Many economists throughout history
have been proven wrong in predicting
that technological progress will cause
irreversible damage to the labour
market. This column shows that so far,
the labour market has always adapted to
the replacement of jobs with capital,
using evidence of new types of skilled
jobs between 1970 and 2007. As long as
the rate of automation of jobs by
machines and the creation of new
complex tasks for workers are balanced,
there will be no major labour market
decline. The nature of new technology,
and its impact on future innovation
potential, has important implications for
labour stability.
Concerns that new digital technologies,
artificial intelligence, and robotics will
create widespread technological non-
employment are now widespread. Various
recent labour market trends, ranging from
declines in US labour force participation
to increases in wage inequality and the
share of capital in national income, are
seen as harbingers of this new normal
(e.g. Brynjolfsson and McAfee 2012, Akst
2014, Autor 2015, Karabarbounis and
Neiman 2014, Oberfield and Raval 2014).
A major shortcoming of the typical
arguments about technological non-
employment is that there is no clear
reason why the effect of new technologies
will be different this time than in the past,
when they did not create such widespread
reductions in employment.
It is not that new technologies weren’t
predicted to be equally calamitous. John
Maynard Keynes stated in 1930:
We are being afflicted with a new
disease of which some readers may not
have heard the name, but of which they
will hear a great deal in the years to
come — namely, technological
unemployment .” (Keynes 1930).
In 1965, economic historian Robert
Heilbroner confidently asserted:
“ As machines continue to invade society,
duplicating greater and greater numbers
of social tasks, it is human labour itself
— at least, as we now think of ‘labour’
— that is gradually rendered
redundant.” (quoted in Akst 2014).
The famous economist, Wassily Leontief,
was equally pessimistic about the
implications of new machines. By
drawing an analogy with the technologies
of the early 20th century that made horses
redundant, he speculated that:
“ Labour will become less and less
important… More and more workers will
be replaced by machines. I do not see
that new industries can employ
everybody who wants a job” (Leontief
So why did these previous dire
predictions not come true? And why
should it be different this time?
Our recent work attempts to answer these
questions (Acemoglu and Restrepo 2016).
Our approach is built on two key ideas.
First, during most times, there is a
continuous process of tasks previously
performed by labour being mechanised
and automated, while at the same time,
new employment opportunities for labour
are created. Second, new employment
opportunities come mostly from the
introduction of new and more complex
tasks in which labour has a comparative
advantage relative to capital. Herein lies
our answer to Leontief’s puzzle – the
difference between human labour and
horses is that humans have a
comparative advantage in new and more
complex activities. Horses did not.
The importance of these new complex
tasks is well illustrated by the
technological and organisational changes
during the Second Industrial Revolution,
which involved not only the replacement
of the stagecoach by the railroad,
sailboats by steamboats, and of manual
dock workers by cranes, but also the
creation of new labour-intensive tasks.
These new tasks generated jobs for a new
class of engineers, machinists, repairers,
and conductors, as well as of modern
managers and financiers involved with
the introduction and operation of new
technologies (e.g. Landes 1969, Chandler
1977, Mokyr 1990).
The importance of new complex tasks can
also be seen in recent US labour market
dynamics. Employment figures document
not just the automation of existing labour-
intensive jobs, but also the rise of new
occupations, ranging from engineering
and programming jobs to those
performed by audio-visual specialists,
executive assistants, data administrators
and analysts, meeting planners, or
computer support specialists. Indeed,
during the last 30 years, new tasks and
new job titles account for a large fraction
of US employment growth. To document
this fact, we use data from Lin (2011)
that measures the share of new job titles
— in which workers perform newer tasks
than those employed in more traditional
jobs — within each occupation. In 2000,
about 70% of the workers employed as
computer software developers (an
occupation employing one million people
at the time) held new job titles. Similarly,
in 1990 a radiology technician and in
1980 a management analyst were new
job titles.
Figure 1 Employment growth by decade
plotted against the share of new job titles
at the beginning of each decade for 330
Notes: Data from 1980 to 1990 (in dark
blue), 1990 to 2000 (in blue) and 2000 to
2007 (in light blue, re-scaled to a 10-year
Source : Acemoglu and Restrepo (2016)
Figure 1 shows that for each decade
since 1980, employment growth has been
greater in occupations with more new job
titles. The regression line shows that
occupations with 10 percentage points
more new job titles at the beginning of
each decade grow 5.05% faster over the
next ten years (standard error = 1.3%).
From 1980 to 2007, total employment in
the US grew by 17.5%. About half (8.84%)
of this growth is explained by the
additional employment growth in
occupations with new job titles, relative to
a benchmark category with no new job
These two key building blocks imply that
one should consider the dynamics of
modern labour markets in advanced
economies as being characterised by a
race between two technological forces:
automation on the side of machines, and
the creation of new complex tasks on the
side of man. While automation is an
ongoing process which, all else equal,
takes jobs away from labour, the creation
of new complex tasks is also an ongoing
process which adds new jobs for labour.
If the first force outpaces the second,
there will be a declining share of labour in
national income and technological non-
employment. If the second force outpaces
the first, the reverse will happen – there
will be a greater share of labour in
national income and rising employment.
Our task-based framework further shows
that automation, though it corresponds to
a technological improvement and
increases GDP, may also reduce the real
wages of workers, not just their share in
national income. This last result is
relevant for understanding a key pillar of
the concerns about the effect of new
technology on wages, which is generally
hard to reconcile with existing models (in
which technological improvements always
increase wages).
Viewed from the perspective of our
theoretical framework, the reason why
illustrious commentators of the past,
including Keynes and Leontief, did not
turn out to be right is that the second
force in the race between machine and
man was every bit the first one’s equal.
Looking into the future, whether the wave
of new technologies will spell doom for
labour will similarly depend on whether
this second force can keep up with the
increased pace of the first.
But in this framework, leaving as
exogenous the rates at which automation
and the creation of new complex tasks
proceed is not fully satisfactory. Though it
helps us understand the forces at work, it
poses an equally deep question: Why was
it that in the past the two forces turned
out to be balanced? Is there any reason
why we should expect the same from
today’s technological developments?
To answer this even more fundamental
question, we develop the full version of
our framework in which the rates at which
automation and the creation of new
complex tasks proceed is endogenised,
and responds to whichever of these two
activities are more profitable. For
example, the cheaper capital is, the more
profitable automation is, which replaces
the relatively expensive labour with
cheaper capital. In the endogenous
technology version of the model, this
greater profitability triggers further
automation. This conceptual structure is
useful for two related reasons. First, it
helps us identify the forces that act as
stabilisers — so that once automation
pulls ahead of the creation of new labour-
intensive tasks, there will be economic
forces that induce a faster creation of new
tasks as well. Second, it helps us
delineate conditions under which the
torrent of new automation technologies
that we are currently witnessing will not
self-correct and will thus have long-term
adverse consequences for the prospects
of labour.
The stabilising forces in the model stem
from ‘price effects’. Because automation
tends to reduce payments to labour, it
also increases the profitability of the
creation of new complex tasks relative to
further automation. This stabilising force
implies that rapid automation tends to
self-correct itself, provided that it takes
place within an environment in which the
technology for creating future innovations
and R&D of different types remains
unchanged. Under these circumstances,
the economy will ultimately return back to
its state before the arrival of these
automation technologies. If so, the
current difficulties of workers in the face
of new technologies notwithstanding, the
future may not be bleak for labour.
Nevertheless, this stabilising force does
not imply that all sorts of changes will
necessarily reverse themselves. If what
has changed is the technology for
creating future innovations, and in
particular, if automation-related
innovations have become easier than
creating new tasks, then the wave of new
automation technologies we are now
seeing will be just the first stage before
the economy settles into a new long-run
equilibrium with worse prospects for
labour. Overall, the extent to which the
future will validate concerns about rising
technological non-employment will
depend on whether we are witnessing a
period of rapid discovery of new
automation technologies or a
fundamental shift in how we are able to
produce technologies for the future.
We also highlight a new implication of our
conceptual structure regarding the
efficiency of the market equilibrium. It is
well-known that models with endogenous
technology have various sources of
inefficiencies resulting from monopoly
markups charged by firms with market
power (which are typically those firms
that introduce new products and
technologies to the market). In addition to
these well-known sources of inefficiency,
we identify a new type of inefficiency,
which leads to too much automation and
too few new complex tasks being created.
This inefficiency arises because
automation, which enables firms to
economise on wage payments, responds
to high wages. When some of the wage
payments accruing to workers are rents
(e.g. efficiency wages or quasi-rents
created by labour market frictions), there
will be more automation than what the
social planner would desire, and
technology becomes inefficiently biased
towards replacing labour.
Finally, we use our framework to explore
the implications of automation for
inequality. When different workers have
different amounts of skills, both
automation and the creation of new tasks
may lead to greater inequality — in the
first case, because machines compete
more strongly against less skilled labour;
and in the second, because the more
skilled workers have greater competitive
advantage than the less skilled in new
complex tasks. However, we show that as
long as over time, tasks become
standardised and are more easily
performed by less skilled labour (e.g. as
in Acemoglu et al. 2010), the introduction
of new complex tasks benefits those
workers as well as the more skilled ones.
Depending on how rapidly this
standardisation process takes place, the
economy might generate powerful forces
self-correcting the inequality implications
of automation technologies as well.
We view our work as a first step towards
a systematic investigation of different
types of technological changes that
impact capital and labour differentially.
Several areas of research appear fruitful
based on this step. First, a more
systematic analysis of the efficiency
implications and how this interplays with
different types of labour market
imperfections (which create wedges
between the opportunity cost of labour
and wages) is an important area for
future work. Second, a richer analysis of
tasks at different parts of the complexity
distribution being automated is an
important area for research, especially in
light of much evidence that automation
will affect not just low-skilled but
increasingly also high-skilled workers in
the near future. Third, since there may be
major differences in the ability of
technology to automate and also to create
new tasks across industries (e.g. Polanyi
1966, Autor et al. 2003), the extent to
which these differences become
constraining factors needs to be
investigated. Finally, and most
importantly, there is great need for
empirical evidence on the impact of
automation and robotics on employment.
Indeed, whether rapid automation does
act as an impetus for the creation of new
complex tasks is of the utmost
importance to provide greater empirical
content to the framework developed here.