Why “How many jobs will be killed by AI?” is the wrong question

By Andrew McAfee and Erik Brynjolfsson  

Over the past few years we've developed artificially intelligent machines that can do many things that used to require human minds: understanding speech, diagnosing disease, checking the terms of a contract, designing a mechanical part from scratch, even coming up with new scientific hypotheses that are supported by subsequent research. As this new software is embedded in hardware we’ll get self-driving cars, trucks, and combines; delivery and inspection drones; and robots of many kinds.

These technologies are improving more quickly than even their creators would have predicted at the start of the decade, and the fact that the world's best players of both the Asian strategy game go and no limit heads up Texas hold-em poker are now AI systems indicates just how deeply they’re encroaching into human territory.

So shouldn't we be preparing ourselves for massive AI-induced technological unemployment? A widely cited 2015 analysis by Carl Frey and Michael Osborne of Oxford University found that 47% of current jobs in the US were susceptible to computerization. And some jobs look especially ripe for automation. As self-driving technology advances it seems likely that many of America's approximately 3.5 million truck drivers could find themselves out of a job.

Despite these scary statistics and scenarios, however, there's no need to panic. For one thing, previous predictions about losses and gains over time in specific jobs have almost always been way off, and there's little reason to believe the current crop will be any better. For another, the Oxford study looked only at destruction, and not also creation. It didn't try to estimate how many new jobs and job categories will come along with future technological progress. There will surely be many of these, from robot wranglers to AI interpreters. Finally, while 3.5 million jobs sound like a lot to lose, there are almost that many layoffs every two months in the United States, and another six million or so people voluntarily leaving their jobs. The American economy is both huge and dynamic; large numbers of jobs are lost all the time, and even more are created.

In fact, a look at recent economic data clearly shows that the demand for good old-fashioned human labor keeps growing, even as AI and other science fiction technologies keep advancing. The size of the total US workforce has increased without fail for eighty months in a row. The total number of hours worked is now at an all-time high, 14.5% percent greater than at the end of the Great Recession. Unemployment now stands at 4.3%, lower than at any other point this century.

But these positive trends don’t mean that there’s nothing to be concerned about. There are two large labor force challenges at present, and they're both due in part to tech progress. The first is that while the engine of job creation is still running it has shifted into a lower gear. This engine used to do a great job of generating lots of solid middle-class jobs that paid more over time. Now it's creating lower middle-class jobs with more stagnant incomes.

There are many reasons for this change, but our MIT colleague David Autor and his collaborators have identified the central one: that the US middle class was built on routine work (both physical, like staffing an assembly line in a factory, and cognitive, like handling payroll for the factory) and this work has been rapidly automated in recent decades. Job growth has continued because there has been a rapid rise in service sector physical jobs like home health aide or short order cook. These are very hard to automate — we are still far more dexterous and agile than robots are — but because they're low productivity they're also low-paid.

The second challenge is that despite very low apparent unemployment, there actually is a serious joblessness problem among some groups. How can this be? It’s because people who have stopped looking for work altogether are not included when calculating the headline unemployment rate. And a surprisingly large percentage of prime-age men, especially less-educated ones, are in this category. According to a 2016 report from the White House, by 2014 more than 16% of US men between the ages of 25 and 54 with a high school education or less had dropped out of the workforce completely. Again, there are many reasons for this phenomenon. It appears that one of them is that many men who did or aspire to stereotypically brawny work like assembly line worker or coal miner are not eager to take available service sector jobs in growing areas like health care, eldercare and education. As automation takes over truck driving and other similar jobs this mismatch between desired and available jobs is likely to grow, as will the joblessness and attendant problems that come with it. 

Furthermore, less skilled workers who do find work often end up with stagnant wages. Real wages are essentially unchanged for the bottom 50% of the income distribution, even as income has grown overall, especially for the most educated and highly paid people in the workforce. This is reflected in growing inequality, but also in greater gaps on other metrics like suicide, alcoholism and drug abuse. As Anne Case and Angus Deaton have documented, “deaths from despair” have increased sharply among the white working class over the past 20 years, after falling in previous decades.

The good news is that we are far from helpless in the face of these challenges. In fact, a strong set of policy interventions is available to help with both income stagnation and concentrated joblessness. One of the most obvious things we could and should do is upgrade our decaying infrastructure, which got an overall grade of D+ this year from the American Society of Civil Engineers. Bringing our highways, ports, bridges, airports, and so on up to world-class standards would be a great investment in our economic future, and it would also create exactly the kinds of hard-hat jobs that lots of currently sidelined men want.

So would supporting the accelerated deployment of renewable energy sources like wind and solar power. These are rapidly becoming cost competitive (which is all the more remarkable given how much the fracking revolution has reduced oil and gas prices), are great news for the planet, and generate lots of jobs. Here again, the AI-powered robots can’t yet install a turbine or array of panels all on their own. The solar industry already employs more Americans than the coal industry does, and there are about twice as many people working in wind as in coal mines. The current administration’s focus on coal is a clear example of trying to drive by looking in the rearview mirror. We need policies that look forward.

We believe that one such policy is a large expansion of the Earned Income Tax Credit, a wage subsidy currently available to low-income workers. A large amount of research shows that the EITC is effective: it not only provides a financial boost but it also directly encourages people to enter the job market. The latter is not true of a universal basic income, which is usually proposed as a cash grant given to everyone, whether or not they’re working. A UBI is thus both less targeted and more expensive that the EITC, but the real problem is that a UBI doesn’t give people any clear reason to get off the sidelines of the economy. Virtually all the social scientists we’ve spoken to and research we’ve seen agree that meaningful work is vital both for individuals and communities. An expanded EITC, which has had bipartisan support, would do even more to encourage participation in the workforce than the current version does, so it should be welcomed.

One of our most urgent policy priorities should be figuring out why entrepreneurship in America has been on a steady decline in recent years. Despite the headlines, the breakout successes from Silicon Valley are the exception, not the rule: business dynamism in our economy is actually decreasing. Fewer new companies are being launched in most industries and regions, fewer people are employed by young companies, fewer people are moving to take new jobs, and so on. The single best way to create more job opportunities for people is to support the creation of lots of new companies that need to hire in order to grow. Ideas from both the right — such as reducing the ever-growing thicket of regulations that confronts a prospective entrepreneur — and the left — like making access to health care more secure and less depending on job status — will help with this.

Tech progress has changed our economy a lot over the past generation, and will change it even more quickly in the years to come. But AI, as impressive and powerful as it is, won’t take over all human work any time soon. Instead of trying to prepare for a jobless future, we should instead be preparing for one that’s a turbocharged version of what we already have: a job creation engine that has shifted into a lower gear, and a large number of people tempted to sit on the sidelines rather than contributing their skills to the economy.

These are serious problems, but not insurmountable ones. The right policies, we believe, can give us the best of both worlds: all the benefits that come from the AI breakthroughs of today and tomorrow and jobs that provide people both dignity and a good paycheck. These jobs and policies are not going to look like the ones of the past, but so what? Throughout our history we Americans have stood out by embracing the future and rising to big challenges. Let’s not stop now.

 Andrew McAfee and Erik Brynjolfsson are the authors of Machine, Platform, Crowd: Harnessing Our Digital Future.