Nearly Half of U.S. Jobs Could Be Done by Computers, Study Says

12,160 8 Loading

If computers become as smart as humans, will they do our jobs better than we can?

A recent study [pdf] out of Oxford University found that almost half of U.S. jobs are vulnerable to being taken over by computers as artificial intelligence continues to improve.

The study, based on 702 detailed job listings, found that computers could already replace many workers in transportation and logistics, production labor and administrative support.

But computers, armed with the ability to find patterns in big data sets, are also increasingly qualified to perform "non-routine cognitive tasks."

"While computerization has been historically confined to routine tasks involving explicit rule-based activities, algorithms for big data are now rapidly entering domains reliant upon pattern recognition and can readily substitute for labor in a wide range of non-routine cognitive tasks," write study authors Carl Benedikt Frey and Michael Osborne.

Software already provides medical diagnoses and does legal research, for example.

robots-and-jobs-alexskopjeTo be clear, the study doesn't predict, based on economics, whether more jobs will be automated, but rather, whether they could.

And it suggests that, notwithstanding smarter algorithms, the jobs most vulnerable to being given to a computer in the near future will be the lowest paying.

“The authors do not claim 47 percent of total U.S. employment will be lost as a result of computerization. They only argue that 47 percent of U.S. employees are in occupations that are at risk as a result of computerization,” Gary Burtless, an economist with the Brookings Institution, told Singularity Hub.

The economic effects of automation nevertheless remain uncertain because of the number of variables in play: possible regulation, the relative costs of labor and computing power, and whether or not workers receive additional training to move into new jobs.

Technological progress has always caused turbulence in the labor market, Burtless said, regularly foreclosing particular areas of work. But workers eventually land in other occupations. How long “eventually” takes also depends on how healthy the overall economy is and whether labor markets were prepared for the changes.

Just as the invention of the automobile squelched all the jobs building horse-drawn buggies, driving buggies and crafting whips for buggy drivers, they also opened up highly paid jobs building cars, economists inevitably point out. (Of course many of those jobs have since been given to robots.)

Economists David Autor and David Dorn back Burtless’s argument. But, in a recent New York Times op-ed, they suggested that it needs refinement.

Computers are better at certain types of jobs, like those the Oxford study identifies as at immediate risk. Those jobs are largely entry-level office jobs. Displaced workers mostly have to move down into lower paying jobs because they can’t move up, Autor and Dorn argue. In this way, technological progress has contributed to the gap between the rich and the poor without reducing the total number of jobs in the United States.

Benedikt and Osborne say automation will return to eating the lowest-paying, rather than middle-income, jobs because computers continue to lack social intelligence. But that still doesn't mean the workers they replace won't find new jobs, Autor told Singularity Hub.

artificial-intelligence-ryger“It could certainly be that ‘this time is different’ -- computers are ‘more human’ than cars, after all -- but the burden of proof is on the ‘it's a serious threat to employment’ camp to show why this era is exceptional, and to present any meaningful evidence that technological change has reduced net employment even in this decade,” Autor said.

It seems, then, the effects of artificial intelligence in the workplace will likely be complicated but not catastrophic.

Studies that point to which jobs are at risk can help make sense of the economic dust storms in the forecast, Burtless suggested.

The Oxford study is useful because it points toward “the occupational areas where technical progress in the form of computerization has the potential effect of reducing or boosting the demand for workers,” he said.

Images: Danomyte, alexkopj, ryger via

Cameron Scott

Cameron received degrees in Comparative Literature from Princeton and Cornell universities. He has worked at Mother Jones, SFGate and IDG News Service and been published in California Lawyer and SF Weekly. He lives, predictably, in SF.

Discussion — 8 Responses

  • Steve Morris September 25, 2013 on 1:52 am

    The lesson of the past few thousand years is that machines create wealth and improve our lives.

  • John Sherrill September 25, 2013 on 10:26 pm

    The quality of life that results from the enhanced manufacturing technology makes it so that even despite having less relative income/employment, the same amount of money can buy much more, so there is an increase of quality of life.

    Lots of the alarmism surrounding this topic is very selfish, only thinking of short term effects on individuals, and ignoring the net effect on society.

    • Dvir Raz John Sherrill September 29, 2013 on 1:13 pm

      When your child has nothing to eat or a place to live, you can’t afford to think of a better future for the rest of society.

  • cabhanlistis September 26, 2013 on 3:10 pm

    “Displaced workers mostly have to move down into lower paying jobs because they can’t move up”

    -Are we to assume that the number of available lower paying jobs will reflect the increase in people moving down into those positions?

  • Ver Greeneyes September 30, 2013 on 1:43 pm

    I wonder how long politicians will continue to push the idea that everyone must have a job to contribute to society in the face of increasing absurdity. Right now, certainly, it’s the way our economy works – and in the face of increasing welfare and medical costs, it makes sense to encourage as many people as possible to support themselves.

    But what about 20 years from now, when AI and robots are increasingly capable of replacing human workers, providing cheaper and more efficient labor? What do we do when people can only get a job as a scientist, an artist, a designer or a social worker? And what if robots become user friendly enough to replace even these jobs? At some point, something will have to give – and we should be preparing for that future. But here in the Netherlands, at least, I’ve not heard a single politician even mention these problems – even the most progressive parties push the idea that everyone should work to their capacity, they’re just more nuanced about it.

    Personally I think the very concept of capitalism is doomed for this very reason – the idea of rewarding people for productivity and innovation is a good one, but we’ll need to find a new way to do it (preferably one that *doesn’t* lead to 1% of the population accumulating 50% of all wealth). How do you encourage people to become scientists or teachers when they could spend their lives playing sports or gaming? The latter may not sound too appealing to many, but I know several people who’d prefer to do just that.

    • John Sherrill Ver Greeneyes September 30, 2013 on 5:17 pm

      Abolishing capitalism is dangerous though, if there is communism (as in Venus Project communism, not idiotic Marxism) then if the governing system fails then unpleasantness will ensue.

      Hopefully, if things go well, then as money becomes less important, food etc will become so cheap that the vendors just stop caring about getting paid. People will just forget about money, and the only people worrying about it are the spacers doing kilo-scale engineering.
      Only then will I feel that it is truly safe to abandon the concept of capital, but then it won’t really be abandoned, just naturally forgotten.

  • Charlie Porcaro October 1, 2013 on 2:35 pm

    The first generation of AIs are already here. They’re called Guatemalans and Cambodians. They are adaptable, easily trainable, and low maintenance. They don’t usually deviate from their programming or attempt to alter their algorithms. Their glitch rate and cost of operation is still lower than the robotics that have been developed over the years….