries and other developed economies (Goos et adoption rate (DeGusta, 2012).
al., 2014; Graetz and Michaels, 2015; Bessen, 2016;
Gregory et al., 2016). In comparison with previous innovations, the new
technological advances are evolving at an exponential
Finally, as discussed in Section C.2, the upcoming pace. Although some experts argue that Moore’s
technological advances may complement some Law, according to which the number of components
tasks or occupations and therefore increase labour in a specific size integrated circuit has doubled every
productivity, which could lead to either higher 18 months since 1965, is approaching its end, it has
employment or higher wages, or both. The new enabled greater computing power and the ability to
workers and/or those benefitting from a pay rise may automate increasingly complex tasks (Brynjolfsson
increase their consumption spending, which in turn and McAfee, 2014; Waldrop, 2016). Graetz and
tends to maintain or raise the demand for labour in the Michaels (2015) report that from 1993 to 2007, mean
economy. Recent empirical evidence suggests that robot density increased by more than 150 per cent
the use of industrial robots at the sector level has led in Australia, 14 European countries, the Republic
to an increase in both labour productivity and wages of Korea and the United States. Boston Consulting
for workers in Australia, 14 European countries, the Group (2017) reports that there are currently

C. I MPACT OF TECHNOLOGY
Republic of Korea and the United States (Graetz and between 1.5 and 1.75 million industrial robots in

OUTCOMES
ON LABOUR MARKET
Michaels, 2015). operation, a number that could increase to between
4 and 6 million by 2025. The pace of progress in
Conversely, the proponents of a future rise in certain areas, such as biotechnology, has even
unemployment due to technological advances exceeded Moore’s Law (Autor, 2015). According
recognize that the fear of such unemployment has to the World Economic Forum (2016), around 65
been proven wrong many times in the past, but per cent of pre-school children will be expected to
consider that the new wave of technological progress undertake tasks and jobs that do not currently exist.
represents a sharp departure from earlier innovations. The speed of technological acceleration could imply
Advances in robotics, artificial intelligence, self-driving that individuals, even those who are flexible and well-
vehicles, “big data” (i.e. data sets too extensive for adjusted to the labour market, may need to retrain
traditional data processing software to process) and and update their skill sets so as to keep up with the
3-D printing are likely to continue to erode lower- occupational rearrangements and new additional
skilled employment specialized in routine tasks, but skills that will be required.
also to impact on medium- and high-skilled jobs
involving physical, cognitive and non-routine tasks Secondly, most previously ground-breaking
requiring knowledge, judgment and experience, which technological innovation, such as light bulbs
were once thought to be exclusively human domains and telephones, did not necessarily occur in
(Brynjolfsson and McAfee, 2014; Ford, 2015). every industry of the economy at the same time,
which allowed affected individuals to look for job
In particular, they argue that the upcoming opportunities in until then undisrupted industries.
technological advances in digitalization and For instance, during the agricultural revolution in
algorithms empowered by big data will continue the 1700s, many individuals who lost their jobs in
to reduce the marginal costs of (re)production the countryside moved to cities in search of work.
to a near zero level, making human workers more Previous technological revolutions often took a
expensive than the additional costs of using the new long time to exhibit significant impacts on the entire
technologies (Rifkin, 2015). This would ultimately economy. While investments in railroads generated
result in a shrinkage of the total number of available initially relatively limited benefits and spillovers, the
human jobs in the medium to long run. 35 The impact latter have gradually expanded thanks to improvement
of the new and upcoming wave of technologies on in railroad productivity and an increase in the share of
future job losses is, in their view, also distinct from rail output in economic activity.
previous ones in terms of speed, scale, and force
(Schwab, 2016). Similarly, the speed of adoption of ICT has been
different across sectors. Some sectors, such as
Firstly, empirical evidence suggests that previous manufacturing, agriculture, forestry and fishing, hotels
technological advances were adopted at a slower and restaurants, and the wholesale and retail trades,
91
 WORLD TRADE REPORT 2017

have experienced a very rapid increase in the use of technologies will also unfold in a different setting in
ICT capital services per hour worked, while other terms of demographic and life expectancy compared
sectors, such as construction and transport, have to previous technological revolutions (Clark, 2017).
recorded a lower ICT intensity growth rate (OECD, Previous significant innovation occurred in a world
2017). Some experts argue that technological characterized by its growing population. Nowadays,
progress in ICT has been less transformative than an increasing number of developed and emerging
any of the three main technologies that emerged countries face an ageing and shrinking working-age
during the second Industrial Revolution at the population with the potential additional pressure to
end of the nineteenth century and beginning of use non-human labour to compensate for the fewer
the twentieth century (namely electricity, cars and working-age workers that used to finance the social
wireless communications) (Gordon, 2014). However, safety net. Conversely, most developing countries are
a recent study comparing US labour productivity in still experiencing a growing population and face the
the electrification era (1890 to 1940) and the ICT challenge of creating conditions in which to provide
era (1970 to 2010) finds that productivity growth in new jobs in addition to existing jobs unaffected
both eras exhibited remarkably common patterns: an by the new technological innovation. Similarly, the
initial relatively slow growth in productivity, followed improvement of the average life expectancy, thanks to
by various decade-long accelerations and then a scientific and technological innovations in health and
slowdown in productivity growth (Syverson, 2013). medicine, implies that individuals will, on average,
Unlike previous important technological innovations, be able to work longer, potentially putting additional
the new and upcoming ones are not limited to one pressure on the labour market.
specific area but combine various elements, such
as energy storage, quantum computing, mobile (b) Prospects of automation
networks, biotechnology, nanotechnology and
material science, potentially affecting all areas of the One of the studies that reignited the debate about the
economy at once, including the services sector as new wave of technologies, in particular automation,
well as the agricultural and manufacturing sectors. and employment was a 2013 research paper by Frey
and Osborne (2017), who classify 702  occupations
Finally, the breadth and depth of these new complex in the United  States in terms of skills that are likely
technologies have the potential to transform entire to be automated. The authors conclude that 47 per
systems of production, management, and governance. cent of these occupations are at risk of automation
For instance, the phenomenon of digitalization has and computerization over the next two decades. In
already led to the emergence of new business and particular, the study identifies retail salespeople,
employment models, often referred to as the “platform administrative assistants, food counter personnel,
economy”, “sharing economy”, “peer-to-peer cashiers, and transport truck drivers as working at
economy”, “gig economy” or “on-demand economy”. occupations that are at a greater risk of automation. 37
In particular, the establishment and development A number of consultancy firms and academics
of new digital transportation, accommodation, and have replicated the analysis for various European
on-demand and freelance labour platforms have countries, Australia, Canada, Japan and New
enabled the creation of new types of jobs as well as Zealand, and report that the share of jobs susceptible
temporary and flexible contracting arrangements. 36 to automation ranges from 30 to 49 per cent (Baert
Some experts also anticipate a surge in superstar- and Ledent, 2015; Deloitte, 2015b; Bouée and El
biased or talent-biased technological change Karoui, 2014; Sproul et al., 2015; Pajarinen and
associated with many digital technologies, where a Rouvinen, 2014; Brzeski and Burk, 2015; Citibank,
limited number of firms and individuals capture most 2016; David, 2017; Durrant-Whyte et al., 2015). 38
of the market share and financial benefits stemming
from the adoption and diffusion of these technologies The risk of automation is not only confined to
(Brynjolfsson and McAfee, 2014). The emergence developed economies. As highlighted in Figure C.5,
of such winner-take-most or -take-all markets could estimates of the share of occupations at risk from
have consequences on the degree of competition of automation actually tend to be higher for developing
different sectors of the economy and on perceptions and least-developed countries than for high-income
of equity and fairness of the consequences of countries. According to the World Bank’s World
technical change. Development Report 2016, two-thirds of all jobs
could, on average, be vulnerable to automation
Besides the speed at which the new and upcoming in developing countries in the next decades. The
waves of technologies could change the systems of estimated share of jobs at high risk of automation
production, distribution and consumption in almost ranges from 55 per cent in Uzbekistan, 65 per cent in
every industry and economy, these new complex Nigeria and 67 per cent in Bolivia and South Africa,
92
 TRADE , TECHNOLOGY AND JOBS

Figure C.5: Proportion of jobs at risk of automation by economic development

90

80
Share of jobs at risk of automation based on
occupation-level analysis (percentage)

70

60

50

40

C. I MPACT OF TECHNOLOGY
OUTCOMES
ON LABOUR MARKET
30
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5
GDP per capita (purchasing power parity US$, natural log scale)

Unadjusted to technological feasibility Adjusted to technological feasibility

Source: WTO Secretariat based on occupation-level analyses published by Baert and Ledent (2015), Bouée and El Karoui (2014),
Brookfield Institute for Innovation + Entrepreneurship (2016), Bruegel (2014), Brzeski and Burk (2015), Centre for Strategic Futures
(2015), Chang and Huynh (2016), Chartered Accountants Australia and New Zealand (2016), David (2017), Deloitte (2015a), Deloitte
(2015b), Durrant-Whyte et al. (2015), Frey and Osborne (2017), Pajarinen and Rouvinen (2014), Pajarinen et al. (2015), Stiftelsen för
Strategisk Forskning (2014), Unionen (2014) and World Bank (2016).

to 74 per cent in Angola, 77 per cent in Bangladesh technological feasibility in low- and middle-income
and 85 per cent in Ethiopia. The estimated share countries range from 34 to 65 per cent, which is
of jobs at risk of automation is also substantial in relatively similar to the estimates for many high-
emerging economies, such as Argentina (65 per income countries.
cent), India (69 per cent) and China (77 per cent).
A recent International Labour Organization (ILO) According to more recent studies, the relatively high
study estimates also that about three in five jobs face share of jobs vulnerable to automation estimated and
a high risk of automation in Cambodia, Indonesia, reported in the above-mentioned studies stems from
the Philippines, Thailand and Viet Nam (Chang and the failure to account for the fact that occupations
Huynh, 2016). tend to adjust to technology by adapting their task
structure. In fact, most occupations adapt regularly
According to the World Bank, the high share of jobs to technological innovation by reallocating routine
susceptible to automation could impact negatively tasks to automation and refocusing human work on
on developing economies’ ability to develop further. management and on non-routine social, interpersonal
However, the adoption and diffusion of automation and creative tasks. This is what happened with many
could be slower and more limited in developing bank tellers following the introduction of ATMs, as
countries given the higher prevalence of barriers to discussed in the previous subsection.
technology, lower wages, and the number of jobs
based on manual dexterity (see Box C.2). Adjusted for Taking into account the difference in the ability to
the slower pace of technology adoption in developing automate specific jobs and tasks within occupations,
countries, the World Bank’s estimates of the share a recent study estimates that 12 per cent and 9 per
of jobs at risk of automation decrease significantly cent of the jobs in Germany and the United States,
for most developing countries analysed. As shown respectively, could be fully automated (Bonin et al.,
in Figure C.5, the estimated shares adjusted to 2015). Based on the same methodology, an OECD
93
 WORLD TRADE REPORT 2017

Box C.2: The future impact of automation on developing countries’ labour market

While there is a growing literature on the potential impact of automation and artificial intelligence on the
labour market in developed economies, the impact in developing countries has received much less attention.
The few occupation-based studies that estimate the share of employment at risk of automation in developing
countries conclude that the latter have a larger share of employment in routine occupations that could be
automated and computerized (World Bank, 2016; Citibank, 2016). Yet, as pointed out in World Bank (2016),
the impact of automation on the labour market of developing countries could occur later and be slower for
two main reasons. First, although the speed of technology adoption has increased in developing countries,
it remains slower than in developed countries. Second, lower wages and a relatively high share of manual
non-routine jobs, which are currently more difficult to automate, could make investment in automation in
developing countries less profitable (at least in the short run). However, regardless of the timing, automation
raises several issues for developing countries.

First, by reducing the labour content of the production process, automation in developed countries could
compete with countries in which labour costs are low (UNCTAD, 2016). Firms in high-income countries could
decide to bring specific manufacturing operations located in developing countries back in order to minimize
production costs and enhance their competitiveness. Reshoring could also apply to business process
outsourcing in financial services (e.g. accounting), telecommunications (e.g. call centres) and medical
services. In such a situation, developing countries may experience a reduction in production and employment
opportunities in certain industries (Citibank, 2016). These potential changes could be particularly challenging
for those developing countries that are already facing deindustrialization and are becoming service economies
sooner and at much lower levels of income compared to countries that were industrialized earlier (Rodrik,
2016). The continuous real wage growth in emerging countries could provide further incentives for potential
re-shoring and the adoption of automation. However, for the time being, empirical evidence suggests that
reshoring is limited, occurring in specific industries and relatively slowly (UNCTAD, 2016).

Second, the new wave of technologies could provide entrepreneurs and firms in developing countries with
the opportunity to establish new business models and offer new goods and services. For instance, additive
manufacturing (i.e. industrial 3-D printing) could, thanks to its mobility, flexibility, energy efficiency and increasing
affordability, enable small-scale manufacturing to become more competitive and efficient in developing and
least-developed economies (Naudé, 2017). However, such opportunities are likely to be challenging for
economies without reliable access to electricity and the internet, as well as to relevant skills in the workforce.

Third, automation and advances in ICT could also create new job opportunities in developing countries through
the development of online work platforms bringing together potential employers and employees (World Bank,
2016). Such online platforms could provide workers in developing countries, including young people and women,
with opportunities to monetize skills which might have limited demand in the local labour market. However,
access to and use of these platforms tend to be higher among young people and highly skilled workers, and
this, along with automation, could further contribute to the polarization of the labour market, with employment
growth at the bottom and top of the skill and income distribution. As discussed in Section B, recent empirical
evidence shows that many developing countries, except those with a large share of low-skill employment, large
natural resources and commodity endowments, are already experiencing job polarization (World Bank, 2016).

Finally, individuals with low educational attainments and low incomes are most vulnerable to technological
changes in the labour market. It is, however, unclear how the informal sector, which, as discussed in Section
B, accounts for a large share of the total workforce in many developing countries, will adjust to automation
and advances in ICT. Empirical evidence suggests that informal firms tend to innovate or adopt technologies
at lower rates than formal firms (Harris, 2014). The literature further shows that as previous technological
innovations have improved energy access, telecommunications, and transport systems, they have enabled
certain informal workers to make productivity gains by improving their work efficiency and organization, as
well as to take advantage of new work opportunities (Casey and Harvey, 2015; 2016). However, for other
informal workers who do not have the financial means to acquire the new technologies and/or are unable
to upgrade their skills to implement them, the disruptive impact of technologies is much more negative.
Several case studies report that informal workers who organize and collaborate with other groups manage
to improve considerably their capacity to upgrade and expand their technology options gradually, to keep up
with the pace of technological change, and to mitigate the setbacks from the negative impacts of technology
disruption in their sector (Casey and Harvey, 2016). Limited and uneven access to future technologies among
94 informal workers, including the most vulnerable, could therefore exacerbate the “digital divide”.
 TRADE , TECHNOLOGY AND JOBS

study reports that, on average, 9 per cent of jobs in to 50 per cent in Brazil and the Russian Federation,
21 OECD countries are susceptible to full automation, 52 per cent in Kenya and Mexico, 55 per cent in
ranging from 6 per cent or less in Estonia, Finland Thailand and 57 per cent in Japan. Given the sectoral
and the Republic of Korea to 12 per cent in Austria, structure of their economy, the activities mix within
Germany and Spain (Arntz et al., 2016b). The authors of these sectors, and their workforce size, China, India,
the study conclude that occupation-level approaches Japan and the United States account for almost two-
overestimate automation potentials, because three thirds of the number of workers whose activities could
out of four jobs in a particular occupation are, on technically be automated by currently demonstrated
average, less automatable compared to the median job technologies (McKinsey Global Institute, 2017).
of the particular occupation, suggesting that workers
specialize in non-automatable tasks within their New research also suggests that the future impact
professions (Arntz et al., 2017). of automation could vary significantly across regions
and areas within a given country (Morgan et al., 2017;
A more recent report prepared by Institute for Spatial Economic Analysis (ISEA), 2017).
PricewaterhouseCoopers (2017) modifies the While digital technologies have enabled firms to
methodology used by Artnz et al. (2016b) by applying enhance their means of communication and further
additional data and developing an algorithm linking segment their production processes, companies still
automatability to tasks’ and workers’ characteristics. tend to cluster specific skills and occupations in certain
It estimates that 35 per cent, 30 per cent and 38 per geographical locations to capitalize on the availability
cent of the jobs in Germany, the United Kingdom and of inputs, including labour force and suppliers, and
the United States, respectively, face a potentially high potential spillovers. As a result, areas with a relatively

C. I MPACT OF TECHNOLOGY
OUTCOMES
ON LABOUR MARKET
risk of automation. Two recent studies by McKinsey larger concentration of tasks and jobs vulnerable to
Global Institute (2016; 2017), based on a different automation, which tend to be small cities, could be
methodology which analyses work activities, suggests impacted more than other larger metropolitan areas.
that even though 46 per cent of all current tasks in the
United States are at risk of automation, and 60 per cent Overall, as shown in Figure C.6, the estimated share
of occupations could encompass 30 per cent or more of country’s employment that could be replaced
automated activities, only 5 per cent of occupations by automation differs significantly depending on
could be entirely automated using currently available the methodology and underlying assumptions
technologies. At the global level, the estimated considered. Yet, independently of the methodology
percentage of work activities that could be automated used, the estimated probability of automation is not
ranges from 41 per cent in Kuwait and South Africa equivalent to future unemployment but could still have

Figure C.6: Comparison of approaches to estimate the share of jobs at risk of automation

70
Share of jobs at high risk of automation (percentage)

60 59
55 56
52
50 48 47 46
43
40 38
35 35
30
30

21
20
12
10 9
10 7 6

0
Germany Japan Republic of Korea United Kingdom United States

Occupation-level approach Job-level approach Modified job-level approach Work activity-level approach

Source: WTO Secretariat based on occupation-level analyses (Brzeski and Burk, 2015; David, 2017; Deloitte, 2015b; Frey and Osborne,
2017); job-level analyses (Arntz et al., 2016b); modified job-level analyses (PricewaterhouseCoopers LLP, 2017); and work activity-level
analyses (McKinsey Global Institute, 2017).

95
 WORLD TRADE REPORT 2017

important labour adjustment implications because Secondly, projections of the development and
of compositional changes in the labour market. adoption of future technologies often underestimate
These estimates should therefore be interpreted the challenges encountered during the development
with caution for various reasons (Arntz et al., 2016b; of experimental prototypes and the adjustment of
McKinsey Global Institute, 2017). the production process. The degree of stability of a
given automation process obtained under laboratory
Firstly, the projections of future technological conditions is often difficult to achieve in practice. The
capabilities are based on subjective assessments automation process very often needs to be tailored
by experts, who are not certain of how much and at and adjusted to the firm’s structure and practices.
what pace technological progress will eventually During that process, the firm has to run tests, develop
be achieved. Technology experts are sometimes prototypes and adjust and improve the automation
viewed as overly optimistic about the forthcoming system until it can be embedded in the production
technological feasibilities in their area of expertise process. A recent survey analysis of German firms
and as potentially overestimating their likely progress reports that, although the share is increasing, only
(Autor, 2015). A recent survey analysis suggests, 5 per cent of firms’ production equipment and 8 per
cent of firms’ office and communication equipment,
however, that experts in artificial intelligence and
on average, are based on smart technology, artificial
robotics tend to be more cautious than non-experts
intelligence and robotics (Arntz et al., 2016a). The
when predicting the number of occupations at risk of
risk of potential disruptions caused by machine
automation in the next decades (Walsh, 2017).
breakdowns, broken or mis-specified parts and
worker mistakes can further slow down the adoption
Many technologies, including artificial intelligence
process. As a result, the implementation speed
and automation, as shown in Figure C.7, have
of new technologies remains often uncertain and
developed in spurts. Very often, occasional volatile. 39
technological advancements are followed by a period
of slower progress because of the presence of certain More generally, the adoption of a new technology
challenging obstacles. As discussed in Section C.2, by a firm depends on the cost of software and/or
some scholars also argue that one cannot discard hardware required to implement it and on whether the
the possibility that rapid automation might also be firm has the necessary financial resources to invest
a transitioning phase towards new technologies in it. Other factors influencing the decision to adopt
benefiting labour by discouraging further automation the new technology include the availability of relevant
(Acemoglu and Restrepo, 2016). skills in the workforce and whether the potential

Figure C.7: Evolution of patents on artificial intelligence granted (2000 to 2016)

3,000

2,500
Number of patents granted

2,000

1,500

1,000

500

0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Biological technologies Knowledge technologies

Mathematical technologies Other artificial intelligence technologies

Source: Fuji and Managi (2017) based on WIPO Patentscope database.

96
 TRADE , TECHNOLOGY AND JOBS

economic benefits in terms of greater efficiency Thirdly, and as discussed in the previous subsection,
outweigh the costs. even when new technologies are increasingly being
adopted and used, their effects on employment
Past experience suggests that the adoption of prospects depend to a large extent on whether the
specific technology, such as the use of personal firms adjust to new divisions of labour made possible
computers, can be relatively slow and challenging, by these new technologies. Each industry, and in
because firms adopting new technologies frequently some cases each firm, develops its own set of job
need time to learn and become familiar with specific roles over the years, which often also encompass
practical implementation. For instance, although cloud their own sets of tasks. While some of these tasks can
computing was first commercialized in the 1990s, potentially be automated or digitalized, others cannot.
less than 30 per cent of small and medium-sized Moreover, different production tasks can often involve
enterprises in OECD countries have currently adopted different types of automated functions, some of which
it (OECD, 2016d). An economy’s level of economic may require more complex and expensive automated
development and firms’ absorptive capacity also systems than others.
seem to play an important role. Although technology
adoption lags across countries appear to have The impact of new technologies on employment
declined significantly over the past two centuries, the also depends on a firm’s managerial and corporate
degree to which new technologies diffuse across firms culture, including its human resource management,
and consumers following their initial adoption seems as well as on organizational and social constraints.
to have widened between developed and developing The adoption of a new labour-saving technology
countries over the same period, as shown in could result in a reduction in the number of hours

C. I MPACT OF TECHNOLOGY
Figure C.8 (Comin and Mestieri, 2017). worked and not necessarily in a reduction in the

OUTCOMES
ON LABOUR MARKET
Figure C.8: Average difference in adoption lags and penetration of significant technological
innovations between Western and non-Western countries (1779 to 2008)

100 0.9

90 0.8

Average difference in adoption lags years between

Average difference in intensity use between

80
0.7
Western and non-Western countries

Western and non-Western countries

70
0.6
60
0.5
50
0.4
40
0.3
30

20 0.2

10 0.1

0 0
8 es

ay y fr s
18 ss ht

le s

40 h

76 5 S il
18 Tele eel

ec ne
8 ity

s
03 3 A 92 cks

19 tion on f tor
El ss ght

19 fu rs
19 Fe ce

19 1 S ar er

54 xy tic r
19 idne fu re
L ra ce

H ns t

Pe 73 sur t
on llp ry

19 pu s
In rs
et
68 tra lan

73 19 art lan
19 O the te
18 85 Ma
ip

Te ger

m e
85 ar
18 rap

rs Ce ge

83 te
tri ge

rn
pa eig

H liz

K gen fib
18 tric
l

co on
o

10 rna

63 y t rna
50 yn ves
c
17 ind
Sh

18 5 C

07 p rei
19 90 18 Tru

19 iver nsp

p
82 ph

te
i
Av via Tra
35 en

ec en

19 12 rt
g

h
Sp
8

El
a

c
a
w
79

ti
1

e
Ra ail

al
17

R
ilw
25

ia

3
18

1
25
18

19

Year of technology invention

Intensity of use (natural log scale) (left axis) Adoption lags (year) (right axis)

Source: WTO secretariat based on Comin and Mestieri (2017).

Notes: Western countries include Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Italy, Japan, the Netherlands,
New Zealand, Norway, Sweden, Switzerland, the United Kingdom and the United  States. The right axis represents the average lags
(expressed in years) with which significant technological innovations arrived to developing countries compared to Western countries from
1780s to the present. The left axis represents the average difference in the penetration of significant technological innovations between
Western countries and developing countries from 1780s to the present. Penetration is defined on the basis of the intensive margin of
adoption for each new technology (expressed in logarithm).

97
 WORLD TRADE REPORT 2017

number of jobs. Workers might also adjust to the technologies, such as robotics. The decision to adopt
new technologies by increasingly performing tasks a new technology may also be resisted by those who
complementary to the new technologies. Empirical expect to be negatively affected. Recent empirical
evidence suggests that most of the adjustments research suggests that individuals in European
caused by technological innovation tend to occur countries and in the United States facing economic
within, rather than between, occupations through positions that are more likely to be negatively affected
tasks-restructuring (Spitz-Oener, 2006). The by robotics are more likely to be fearful of robots at
mechanisms by which technology complements work (Dekker et al., 2017; McClure, 2017). Similar
human work are, however, less well understood in the findings were found in the case of Japan, where
literature than those by which technology substitutes workers with limited professional experience, non-
for human work. A recent survey-based analysis of regular contracts, and who were engaged in clerical
Japanese firms operating in the manufacturing and and manufacturing occupations, tended to perceive a
services sectors suggests that the surveyed firms higher risk of being replaced by artificial intelligence
with a relatively larger share of high-skilled workers and robotics (Morikawa, 2017b).
tend to express more positive views on the impact of
artificial intelligence and robotics on the prospective Empirical evidence further suggests that the
number of their employees, while firms with a larger perceptions of workers in the services sector in
share of low-skilled workers tended to anticipate a New Zealand regarding potential changes in their
negative impact on employment (Morikawa, 2017a). workplaces due to artificial intelligence and robotics
seem to be negatively related to their commitment
Fourthly, the studies attempting to quantify the and career satisfaction and positively related to their
share of jobs vulnerable to automation consider only turnover intentions and pessimism (Brougham and
existing jobs. They omit to analyse the new jobs that Haar, 2017). In that context, some experts argue
these new technologies could create. According to that some professional occupations, such as those
the World Economic Forum (WEF) (2016), a large of engineers, lawyers and doctors, may have a
number of today’s most in-demand jobs did not exist greater negotiating power in a firm than other types
10  years ago. For instance, technological progress of occupation to ensure that new technologies
in digitalization has created requirements for app extend and complement their work (Brynjolfsson and
developers, big data analysts and social media McAfee, 2014; Hughes, 2017). 40
managers. The upcoming wave of new technologies
could thus support the growth of different types of More generally, public acceptance of technologies
jobs, including those in charge of developing the new can be a key factor in determining their impact on
technologies, implementing them, and/or supervising society, including on the labour market. Acceptance
and repairing them (Executive Office of the President of new technologies encompasses political
of the United States, 2016). In addition, new acceptance by public and key stakeholders, but also
technologies are likely to require changes in legal by consumers and investors, and by the communities
frameworks and physical infrastructures, which would and regions in which the new technologies are being
create specific new occupations and jobs. developed and implemented. Past experiences show
that high public concern can determine the direction,
New technologies may also have positive effects on speed and diffusion of technological advances
labour demand by raising the demand for existing and and, in some cases, impede their progress even
new products and/or services if they lead to improve when technical and economic feasibility have been
firms’ productivity and increase workers’ wage and established, the rationale for adoption seems sound,
income. As discussed in Section B, frictions in the and important investments have been made. Empirical
labour market can alter the process of allocating evidence shows that public ignorance about the
individuals to jobs and increase unemployment. In this true benefits of particular technologies is often
context, further advances in ICT could also facilitate not the main reason for public opposition to these
the matching of the labour demand and supply by technologies. Other, more important, factors include
reducing the time and resources spent by firms and value conflicts and distributive concerns related,
individuals and improving firms’ efficiency (Dehaze, among others, to jobs and welfare, as well as failures
2016). of trust in institutions, such as regulatory authorities
and technical advice bodies (Winickoff, 2017). Public
Fifthly, the adoption and diffusion of new technology opposition to technologies can, in some cases, lead
does not take place in a vacuum but in a specific to the adoption of new regulations that improve trust
legal and regulatory framework. Some labour and confidence, and orient technological progress
market regulations may make it difficult and costly along pathways that become acceptable to the public
for firms to replace workers with new labour-saving (Davis, 2014).
98
 TRADE , TECHNOLOGY AND JOBS

(c) Implications for skills development experts to be immune to automation, such as perceptual
and supervisory skills, has been experiencing a
While a definitive conclusion on the exact outcome decrease in the United States (MacCrory et al., 2014).
of the new wave of technological innovation on This seemingly contradictory result could be explained
labour markets remains elusive at the present time, by the fact that workers may take on more managerial
the upcoming technological advances will certainly and organizational responsibilities within the same
continue to have an effect on the labour supply, occupations. On the other hand, interpersonal skills
especially skills development through changes and workers’ facility with technology have gained
in labour demand, work organization and skills importance in the last few years. Recent empirical
requirement. In particular, technological advances are research further suggests that individuals who are
likely to continue to be disruptive by rendering specific more intelligent and show an interest in the arts and
qualifications and skills less relevant and obsolete sciences during high school in the United States are
whilst requiring and enhancing other and new ones. less likely to select jobs that are more likely to be
automated in the future (Damian et al., 2017).
Several recent studies, many of which are based on
the methodology used to estimate the share of jobs Many of the skills potentially less exposed to
at risk of automation, attempt to identify the types of automation are already being highlighted as important
skills less likely to be subject to automation. Some by many firms. A recent survey of employers
of these studies identify the jobs least vulnerable conducted by the WEF (2016) highlights an
to automation as those occurring in dynamic and important increase in the future demand for cognitive
changing environments and involving non-routine abilities, systems skills and complex problem-solving

C. I MPACT OF TECHNOLOGY
manual and cognitive skills that have so far been

OUTCOMES
ON LABOUR MARKET
skills, such as mathematics and logical reasoning,
proven difficult to automate. These skills include visualization, systems analysis and creative thinking,
perceptual judgment and manual dexterity skills (used by 2020. 41 As discussed in section E, access to
by nurses and surgeons, as well as housekeepers and higher education, digital literacy and quality training
cooks), social-emotional intelligence skills, such as have been identified by some countries as important
empathetic and negotiating skills (used by educators, means to providing individuals with the responsive,
managers and social workers) and creative skills flexible and complementary skills needed to alleviate
(used by scientists, designers and artists) (Frey and and respond, at least in part, to the current and future
Osborne, 2017; McKinsey Global Institute, 2017). challenges of the labour market. 42

As explained above, some experts also anticipate

that automation will be applied to tasks and jobs 5. Conclusions
located at increasingly elevated levels of the skills
ladder (Susskind and Susskind, 2016). For instance, This section has considered the effects of technology
the “Internet of things”, which enables smart devices on the level and composition of employment and
to send and receive data, could apply to higher- wages. Technological progress is the ultimate source
skilled and complementary tasks undertaken by of economic growth, as it allows for the production of
skilled workers, such as providing online instructions the same amount of output with fewer resources, or
to workers. Where complex digital technologies more output with the same amount of resources.
increase the importance of experiential knowledge,
some specific experiential knowledge could also be Technological progress has ambiguous effects on
eroded or become obsolete. Given the potentially aggregate employment. When such progress takes
shorter life cycle of skills, the development of deep the form of a new product (such as flat screen
soft skills, such as adaptability and learnability, televisions) which replaces an old product (such as
defined as the desire and ability to learn new skills, cathode ray tube televisions), firms producing the
has been identified as essential to grasp complexity, old product go out of business, but labour demand
handle unexpected situations under time pressure, may increase due to additional demand from firms
and take the right actions in those situations without producing the new product. When such progress
necessarily having clear information (OECD, 2016c). takes the place of labour-replacing automation,
In other words, the ability to get and keep certain technological change leads firms to adopt more
types of jobs is likely to depend less on what the capital-intensive technologies and to substitute
individuals already know and more on how and what labour for capital. However, various compensation
new knowledge and skills they are likely to learn. mechanisms (e.g. price-productivity effects, scale-
productivity effects, additional demand in other
Empirical evidence shows, however, that the demand sectors of the economy) can counterbalance this
for some of the skills that are considered by many type of reduction in labour demand. The evidence
99
 WORLD TRADE REPORT 2017

reviewed in this section shows, with some exceptions, This implies that current technological change is likely
few overall effects of technology on the level of to have long-lasting and potentially disruptive effects
employment. on the world of work. This section has evaluated the
arguments put forward by both technology optimists
While having few effects on the level of employment, and by technology pessimists. Technology optimists
technology strongly affects its composition. This is recognize that each wave of technological change
because technological change has different effects in the past generated technological anxiety and led
on different workers, depending, for example, on their to temporary disruptions with the disappearance
skills and on the work tasks they perform. This section of some occupations and jobs, but note that other
has presented theoretical and empirical evidence jobs were modified and new, and jobs which were
showing that current technological change tends to often better were eventually developed and filled.
be skills-biased, in the sense that it increases the Technology pessimists, while recognizing that the
relative demand for skills, and routine-biased, in the fear of technological unemployment has been proven
sense that it decreases demand for routine tasks. wrong many times in the past, consider that the new
Therefore, skilled workers performing non-routine wave of technological progress represents a sharp
tasks tend to benefit from technological change, departure from earlier innovations in terms of speed,
while the latter can be disruptive for unskilled workers scale and force. Definitive conclusions on the exact
employed in routine tasks. outcome of the new wave of technological innovation
on labour markets, however, remain elusive for the
Technological progress is ever-increasing. There present.
is indication that advances in smart technology,
artificial intelligence, robotics and algorithms, often
referred to as the fourth industrial revolution, are
taking place at an unprecedented pace. However,
technological revolutions often take a long time to
have significant impacts. The maximum impact of
steam power on British productivity growth was not
felt until the third quarter of the nineteenth century,
nearly 100 years after James Watt’s patent. The
benefits of railroads were fairly small initially, but
grew as railroad productivity improved and rail
output rose as a share of economic activity. Similarly,
investments in electrical capital equipment did not
have important spillovers until the 1920s. Initially,
factory owners simply replaced large steam engines
with large electric ones. It took nearly 40 years after
electricity was widely available in the United States
for organizational methods to catch up and develop
more efficient decentralized production lines.

100
 TRADE , TECHNOLOGY AND JOBS

Endnotes
1 Automation refers to the use of technologies and automatic 15 Lewis (2004) investigates the technology adoption effects
control devices that results in the automatic operation and of the Mariel boatlift. The Mariel boatlift, which occurred
control of production processes (Electrical Technology, in April 1980, authorized Cubans to leave their country for
2017). a limited period of time. It brought 125,000 Cubans from
Mariel to Miami, creating a 7 per cent increase in the local
2 In this case, labour productivity increases go hand in hand
labour force in five months in the American city (see Card,
with improvements in working conditions.
1990). Lewis finds that post-boatlift computer use at work
3 The expression “technological unemployment” was coined was lower in Miami than in other cities with similar levels
by Keynes. of computer-based employment before the event. This
suggests that the boatlift induced Miami’s industries to
4 Evidence of a declining share of manufacturing employment employ more unskilled intensive production technologies
in several other developed economies is presented in and supports the idea that markets adapt production
Section B. technology to local factor supplies.
5 Industrial robots are defined by the International Organization 16 For other theoretical contributions on the overall
for Standardization (ISO) as “automatically controlled, employment effects of labour-saving technologies, showing
reprogrammable multipurpose manipulator programmable in that the net effects are indeed ambiguous, see Blien and
three or more axes” – see the website of the International Ludewig (2016), Benzell et al. (2015), Sachs et al. (2015)
Federation of Robotics (IFR) at www.ifr.org and Nordhaus (2015). Blien and Ludewig (2016) show
6 IFR data, as elaborated by Graetz and Michaels (2015). that although labour-saving technology may generate
unemployment initially, it may also attract higher product
7 Other factors behind the falling shares of employment but demand. The relative strength of the two forces depends

C. I MPACT OF TECHNOLOGY
ever-increasing output in agriculture include more effective

OUTCOMES
ON LABOUR MARKET
on the demand conditions on product markets. Benzell
land use through crop cycling and fertilization, following et al. (2015) and Sachs et al. (2015) show that a rise in
soil analysis. robotic productivity which substitutes for labour can result
in declining product demand if the output produced by
8 Another typical reason why occupations disappear is lack
robots is sufficiently substitutable for the output produced
of demand, such as in the case of boarding-house keepers
by humans. In the paper by Nordhaus (2015), a situation in
(Bessen, 2017).
which technological change makes human labour obsolete,
9 Similarly, Harrison et al. (2014) show that product innovation denoted “economic singularity”, can arise either if product
has an ambiguous labour displacement effect (which demand is elastic, so that demand restructures to only
depends on productivity differences between old and new ICT-produced goods, or if production is elastic, shifting
products), and a positive compensation effect (related to production to ICT-inputs only.
demand enlargement). Overall, product innovation can
17 For instance, demand rigidities may prevent product
therefore have net positive or negative employment effects.
demand to increase as prices fall. For a more detailed
10 Non-tradable sectors are those which do not trade discussion, see Vivarelli (2015) and Ugur and Mitra (2017).
internationally. Typically, the non-tradable sector comprises
18 Equating technological change with routine task
services where the demander and producer must be in the
specialization has advantages and disadvantages. If the
same location, such as electricity, water supply, all public
aim is to measure automation technologies, routine task
services, hotel accommodation, real estate, construction
measures capture such technologies more broadly than
and local transportation. Commodities which have low
robotics data, as the former include computers, machines,
value relative to either their weight or volume can also
algorithms, robots and the like. However, task allocation
be non-tradable if the transportation charges prevent
is affected by several factors other than technological
producers from profitably exporting their goods (Jenkins et
change, including offshoring, migration and organizational
al., 2011). Due to advances in ICTs, however, the distinction
change.
between tradable and non-tradable sectors becomes
ever thinner, particularly if one considers all the modes of 19 These results are subject to the same methodological
services supply contemplated in the General Agreement on critiques as those of Autor et al. (2013), which are detailed
Trade in Services (GATS). in Section D of this report, and should be interpreted with
caution. In particular, only differential effects between
11 A list of ICT sectors is provided by OECD (2002, Annex 1).
locations, and not a national effect, can be identified by
12 A list of ICT occupations is provided in ILO (2006). the underlying “difference-in-differences” econometric
approach.
13 Brynjolfsson and McAfee (2014) report the telling example
of Instagram, a photo-sharing app. When it was bought by 20 Put differently, the labour demand effects of technology
Facebook in 2012, Instagram had just 13 employees, while substantially depend on who owns the capital, as
Facebook had 5,000. These numbers are only a tiny fraction highlighted by Benzell et al. (2015) and by Sachs et al.
of the number of people employed by Kodak (around 145,000) (2015).
at the peak of its success in photographic film in the 1990s.
21 Several single-country firm-level studies for developing
14 Among the various other factors affecting technology countries also find that introduction of new products is
adoption by firms, there is uncertainty over future profit associated with employment growth. Moreover, they find
streams, sunk costs, the opportunity to delay (Hall and no negative employment effects of process innovation (see
Khan, 2003) and the structure of incentives within firms Crespi and Tacsir, 2013 for a comparative analysis of firm-
(Atkin et al., 2017). level studies for Argentina, Chile, Costa Rica and Uruguay).
101
 WORLD TRADE REPORT 2017

22 Technology can also have an impact at the level of other training (Autor and Dorn, 2013; Brynjolfsson and McAfee,
individual characteristics. For instance, it has been argued 2014).
that younger cohorts are more productive than older ones
32 For the United States, evidence that routine employment
because they are more adept in using new technologies
has declined, while non-routine manual employment has
and keeping up with technological change (Meyer, 2011).
expanded, is also provided by Autor and Dorn (2013) and
In this sense, technology might be biased in favour of
Mazzolari and Ragusa (2013). The World Bank (2016)
young generations. There are also some studies on gender
shows that employment is shifting away from occupations
showing that, especially in developing countries, women
that are intensive in routine tasks in most countries, both
are much less likely to work in ICT sectors or occupations,
high-income and low- and middle-income.
which are well paid, because they are less likely to receive
education in subjects such as science, technology, 33 Autor et al. (2015) conclude from this that technology
engineering and mathematics (World Bank, 2016, Box affects local labour markets only by shifting occupational
2.10). Technology can, therefore, also be biased in favour of composition within sectors. This is in line with the
male workers. conclusions of Section C.2 that the overall employment
effects of technological change are very small and even
23 Goos and Manning (2007) introduce a finer distinction in
positive.
the sphere of non-routine non-manual tasks, distinguishing
between cognitive tasks (e.g. testing hypotheses) and 34 According to “Baumol’s disease”, industries with
interactive tasks (e.g. managing others). This distinction is occupations for which it is difficult to enhance workers’
not crucial for the results discussed in this section and will productivity or in which productivity does not increase
therefore not be considered here. as fast as the economy’s growth, tend to capture a larger
share of the economy’s workforce.
24 The skill premium is the wage of skilled (or production)
workers relative to the wage of unskilled (or non-production) 35 Besides the fear that many jobs could be lost to automation
workers. and robots, other forms of anxiety related to technological
advancements have been discussed in the literature. One
25 Evidence that computer technology is complementary with
of them relates to the potential risk of dehumanization of
human capital is presented by Krueger (1993), who shows
work and society. Conversely, another fear expressed by
that more skilled workers, especially those with higher
some experts is that technological progress is too slow
educational attainment, are more likely to use computers on
because the greatest technological advances have already
the job.
occurred. The lack of technological progress could limit
26 It should be emphasized that different technology indicators the prospect of future productivity gains and ultimately of
have been used in the literature, making it difficult to economic growth.
directly compare the different studies.
36 Recent literature also discusses how the changes
27 The combination of computerization and occupation associated with the “gig economy” offer opportunities for
demand shifters explain roughly 80 per cent of the rise in some individuals, including those excluded from traditional
the skill premium, and almost all of the rise in inequality work modes, such as economically inactive or long-
across more disaggregated education groups (Burstein et term unemployed individuals, but also present a series of
al., 2015). challenges for other individuals (De Stefano, 2016).

28 For an alternative theoretical approach to the routine- 37 Driverless vehicle technology is one area that has attracted
biased nature of technical progress, see Jung and increasing research interest given its potentially large
Mercenier (2014). Cortes et al. (2016) demonstrate disruptive impact on the labour market of truck drivers
analytically that advances in automation cause workers to (Executive Office of the President of the United States,
leave routine occupations in favour of non-routine manual 2016; Davey and Toney, 2016).
jobs and non-employment.
38 A different approach was adopted in a 2016 report
29 Recall from Section C.2 that labour-saving technology published by the WEF (2016), in which the results of a
substitutes labour for capital (substitution effect). This survey of the main global employers in 15 developed and
substitution effect operates mostly in the upper row emerging economies were used to estimate the expected
of Table C.1, because it mostly applies to workers that level of changes in job families. The report concludes that
perform routine tasks. technological advancements, including automation, could
lead to a net loss of more than 5.1 million jobs between
30 The elasticity of labour supply is the percentage change
2015 and 2020 (WEF, 2016). Similarly, Willcocks and
in labour supply following a 1 per cent change in wages.
Lacity (2016) surveyed a large number of firms in the United
The more elastic the labour supply, the more employment
Kingdom and conclude by extrapolation that for every 20
responds to wage changes. Graphically, an elastic labour
jobs lost through robotic process automation, 13 new
supply is flatter than an inelastic labour supply. A perfectly
ones would be created. In addition, the authors estimate
inelastic labour supply, represented by a vertical curve,
that robotic process automation is expected to change at
implies a fixed supply of labour at any wage rate.
least 25 per cent of each job in the economy in the next
31 Middle-skilled workers displaced from routine occupations five to seven years. Combining the projected likelihood of
can also compete with middle-skilled workers in non- skills becoming outdated with survey information regarding
routine occupations with cognitive content and low market the occurrence of previous technological change in
entry barriers. Hsieh and Moretti (2003) show socially workplaces, the European Centre for the Development of
inefficient new entries into the occupation of real estate Vocational Training Combining (2016) estimates that about
broker (a non-routine cognitive occupation with low entry 10 per cent of the jobs of EU employees could be at risk
barriers) in response to rising house prices in the United of technological skill obsolescence. Another approach,
States. Some middle-skilled workers may also compete adopted by Elliott (2017), is based on a literature review
with high-skilled workers, conditional on getting adequate of recent computer science research studies in order to
102
 TRADE , TECHNOLOGY AND JOBS

identify the IT capabilities related to skills used in different 42 Some experts argue that while education and training
jobs that have already been demonstrated to work. The have made it possible to adapt to previous disruptive
author estimates that occupations representing 82 per technological innovations, they are unlikely to mitigate the
cent of current employment in the United States could be impact of future automation, because the new wave of
vulnerable to displacement by IT over the next few decades. technologies are likely to substitute rather than complement
skills, implying that that the number of jobs requiring
39 From a labour market perspective, uncertainty and volatility
an advanced degree could become limited. In addition,
in technology adoption may create an additional burden
increasing educational attainment, which is already high in
on the labour market as it needs to absorb excess labour
many developed and emerging economies, would increase
turnover beyond the long-term trend.
the supply of highly skilled workers and potentially reduce
40 A strand of the literature also analyses the attitude of trade the wage levels of highly skilled workers because of greater
unions towards technological changes, including with competition in the labour market. Lower high skill wages
respect to the risk of job displacement, reorganization of could further reduce the incentive to actually acquire higher
work routines and wage formation (Lommerud et al., 2006), education (Avent, 2016; Brynjolfsson and McAfee, 2014).
and the mechanisms by which trade unions can influence
a firm’s technology choices (Haucap and Wei, 2004;
Addison et al., 2017).

41 Some governments and firms already lament a current

labour supply shortage in some science-, technology-,
engineering-, and math-related skills required to fill the
new job openings fostered by the recent technological
developments (Dehaze, 2016). However, several academics
and experts have questioned the validity of the claim of
these particular labour market shortages, in particular in the

C. I MPACT OF TECHNOLOGY
United States, by noting the increasing number of studies

OUTCOMES
ON LABOUR MARKET
that directly contradict such claims (Charette, 2013). In
particular, the real wage evidence in the United States over
the past decade is not suggestive of a strong increase in
skill demand in science and engineering occupations. If
skill demands were strong and matching skill supplies
weak, wage growth should have been faster over the past
decade.

103