Better skills will not save middle-class jobs from automation

Employment & Income,Inclusive Economy28 Apr 2015Jo Michell

Debunking the education versus technology myth

As technology advances, automation will affect a growing number of European jobs. The conventional view is that education is the key to avoiding unemployment, falling wages and stagnation. This view is undermined, however, by the failure of high-wage employment to increase, even as educational attainment rates have risen over the last decade. Instead, the redistribution of wealth and income through taxation, the strengthening of labour bargaining rights and stronger social protection are vital if the European middle class is to share in the benefits of technological change.

Recent decades have seen major shifts in the economic lives of the middle classes in both advanced and developing nations. Advances in technology have led to significant changes in the types of jobs available and working conditions. Much manufacturing production has shifted to low-wage industrializing nations. In advanced nations, de-industrialization and automation have led to the loss of middle-income jobs and expansion of employment in the low-skilled service sector (see more in article ‘Making globalization work for the European middle class‘ of Evert-jan Quak).

Over the same period, the distribution of wealth and income has undergone dramatic shifts, both globally and at the national level. Convergence in income levels between nations, as poorer countries catch up richer ones, has taken place alongside a dramatic widening of inequality within nations. 1 In advanced nations, wages at the top of the distribution have accelerated sharply away from the median, leaving the majority of workers with stagnating, or even falling, incomes. 2 The rise of middle-income nations has seen the emergence of both a global middle class and a global ultra-wealthy elite: the richest Chinese, Russians and Brazilians now receive incomes comparable to those of the wealthiest Americans. 3

Shortage of skilled workers?

It is widely claimed that the emergence of a global middle class coincided with another phenomenon: the ‘hollowing out’ of the middle class in advanced nations. 4 This suggests that traditional middle-class jobs have disappeared while employment in low-skilled service jobs and high-paying elite professions has expanded.

The most commonly cited explanation for these trends is the ‘education versus technology’ hypothesis. Proponents of this theory, such as economists Claudia Goldin, Lawrence Katz and Daron Acemoğlu, claim that changes in technology – the use of computers, in particular – have led to higher educational requirements for workers. 5 New technologies introduce opportunities for workers to raise their productivity and, thus, earn higher wages. For these gains to materialize, however, workers must obtain the requisite skills.

These authors argue that the level of educational achievement over recent decades has been insufficient, leading to a shortage of highly skilled workers and an excess supply of relatively unskilled workers. An imbalance between the high demand for skilled labour and lagging supply (often referred to as the ‘skills mismatch’) generated the sharp rise in wages at the top of the distribution and stagnation elsewhere.

The proposed solution: better quality education

If this view is correct, policy responses intended to support middle-income earners should aim to increase the availability and quality of education. For middle-income workers to avoid being left behind by technology they must upgrade their skills.

Despite predicting an increase in inequality, the ‘education versus technology’ theory denies the possibility of long-term ‘technological unemployment’. 6 In the short term, it is argued, workers may have to resort to menial employment in the service sector, but once they raise their skill levels through education they will shift to higher-wage jobs, as the labour-saving potential of robots increases the value of skilled labour. Lower production costs and higher potential profits will lead to an increase in demand for skilled engineers to tend to the robots and advertising executives to find markets for their output.

An effect of rising incomes at the top of the distribution is to generate demand – conveniently for those displaced by automation – for low-skilled service-sector workers. The demand for service-sector employees to transport, feed, groom and otherwise service engineers and advertising executives should expand accordingly. This, it is claimed, explains increases in both high-wage professions and low-wage service-sector jobs from around the 1990s, while the number of people employed in middle-income jobs faltered – the ‘hollowing-out of the middle class’. 7

Aside from its simplicity, the narrative is appealing because appropriate policy responses are relatively lacking in disruption of the status quo. Educational standards must be raised; if those in the middle class wish to remain there, they should up their pace of skills acquisition. If markets fail to provide the requisite volume of education, the state has a clear justification for policy interventions aimed at raising the supply of education.

More than education alone

Critics of the ‘education versus technology’ view, such as economist Lawrence Mishel, point out that the evidence does not support the theory. 8 The erosion of middle-income blue collar and clerical jobs can be traced back over a much longer period, beginning in the 1960s, rather than in the 1990s when the adoption of computers became widespread. 9 From the 1980s onwards, the most significant trend has been divergence between productivity and wages in the majority of occupations in advanced countries – not just in middle-income jobs. This finding is incompatible with the ‘education versus technology’ theory, which assumes a close relationship between wages and labour productivity.

Crucially, while wage rises at the top of the distribution continued to accelerate, the growth in the number of people in high-wage employment slowed sharply in the 2000s – even as university completion rates continued to rise – casting serious doubt on the claim that excess demand for skilled labour is the key factor in rising inequality. 10 While the period since the 1990s did see a moderate increase in low-end service sector jobs, the significance of this is generally overstated and the sector remains a relatively small proportion of total employment in many advanced countries. Further, wages at the bottom of the income distribution stagnated or fell, despite the claim that demand for low-skilled workers was increasing. 11

The view that the middle class has been ‘hollowed out’ as a result of a technology-induced skills mismatch is not supported by the evidence. 12 What has instead been observed is an initial phase of pressure on skilled manual workers from automation, leading to de-skilling and menial production-line factories. This was followed by major changes in the global structure of production as de-industrialization in advanced nations was accompanied by the ‘off-shoring’ of unskilled labour to developing countries and the moderate expansion of low-skilled service-sector jobs. Initially, the introduction of computers undoubtedly added to the pressure on middle-income manual and clerical workers, but the jobs most recently affected by automation (for example, supermarket checkouts, train ticket sales and passport checking) are concentrated in the low-skilled service sector.

Advanced automation

Where, then, will computers next compete with humans? While automation has so far mostly been confined to occupations involving well-defined routine tasks recent advances have significantly broadened the range of tasks that computers can perform. Computers are now able to cope with unstructured tasks requiring the ability to deal with unpredictable situations. Advances in machine learning, pattern recognition and natural language processing are opening up large new sections of the workforce to competition from automation. 13

One example is the transport industry. While driverless trains have been in operation for some time, driverless cars were, until only a few years ago, thought to be the stuff of science fiction. Google cars now routinely navigate the expressways of California, driving entirely unaided, if not yet unsupervised. Driving a car is a computational task an order of magnitude more complex than driving a train. Taxi drivers, already under pressure from mobile technologies such as Uber, will soon face direct competition from cybernetic drivers. Lorry drivers, couriers and chauffeurs will eventually go the way of telephone switchboard operators and supermarket checkout staff.

Other manual tasks that robots will soon be capable of undertaking include cleaning buildings, picking and sorting fruit and vegetables, serving food in hospitals, and otherwise caring for the sick and elderly. Non-routine cognitive tasks involving natural language processing will be driven not only by increased computational capability but also by access to large datasets. Computers are becoming increasingly competent at such tasks as fraud detection, medical diagnosis, preparing briefs for lawyers and compiling facts from a variety of sources into basic journalistic articles. The ability of top academics to teach thousands of students at a time over the Internet threatens lower-ranked universities with obsolescence. Recent estimates by McKinsey suggest that automation threatens 140 million workers worldwide. 14 Research undertaken at Oxford University predicts that 47% of US jobs and a third of UK jobs are threatened by the next wave of automation. 15

How automation affects the next generation of middle class

What are the implications of all this for the middle class? And what is a middle class job? Defining the ‘middle class’ is a thorny problem. Marx accepted the existence of only two social classes within capitalism: owners of the means of production (capitalists) and workers. Since Marx’s time, a large professional and managerial class has grown to become an important feature of both advanced and, more recently, developing economies.

How is this ‘middle class’ differentiated from other social classes? There are a number of possible criteria: level of income, type of work (manual or cognitive), and level of educational achievement. Cultural indicators, such as consumption patterns, are increasingly used; for example, does a household eat organic home-cooked food or mass-produced processed products? (See more in article ‘Who are the ‘middle’?‘ by Josefine Ulbrich.) All of these criteria are problematic: is the low-ranked government clerk earning 25,000 euros a year in the same class as the banker earning a million? Is a London taxi driver who studied for six years to gain his qualification working class? Is the ex-banker who left the city to open an organic farm middle class? What about a call-centre worker in the UK? Or in India?

It is, therefore, difficult to discuss the impact of automation on the ‘middle class’ in isolation from the rest of the labour force. Since the 1980s, as described above, wages have lagged productivity, not just for middle-income occupations, but for all except top earners. As computational power increases, the potential for automation to replace humans will broaden, placing a growing number of jobs under threat. These will range from relatively low-skilled service-sector jobs such as transport, care, sales and administration, to high-skilled occupations in medicine and law.

Automation will no longer be confined to developed nations; low-skilled, low-wage workers in middle-income countries will increasingly face the threat of replacement by robots. Nike cut employment of low-skilled manufacturing workers by over 100,000 in 2013 by shifting towards automation, despite the fact that the majority of its production takes place in very low-wage countries. 16 iPhone producer Foxconn plans to replace many of its over one million Chinese workers with robots. 17 Call-centre workers in low-wage countries will soon be replaced with computers.

Optimists versus pessimists

How should policy-makers respond to the challenges posed by such developments – and to what extent will raising educational standards help? Opinion is divided. Optimists, 18 such as Erik Brynjolfsson and Andrew McAfee argue in their book, ‘Race against the machine’, that despite disruption during the introduction of new technology, previous episodes of automation have eventually resulted in new jobs and productivity gains in the form of higher wages spreading throughout the workforce. 19 New technology may generate temporary displacement and unemployment, but it eventually leads to the creation of new high-productivity jobs to replace those lost to automation. As the number of working hours needed to produce goods decreases, workers will choose to work less and spend more time on leisure pursuits.

What the optimists fail to take into account, however, is that wage rises following technologically-driven productivity gains have not occurred automatically, but have more often than not been the result of political organization and changes in the bargaining position of different social classes. In Britain, between 1750 and 1850, despite rapid technological change and increases in productivity, real wage growth was sluggish— and completely flat for much of the period. 20 It was only in the second half of the nineteenth century, as the concentration of workers in factories led to their political organization, that wage growth accelerated and workers began to participate in the rewards of technological progress.

This suggests that the simplistic ‘education versus technology’ story is inadequate. The stagnation of wages in advanced nations since the 1980s is due to a complex set of factors including the weakening of organized labour and the shrinking of welfare states, increases in global trade and ‘off-shoring’, and the rise of a dominant financial sector. The belief that political imperatives are key to explaining recent trends in inequality is gaining acceptance.21 While it is undoubtedly true that cost-saving and labour-replacing automation adds to pressure on workers, and that the unemployment that results acts as a disciplining mechanism that serves to restrain wage demands, automation is only one aspect of a broader trend.

Shifting debate

The conventional view holds that education provides a means by which anybody may obtain a high-paying job. However, the evidence increasingly contradicts this view. Researchers – including those who initially developed the ‘education versus technology’ hypothesis – are finding growing evidence that, from around the 2000s, demand for high-skilled workers in advanced countries has stalled, despite continued automation, even as wages at the top continue to accelerate. 22 Employment growth is increasingly concentrated in low-skilled service sectors. Attempts to find evidence of a relationship between educational attainment and productivity have been largely unsuccessful.

With high-quality jobs disappearing, education increasingly becomes a means by which workers compete amongst themselves for a shrinking number of positions. Paradoxically, increasing the availability of education may worsen the situation for many, because once the majority of the population are university educated the competitive advantage from obtaining a university education is wiped out. Ha-Joon Chang uses the analogy of people in a theatre standing to gain a better view: the first to stand succeed, but once everyone in the theatre is on their feet, the advantage is lost and everyone becomes uncomfortable. 23

Neglect of wealth distribution

A crucial problem with the conventional view is the neglect of wealth distribution – a subject which, after being ignored for decades, has recently been resurrected by Thomas Piketty.24 Taken to its logical conclusion, the ‘education versus technology’ argument suggests that, with enough education, everyone can become the owner of a ‘Facebook’ or a ‘Google’.

A more likely outcome is that, despite rising educational levels, uneven capital ownership will concentrate the gains from automation in ever fewer hands. As technology advances and income inequality rises, an increasing share of income will be saved, not spent – as J.K. Galbraith observed, ‘the rich cannot buy great quantities of bread’. 25 The same point is made in the apocryphal quote attributed to automobile union leader Walther Reuther who, during a tour of a newly constructed automated factory was asked by a company official, ‘How are you going to collect union dues from these guys?’ Reuther is said to have responded by asking, ‘How are you going to get them to buy your cars?’

Policy-makers must act now

Policy-makers must be bold in the face of the challenge posed by technological change. If managed well, these advances have the potential to increase the quality of life for the majority. But, if managed badly, as has so far been the case, the result will be a deepening of the current trend of increased concentration of wealth and income in the hands of the few while the majority face job insecurity, unemployment and stagnating real incomes.

An essential first step in reforming policy is for governments to recognize the damaging effects of the emphasis on labour market flexibility and instead focus on protecting the rights of workers. At the same time, policy-makers must reject the dogma that governments cannot ‘pick winners’. Either directly or indirectly, through institutions such as publicly-owned banks, governments must ensure that sufficient investment spending is directed to socially useful activities – such as adaptation to climate change. In doing so, governments will generate decent jobs in the present and the promise of a better life in the future.

As automation spreads, the alternative is worsening inequality, leading to stagnating demand for goods and rising unemployment. 26 Without political intervention to tax and redistribute wealth and income, the result will be secular stagnation and the creation of a permanent ‘technological underclass’. Having a university degree will be of little consolation to the swelling ranks of insecure menial workers and the unemployed.


  1. World Bank economist Branko Milanovic has produced extensive research on this topic, for example, Lakner and Milanovic (2013) ‘Global income distribution: from the fall of the Berlin Wall to the great recession’, Milanovic (2012) ‘Global income inequality by the numbers: in history and now’.
  2. For discussion of the mechanisms behind these trends and their role in the financial crisis of 2008, see Michell (2014) ‘Factors generating and transmitting the financial crisis: functional distribution of income’.
  3. Milanovic (2014) ‘Winners of globalization: The rich and the Chinese middle class. Losers: the American middle class’
  4. For example, Warren and Tyagi (2003) The two-income trap: why middle-class mothers and fathers are going broke, Basic Books; Hollinger (2013) ‘A hollowing middle class’; Thompson (2010) ‘the hollowing out of America’s middle class’
  5. See Goldin and Katz (2009) The race between education and technology, Harvard University Press; and Acemoglu (2002) ‘Change, inequality and the labour market’.
  6. See, for example, Greenwood (1999) The third industrial revolution: technology, productivity and income inequality
  7. Goos, Manning and Salomons (2009) Job polarization in Europe
  8. Others include Paul Krugman, Joseph Stiglitz, and Larry Summers. From other schools of thought, post-Keynesians like Engelbert Stockhammer have long argued that inequality is not mainly driven by technology. Stockhammer has shown the flaws in the econometric evidence presented to support the Skill-Biased Technological Change theory.
  9. Mishel, Shierholz and Schmitt (2013) ‘Don’t blame the robots: assessing the job polarization explanation of growing wage inequality’.
  10. See OECD (2014) ‘Education at a glance’, for detailed data on completion rates.
  11. See Christopolou, Jimeno and Lamo (2010) ‘Changes in the wage structure in EU countries’ for detailed data for EU countries.
  12. For a thorough disection of the ‘education versus technology’ story, see Mishel, Shierholz and Schmitt (2013) ‘Don’t blame the robots: assessing the job polarization explanation of growing wage inequality’.
  13. See, for example, Levy and Murnane (2013) ‘Dancing with robots: human skills for computerized work.
  14. McKinsey (2014) ‘Automation, jobs and the future of work’
  15. Frey and Osborne (2013) ‘The future of employment: how susceptible are jobs to computerisation?’; Frey and Osborne (2014) Agiletown: the relentless march of technology and London’s response.
  16. McAfee (2014) ‘Even sweatshops are getting automated. so what’s left?’
  17. Roubini (2014) Where will all the workers go?
  18. The optimisism is implicit in the ‘skill versus technology argument’ – productivity and wage increases are automatic as long as there is enough education.
  19. Opinon is also divided on the macroeconomic outlook. Technological optimistists argue we are at the leading edge of a new technological revolution which, while disruptive, will produce enormous increases in productivity, e.g., Brynjolfsson and McAfee (2011) Race against the machine: how the digital revolution is accelerating innovation, driving productivity, and irreversibly transforming employment and the economy, Digital Frontier Press. Pessimists see demographics, technology and income inequality combining to produce a prolonged period of low growth or stagnation. The most high-profile statement of this kind is Larry Summers’ ‘secular stagnation hypothesis’.
  20. The Economist (2014) The onrushing wave
  21. For example, Krugman (2007) The conscience of a liberal; Stiglitz (2010) Freefall : America, free markets, and the sinking of the world economy; Rajan (2010) Fault lines: how hidden fractures still threaten the world economy.
  22. For example, Beaudry, Green and Sand (2013) The great reversal in the demand for skill and cognitive tasks; Autor (2014) Polanyi’s Paradox and the shape of employment growth. Larry Summers has recently added his voice to the growing criticism of the ‘education versus technology’ view: The one where Larry Summers demolished the robots and skills arguments
  23. Chang (2011) 23 things they don’t tell you about capitalism, Penguin.
  24. Piketty (2014) Capital in the Twenty-First Century, Harvard University Press.
  25. Galbraith (1955) The Great Crash, 1929, Houghton Mifflin
  26. FootnA rather unsophisticated argument against this view can be found in Coyle (2014)Worry about robots or secular stagnation – not both, in which she argues that technological advances cannot simultaneously lead to both productivity gains and stagnation.ote