Automation and Employment


Somewhere between Stephen Hawking’s pronouncements on the dangers of superintelligent artificial intelligence and AlphaGo’s victory over Lee Sedol, AI seems to have unexpectedly captured the public imagination, leading to a plethora of news articles, podcast episodes, and even a recently-concluded series of workshops organized by the White House.  Much of this attention has focused on what are presumably very distant scenarios.  More recently, however, the debate seems to have shifted to a much more immediate concern that we may need to confront, one that does not necessarily depend upon the kind of radical advance in technical progress that superintelligence presumes; namely, what will be the impact of robotics and automation on jobs, employment, and society at large?

When I first started reading about this topic, I already had fairly strong preconceived notions about how things would develop.  Given the pace at which technology is advancing, and our increasing ability to automate tasks which previously could only be done by humans, it seemed entirely plausible that we could soon be in a world in which it would be increasingly difficult for the majority of people to find any kind of remunerative work.  And that may still turn out to be the case.  But the economy is a complex beast, and exactly how things will play out depends on many factors, including both technological developments and policy decisions, making it very hard to predict the future with any certainty.

The two most recent books I’ve read on this subject are Race Against the Machine, by Andrew McAfee and Erik Brynjolfsson and Rise of the Robots by Martin Ford.  Published in 2012 and 2015, respectively, both books make the same basic argument: historically, the benefits of new technologies have outweighed the costs; but this time is different.  The advance of technology, they argue, is moving too fast for society to be able to keep up, and poses a serious threat to the society in terms of its effects on employment, opportunity, and equality.

To support this position, both books draw on a similar set of data and anecdotes, both technological and economic.  While I don’t find their evidence entirely convincing, it seems likely that we are headed for a potentially major restructuring of at least some aspects of society, and this is very much a conversation worth having.  Moreover, this discussion also opens the door to a wider and even more interesting conversation about the nature of work, values, and what sort of society we want to live in.  But first, let’s examine the question: is this time different?

1. Precursors

Let’s begin with the obvious.  Developments in technology have had, and are continuing to have, a dramatic impact on what it is possible to do without human labour, and how much it costs to do those things.  Historically, the agricultural sector is the most obvious example of this.  90% of Americans were involved in agriculture in 1800; by 1900 it was was approximately 40% and today it is less than 2%. 1

Moreover, this has not merely been a process of simply automating the work that was once done by humans and draft animals.  The entire food production and distribution chain has been reinvented from the ground up, to the extent of genetically engineering chickens to make them easier to process in automated factories.  Among other things, this transformation effectively made obsolete the millions of horses and draft animals that were employed in farm labour and transportation.  The question currently being raised in many circles is: do recent advances in AI and robotics imply the same thing for humanity?

Concerns of this kind are certainly not new. As far back as 1930, Keynes could write:

We are being afflicted with a new disease of which some readers may not yet have heard the name, but of which they will hear a great deal in the years to come – namely, technological unemployment.  This means unemployment due to our discovery of means of economising the use of labour outrunning the pace at which we can find new uses for labour. 2

As he and others, much of the work being done at that time is no longer being done by humans.  Indeed, from the earliest automated looms, to mechanized agriculture, to nearly fully-automated automotive plants, the history of technology (or alternatively, the history of labour), can be seen as one of people inventing machines to do the work of people.

The profit motive for this is obvious.  Once a robot or a computer can be purchased to do the job of a human for a cost that is commensurate with the human wages, there is little incentive to keep paying the human, especially when the automated system can do the job faster, better, or more reliably.  To quote an article from 1947 that sounds vaguely like a line from The Terminator,

[machines] are not subject to any human limitations.  They do not mind working around the clock.  They never feel hunger or fatigue.  They are always satisfied with working conditions and never demand higher wages based on the company’s ability to pay. 3

But what people did not necessarily expect was the way in which technology would simultaneously lower the costs of goods, and create new forms of employment.  As people moved from agriculture into manufacturing and from manufacturing into service, not only were more jobs created than eliminated, but the new jobs tended to be both less physically demanding and higher paying, leading to large increases in the standard of living for the average person.

Of course, as with all things in economics, the costs and benefits of this transformation were not evenly distributed, and there is no doubt that many people suffered profoundly as a result of technological change.  On the whole, however, the advance of technology has brought enormous benefit for the average person in the form of cheap and widely available goods, unprecedented access to information, and medical advances that not long ago would have been considered miraculous.

2. The Question of Technological Unemployment


Given this positive historical pattern, why should we be concerned?  First, these books argue, technology is advancing at a much faster pace today.  When mechanical devices began to displace human labour on the farms, this transition happened relatively slowly, and at least some people were able to adapt by learning new skills, and transitioning to other types of work.  Today, by contrast, people’s skills are being made obsolete much more quickly, and education is perhaps unable to keep up.

Second, they argue, a much broader range of tasks are now within the purview of automation, thereby threatening a much wider range of jobs.  While previous waves of automation were largely limited to physical processes or simple cognitive operations such as tabulating numbers, we now have smart machines that are threatening to replace a wide range of cognitive tasks, leading to the question of what sorts of jobs people will possibly transition into.

Both Ford and McAfee / Brynjolfsson provide sweeping tours of recent technological developments, and those on the horizon.  Of the two books, Rise of the Robots betrays far greater exuberance about these technologies.  Ford, a silicon-valley entrepreneur, writes excitedly about what he thinks are revolutionary demonstrations of the power of big data and machine learning: the victory of Deep Blue of Kasparov, the amazing power of Google translate, software that can write news stories about sporting events, and of course, IBM’s Watson.

Ford’s main point, I think, is to alert us to the fact that we should not assume that white-collar jobs are safe simply because they require cognitive skills.  Indeed, various professionals, such as lawyers and accountants are already experiencing shocks to their sectors due to a combination of automation and offshoring.  Although the technology is still relatively simplistic in some cases, an 80% solution can still offer a lot of value to a company.  Automated translation between languages is a classic example of a problem, where, despite impressive progress, we still have nothing that approaches human level performance for realistic scenarios.  However, despite the fact that rivaling humans may remain elusive for the foreseeable future, companies may find that today’s solutions are good enough for certain scenarios, particularly as the quality continues to improve and the costs continue to fall.

In other cases, seemingly impressive results can be misleading.  The fact that computers can be trained to write formulaic summaries of sporting events certainly has implications for the people who are currently paid to do that, but we should be careful in extrapolating from that to the idea that all journalists will soon be out of a job.  In addition, the commercial adoption of automated systems, despite their capabilities, may be slower than one would think.  The performance of IBM’s Watson on Jeopardy was a remarkable engineering achievement; despite the hype, however, the company seems to be struggling to find any commercial viable use of its system.

Perhaps more importantly, the follow on effects of these developments can be difficult to predict.  Automated teller machines (ATMs) are a classic example of technology that seemingly threatened to eliminate a large number of service jobs.  It turns out, however, there are actually more bank tellers employed at US banks now than there were in 1990. 4  The reasons for this are complex, but one explanation is that as ATMs were introduced, the cost of running a branch dropped; as a result, banks opened additional branches to compete for market share, thereby increasing overall demand for tellers.  Of course, the nature of what it means to be a teller also changed to incorporate more complex requests from customers that could not be so easily handled by ATMs.

Does this mean that tellers are immune to the effects of automation?  Of course not.  Indeed, we might expect to see a drastic reduction in the number of physical bank branches in coming years, as more people transition to move to online banking.  The point, rather, is that the impacts of automation may be both more complicated and play out over a longer time span than we might at first assume.  One can certainly imagine something similar happening, for example, with translation; if human translators are able to incorporate automated systems into their process, this could potentially increase their productivity, thereby lowering the cost of translation, and leading to an increased demand for human-in-the-loop translation.

If we look at the current landscape of of employment in the United States, there are reasons for both optimism and pessimism.  Certain jobs, particularly those that combine physical and cognitive skills, such as nursing, seem likely to remain immune to technological obsolescence for many years to come, as producing affordable, strong, and flexible, general purpose mobile robots is still an immense challenge on many levels.  Similarly, while automating specific routine tasks is getting ever easier, those jobs that involve a variety of non-routine tasks will likely be less at risk.

On the other hand, some of the most common forms of employment seem dangerously susceptible to virtual elimination, even without any further technological development.  The three most common occupations in the US are currently retail salesperson, cashier, and food preparation / service worker.  Together these three categories include over 10 million workers, or approximately eight percent of the total US workforce.  Both Ford and McAfee/Brynjolfsson emphasize that the replacement of these types of employment is already underway, with dramatic growth in online purchases, deployment of vending machines in place of kiosks (especially at places like airports), and investments in automated fast-food restaurants.

On the other hand, consumer demand is complex, and people find all kinds of surprising ways to make use of their disposable incomes.  While high-end status items such as sports cars are perhaps the most obvious example of luxury goods, one could argue that the equivalent of luxury goods exist across the income spectrum, from professionally brewed coffee and restaurant meals, to live entertainment, to spas and personal services.  Many of these are effectively social services, where part of the value comes from interacting with a human; given the premium people are willing to pay for these services, it seems plausible that at least some people will continue to find ways to market their physical presence as a way of earning income.  While it may already be technologically feasible to have a fully automated fast-food restaurant, it seems highly unlikely that wait staff in traditional bars and restaurants will be replaced by robots anytime soon.

Over the very long term, it is hard to dispute that virtually any task now down by humans could at some point be done instead by a robot or a computer.  There is no good reason to suppose there is any limit on what is technologically possible when it comes to replacing human labour, and certainly nothing special about human cognition that makes it impossible to duplicate in silicon.  In the shorter term, however, the complexities of demand, the adaptability of humans, and the challenges of making robust and flexible machines make it far from obvious to me, based solely on the technologcial evidence, just what the effects of all this technology will be.

3. Unemployment and Inequality

If the technological evidence is somewhat ambiguous, what does the economic evidence tell us?  Both of these books were written in the wake of the 2008 financial crisis, and at that time, the US economy did truly seem to be in almost unprecedented territory.  In particular, the years following 2008 seem to have been a particularly severe kind of “jobless recovery.”  By most metrics, the overall economy rebounded quite quickly after the recession ended; employment figures, however, did not.

By 2011, key measures such as unemployment rate, mean length of time unemployed, and workforce participation rate, all remained poor – worse than or nearly as bad as during the height of the crisis.  Both books present this as evidence of the already-present impact of technology, with companies laying off workers in the bad times, and replacing them with machines when things improved.  In the years since then, however, some of these economic indicators have more or less returned to normal, while others, such as inequality, continue to worsen.

If we look at unemployment, we see that following a truly dramatic increase in unemployment around 2008, things have finally returned to pre-recession levels. 5

Of course, these figures do not count those who have given up looking for work – something that is better captured by labour force participation.  Taken as a whole, it does appear that we are seeing a shocking reduction in labour force participation since 2008, following a decade of stability. 6

These numbers are deceptive, however, as a big part of this is attributable to demographics, with large numbers of baby boomers leaving the workforce as they retire.

If we break labour force participation down by age group the available data is more sparse, but we can still see some interesting trends.  First, there is a slight decrease in the labour force participation rate of people during their prime earning years.  More noticeable, however, is the increase in the proportion of people who are still working after 55, 65 and even 75 years of age.  The biggest decreases are seen in the youngest age groups, which is easily explained by people spending more years in school.  Again, however, this trend is not simple to interpret, as increased investments in education could represent a necessary development of skills, or the only viable alternative when jobs are not available. 7

The main economic measure that both books emphasize, however, is the link between wealth and productivity.  As McAfee and Brynjolfsson write, “In the long run, productivity growth is almost the only thing that matters for ensuring rising living standards.”  Following WWII, as productivity increased (measured as economic output per hour of labor), workers’ incomes kept pace, ensuring rising living standards across income groups. This relationship began to diverge, however, during the 1970s.   Labour productivity growth has continued to be strong in recent years, but median wages have not kept pace.

Although its significance is debated, the graph of “labor share” shows this trend.  Calculated as employee compensation divided by GDP, this metric is notable because it remained remarkably stable for decades following WWII, despite massive changes in technology and productivity.  In the past decade, however, this stability has been shaken, and labour’s share has dropped precipitously. 8

The consequence of this break between productivity and compensation is growing inequality.  As measured purely by GDP per capita, the US economy has mostly been following a steady upwards trajectory (aside from a shockingly large blip around the 2008 crisis). 9 10

The distribution of this growth, by contrast, is far from equal.  The figure below shows the change in household income for each quintile of US households (with the top bracket being represented by the 95th percentile).  As can be seen, the majority of household are effectively no better off than they were 50 years ago (in terms of income). 11

According to a statistic cited by McAfee and Brynjolfsson, 100% of the economic growth between 1983 and 2009 went to the top 20% of households.  Moreover, this pattern repeats itself at finer and finer levels, with the top 1% of households claiming 65% of the growth in the economy since 2002, and  the wealthiest 15,000 families doubling their share of the national income double from 3% to 6% between 1995 and 2007.

Why is inequality so important to this debate?  McAfee and Brynjolfsson see it as an indicator that something fundamental has changed in the economy.  In particular, they emphasize the “winner-take-all” aspect of the current environment, in which a single company, such as Instagram, can capture an enormous share of the market with only a small number of employees, effortlessly scaling up thanks to the power of digital technology. 

Arguably, this trend toward increasing inequality is also evident in the types of jobs being created, with a more stark dichotomy emerging with very well paying jobs on the one hand – jobs which require very high skill levels, and concomitant investments in education – and very low-wage, temporary, and part-time jobs on the other.  Many people have pointed to Uber as a success story in which people are able to earn extra income in a flexible manner without further investments in skills or knowledge; by now, however, it seems clear that from Uber’s perspective, the most valuable function of a driver is to provide training data for the development of automated vehicles.

In the years when these books were written, there did seem to be some economic evidence that technological change was starting to make life worse for the average person, in terms of income, employment, and inequality.  In the years since then, employment levels have normalized, but inequality continues to rise and incomes remain stagnant.  Unfortunately, it is very difficult to attribute any of these trends to technology in particular, as opposed to globalization or many other aspects of macro-economics.

If we put together the data from inequality and employment, there does seem to be an indication that things have started moving towards a situation in which more people are struggling to find a way to earn sufficient income, and part of this may be due to advances in technology.  On the other hand, it is certainly possible to imagine changes to taxation and social support that could offset this and keep the whole machine functioning, even with less people in formal jobs.  Of course, the political challenges to such a change would be immense.  Perhaps more importantly, even if such changes were politically feasible, would that be the sort of world in which we want to live?  Is something like a guaranteed minimum income a real substitute for employment, or is there something in the nature of work that is inherently valuable?  And in the long term, what would society look like if humans can be seen as increasingly superfluous to the economy? 

3. Work and Society

The-end-of-work-bookcoverIf we accept that things are changing as these authors argue, and that we will see a continued rise in unemployment, the question becomes, what impact will this have on society, and what should be done about it.  Here, an older book, The End of Work, by Jeremy Rifkin is more helpful.

Although it was written over 15 years ago, Rifkin anticipates many of the arguments made by the more recent works.  The facts and figures are of course out of date, but the arguments seem fresh, and the overall picture remains the same.  Moreover, Rifkin provides a much better and more comprehensive historical background to the way automation can impact society, including early efforts to shorten the workweek, the difficulties presented by lower production costs, and prescient thoughts about inequality.

Most usefully, Rifkin goes beyond economics, and devotes a considerable amount of space to thinking through the broader societal impacts.  One of the most interesting sections deals with the disparate impacts of automation.  According to Rifkin, in 1949, 6% of cotton in the South was harvested mechanically; by 1964, it was 78%; at least partly as a result, 5 million African Americans migrated north in search of work between 1940 and 1970.  Rifkin argues that this development contributed greatly to the racial unemployment gap that persists to this day.  More generally, he makes a strong case that not only does automation affect different groups in different ways, but that the most disadvantaged groups are likely to be most negatively affected.

Employment, however, is not just about the distribution of wealth, and Rifkin also explores how automation may change our relationship to the very idea of work.  As he writes,

Americans, perhaps more than any other people in the world, define themselves in relationship to their work. . . . The notion of being a “productive” citizen is so imprinted on the nation’s character that when one is suddenly denied access to a job, his or her self-esteem is likely to plummet.  Employment is far more than a measure of income: for many it is the essential measure of self-worth. 12

Indeed, these claims seem to be borne out by studies finding that the physical and psychological costs of long-term unemployment can be quite devastating, akin to losing a loved one, or experiencing a life-altering injury.

Derek Thompson, writing in The Atlantic, noted that “by and large, the jobless don’t spend their downtime socializing with friends or taking up new hobbies.  Instead, they watch TV or sleep.” Although many people are able to use their time creatively, this does not hold for everyone.  It may seem paradoxical, but Thompson’s finding is that “most people want to work, and are miserable when they cannot.”

4. The Future

Despite their gloomy forecasts, all three books end on primarily optimistic notes.  Rifkin, for example, puts his faith in government policy.  He does discuss a guaranteed minimum income, but overall shows much greater enthusiasm for the emergence of what he calls the “third sector” – a kind of volunteer labour force engaged in socially productive activities, remunerated with a living wage more akin to social security than an actually fee for service.  In his conception,

Unlike the market economy, which is based solely on ‘’productivity’’ and therefore amenable to the substitution of machines for human input, the social economy is centered on human relationships, on feelings of intimacy, on companionship, fraternal bonds, and stewardship – qualities not easily reducible to or replaceable by machines. 13

Even ignoring the possibility that these qualities are perfectly replaceable by machines, it seems hopelessly optimistic to believe that such a system could stand up to the vicissitudes of capitalism.  Nevertheless, the combination of a supplemental income and incentivizing labour for the public good does seem to have some potential for keeping more of the population as active participants in the economy.

Ford, the silicon-valley entrepreneur, covers a lot of grounds in terms of the challenges we face, but perhaps inevitably ends with a discussion of the singularity, which I won’t bother to discuss.  McAfee and Brynjolfsson, by contrast, view the future primarily in economic terms.  In particular, the believe most strongly in power of entrepreneurship, and are confident that entrepreneurs will find ways to make productive use of idle labour, especially through strategies which make use of humans working in collaboration with intelligent technologies, combining the strengths of both. 14

To put it another way, the stagnation of median wages and polarization of job growth is an opportunity for creative entrepreneurs.  They can develop new business models that combine the swelling numbers of mid-skilled workers with ever-cheaper technology to create value.  There has never been a worse time to be competing with machines, but there has never been a better time to be a talented entrepreneur. 15

They also, it should be said, end their book with a set of concrete policy recommendations, ranging from  education, to entrepreneurship, to investment in infrastructure, to laws related to employment, regulation, patents, and taxation.

Despite these concluding notes of optimism, it is easy to imagine a darker future.  As many people have pointed out, the most ominous looming development is that of driverless cars and trucks.  There are millions of people in the US employed in driving cars, trucks, and buses, and driving is the most common occupation for American men. 16  While there are still many hurdles to widespread deployment of automated vehicles, including technological, legal, and societal, self-driving cars have already begun to appear on our roads, and it is conceivable that the transition to full automation could come sooner and more swiftly than anyone expects.  If it does, we might eventually see the kind of growth through increased demand that we did with bank tellers, or we might suddenly see a huge number of people added to unemployment rolls with which the government would find it very difficult to cope.

While all authors emphasize the importance of education, it seems hopelessly unrealistic to think that everyone will be able to transition into highly-skilled technical roles; even among the next generation it seems unlikely that we will see more than a relatively modest increase in the number of people who choose to pursue a scientific or engineering education.

Ultimately, the rate of change at which this happens will likely be a critical factor.  Over the very long term, we can imagine all kinds of ways in which society could gradually adapt to gradual changes, and even utopian scenarios in which the need to work is gradually eliminated.  If changes are less gradual, however, it is equally easy to imagine society’s ability to adapt being overwhelmed by technological forces.  Assuming the economy continues to grow despite the lack of formal participation by an ever-increasing number of people, are we headed toward a Star Trek-type scenario, in which free goods and energy leads people to transcend the competitive nature of capitalism, or to the hypercompetitive, class-based genocide of Atlas Shrugged?  To borrow Robert Reich’s somewhat arrogant phrasing, “What do we owe one another as members of the same society who no longer inhabit the same economy?” 

While the evidence that “this time is different” is not nearly so strong as I had expected it to be, we cannot ignore the real possibility of the displacement of large numbers of jobs by automation, possibly sooner than we expect. I personally believe that the effects of all this might be stranger than we can predict, and that the economy may prove more resilient and adaptable than one might think.  But I still think it’s enormously valuable that we are having this conversation, and that these questions are being seriously considered.  We can only hope that the discussion will consider these questions broadly – not just in terms of GDP and retraining, but in terms of the bigger picture of what sort of society we want to live in, and how to get there.




  1. Race Against the Machine
  2. Keynes, J. M. Economic Possibilities for our Grandchildren. 1930.
  3. Leaver and Brown, “Machines Without Men”, Fortune, November 1946
  4. James Bessen, “Toil and TechnologyFinance & Development, March 2015, Vol. 52, No. 1
  5. Federal Reserve Bank of St. Louis: Civilian Unemployment Rate
  6. Federal Reserve Bank of St. Louis: Labor Force Participation Rate
  7. Bureau of Labor Statistics: Civilian labor force participation rate by age, gender, race, and ethnicity
  8. Federal Reserve Bank of St. Louis: Nonfarm Business Sector: Labor Share
  9. Federal Reserve Bank of St. Louis: Real gross domestic product per capita
  10. Because population growth has been approximately linear, graphs of GDP and GDP per capita show approximately the same trend.
  11. United States Census Bureau: Table H-1. Income Limits for Each Fifth and Top 5 Percent (All Races)
  12. The End of Work
  13. The End of Work, p. 292
  14. Rifkin similarly quotes William Leiserson: “the army of the unemployed is no more unemployed than are firemen who wait in fire-houses for the alarm to sound, or the reserve police force ready to meet the next call.”
  15. Race Against the Machine.
  16. United States Census Bureau: Detailed Occupations and Median Earnings: 2008

Leave a Reply

Your email address will not be published. Required fields are marked *