Sign In

Communications of the ACM


The New Jobs

The New Jobs, illustrative photo

Credit: Antonio /

Rarely does a day go by without more news predicting the end of work. After all, autonomous vehicles are all but certain to replace truckers and taxi drivers in the coming decades, and robots have already taken over many jobs in factories and warehouses, and will continue to expand their reach beyond heavy industry as they become smarter and ever more affordable.

Perhaps most frighteningly, even professional services no longer seem safe from the encroachment of increasingly sophisticated artificial intelligence (AI). Law firms, for example, employ electronic-discovery software, which uses natural language processing to sift through reams of documents faster and more cheaply than the entry-level lawyers who used to do this tedious work. Deep-learning image recognition tools can flag and classify worrisome tumors in digital scans as well as, or better than, experienced radiologists. Websites like Wealth-front and Betterment, which algorithmically optimize investment portfolios, are giving financial advisors a run for their money.

No one doubts that automation is replacing many jobs, but will advances in technology also create plenty of new jobs? If so, what kind(s)?

These questions have taken on new urgency in the past year, as the startling rise of populism and the election of U.S. President Donald J. Trump have awakened previously complacent white-collar workers to the plight of the large numbers of Americans left behind by automation and other changes in the U.S. economy over the past several decades.

For example, according to a widely reported recent study by economists Daron Acemoglu and Pascual Restrepo, industrial robots alone have eliminated as many as 670,000 U.S. jobs between 1990 and 2007. As some displaced workers stay unemployed, while others must settle for significantly lower-paying work in the retail and service sectors, society pays the price in the form of lower consumer spending, higher crime rates, and a lower tax base.

"Two-thirds of people in this country don't have a four-year degree," points out Moshe Vardi, senior editor of Communications and a professor of computational engineering at Rice University who writes and lectures on the impact of automation on work. "We have to make sure we have an economy that creates jobs for them—and not only jobs for the people who have a Ph.D. in computer science."

What kinds of jobs will the tech-driven economy create? Although no one can peer into the future, experts make educated guesses by looking to the effects of automation in the past, which have followed a pattern of initial disruptions followed by a period of adjustment that culminated in long-term gains.

A classic example is the Industrial Revolution, the massively disruptive transformation of the economy beginning in the 18th century. Industrialization created decades of problems, from changes in job structure and skills demanded to horrendous working conditions for those who did get the new factory jobs; eventually, though, industrialized economies adjusted, with enormous increases in prosperity. A major expansion of education and the rise of labor unions made sure that workers shared in these gains.

Similarly, in the first part of the 20th century, in what's been called the second Industrial Revolution, technological improvements increased job turnover but lowered unemployment, according to a study published last year in the Review of Economics and Statistics (

The big question these days, though, is whether this time will be different. Lacking a crystal ball, scholars are again looking to history—analyzing data from recent years, as the rate of automation has accelerated.

That is the approach used by economists like Anna Salomons of the Utrecht University School of Economics and David Autor of the Massachusetts Institute of Technology, who analyzed 35 years' worth of data for 19 industrialized countries to study the relationship between technology-fueled productivity growth and employment. Across the board, the researchers found that although employment certainly falls in industries with rising productivity, overall employment actually goes up somewhat. "Job losses are occurring in some places," Salomons says, "but they are overcompensated by gains in others." Those countervailing job gains come from two sources, according to Salomons, who holds her school's Chair of Applied Econometrics.

The first is higher incomes, which tend to increase consumption of products and services. Put simply, "we keep buying new stuff, and that keeps producing new jobs," she says. (In Silicon Valley, for example, the tremendous growth in the tech sector has fueled job growth in real estate, the restaurant industry, and personal services like massages.)

The other source of new jobs from automation is what economists call inter-industry demand linkages. "As one industry becomes more productive," Salomons explains, "other industries that use that [industry's products] as an input will also demand more, and as a result you'll also get more labor demand."

Whereas many of us see job titles—factory worker, doctor, engineer—experts on technology's economic impact find it more helpful to think in terms of tasks as the basic unit of work. This more granular view is useful because most jobs don't consist of just one task. "It's a variety of skills and responsibilities," points out computer scientist and entrepreneur Jerry Kaplan, who teaches a Stanford University course on the social and economic impact of artificial intelligence. "So the way to think of it is which tasks to automate."

It is relatively easy to automate routine tasks, since these can be codified in an algorithm; non-routine tasks, on the other hand, have been resistant to even the most sophisticated AI, requiring complex capabilities such as analytical skills, creativity, and compassion. The more of a job's tasks that are non-routine, therefore, the more technology can serve as a complement, rather than a substitute, for the worker—helping them to be more effective, instead of threatening their livelihood.

That reasoning gives cause for optimism about the future of work. This is especially true if you look at it from the perspective of AI developers, for whom augmenting human capabilities makes economic sense, suggests Guruduth Banavar, chief technical officer of Viome, a startup that develops AI for wellness using information from people's microbiomes. "The field of AI is going to take the route of complementing people much more than replicating people," says Banavar, who previously headed research and development for IBM's Watson.

He points out that although humans (like machines) have both strengths and weaknesses, for now "unfortunately we are making people do things that humans are bad at," Banavar says. Think of a radiologist sitting in a dimly lit room looking at hundreds of images each day, searching for anomalies. "It's a boring job, you can be error-prone, and you can get tired." Machines doing the same task make fewer errors—without getting bored or tired. A well-trained AI system, therefore, enables the radiologist to play to uniquely human strengths, such as the ability to make a differential diagnosis.

Differential diagnosis, Banavar explains, requires the doctor to mentally connect images to the patient's history, as well as to textbook knowledge and to the latest scientific findings. "You have to connect all those dots to make hypotheses about what is possibly going on, and then you have to analyze enough observations to decide among those possible hypotheses, and test some of them, then proceed based on what the outcomes of those experiments are."

The doctor also has to interpret the results for each patient and explain the patient's options. "That's a social skill, and those social skills are extremely complicated for machines to learn," Banavar says. Similarly, jobs in management require understanding a complex mix of human and social factors. "I think it's going to be a very long time before we can get machines to do those things," Banavar says. "For social skills, I think the demand will increase over time."

Recent research by Harvard economist David Deming on the growing importance of social skills in the labor market supports Banavar's prediction. "Between 1980 and 2012, social skill-intensive occupations grew by 11.8 percentage points as a share of all jobs in the U.S. economy," Deming writes, pointing to rapid growth in demand for managers, teachers, nurses and therapists, and doctors and lawyers. These socially demanding jobs have also seen wage growth during this period, Deming adds, noting that jobs that require high levels of both cognitive and social skills have fared particularly well.

"The field of AI is going to take the route of complementing people much more than replicating people."

"Computer programmer" might well be one of those cognitive-and-social jobs of tomorrow, especially as programmers continue to boost their efficiency by working with others–seeking and giving answers on Stack Overflow, for example. Busting the myth of the lone coder will encourage more people prepare for tech jobs, believes Caleb Fristoe, the founder of a Tennessee nonprofit called CodeTN, which partners with Knoxville-area schools to teach basic software development skills to low-income students, such as the children of laid-off factory workers. These days, Fristoe says, coding doesn't require extraordinary smarts, or getting a job at the likes of Google.

Instead, CodeTN's goal is to prepare working-class children for generalist coding jobs consisting of unglamorous tasks like managing a firm's login page or setting up a customer database—work that requires only modest aptitude and little formal education.

"New frameworks are lowering the barrier to entry," Fristoe says; that's a far cry from the days when you had to learn the syntax of several programming languages to build useful software. "Rather than typing these seven lines of code to get a menu to pop down, you just download the framework from a code base that allows you to do that in a simpler way," he explains. "Frameworks are taking the hard work that developers prided themselves on out of the equation."

In other words, programming itself has become more automated, potentially opening up job opportunities for more people within a widening circle of employers.

Not all experts are sanguine about the future of jobs, however. "There's a friction in economic systems that economists aren't thinking about," says Vardi, referring to the challenges of matching workers to available jobs. Training hundreds of programmers or plumbers, for example, doesn't guarantee that the local economy can absorb that many workers—nor is it simple for people to move to where the jobs actually are. Just think of Silicon Valley, which has created so many jobs outside the tech sector: the Bay Area is also, not coincidentally, one of the most expensive places to live. And just because there's high demand for nurses doesn't mean men who've lost their jobs in manufacturing will even want to become nurses, let alone retrain and try to get hired.

"You can argue that this is culture, this is bias, or whatever, but they're not moving to those jobs," Vardi says.

* Further Reading

Deming, D.
The Growing Importance of Social Skills in the Labor Market, working paper, May 24, 2017

Kaplan, J.
Humans Need Not Apply: A Guide to Wealth and Work in the Age of Artificial Intelligence Yale University Press, 2015

Autor, D. and Salomons, A.
Robocalypse Now—Does Productivity Growth Threaten Employment? (ECB Sintra Forum on Central Banking Conference Paper, June 19, 2017)

Back to Top


Based in the San Francisco Bay area, Marina Krakovsky is the author of The Middleman Economy: How Brokers, Agents, Dealers, and Everyday Matchmakers Create Value and Profit (Palgrave Macmillan).

©2018 ACM  0001-0782/18/1

Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and full citation on the first page. Copyright for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior specific permission and/or fee. Request permission to publish from or fax (212) 869-0481.

The Digital Library is published by the Association for Computing Machinery. Copyright © 2018 ACM, Inc.


No entries found