The Traditionalists: Running the Numbers and Worrying About It.
By Alberto Moel, Vice President Strategy and Partnerships, Veo Robotics
And we’re back! By now you are surely familiar with our 2x2 matrix of four possible positions in the debate on the macroeconomic impacts of automation.
To recap, the techno-optimists embrace AI and automation and believe they will be good for employment and productivity growth. The techno-pessimists dismiss AI and automation and expect them to never make a difference—they'll neither improve productivity nor create jobs.
The traditionalists, on the other hand, fear AI and automation. They expect these technologies to have a slow but material impact on jobs and job creation, but believe that impact will be negative and lead to long-term shifts in the nature of employment. Rather than believing that AI and automation will expand the pie, traditionalists believe these technologies will only change the widths of the slices. And they are most concerned with quantifying those changes and preparing policy responses.
During 1995-2002, 22 million manufacturing jobs were eliminated in the global economy even as production rose 30%, all thanks to automation technologies. In the United States, between 1982 and 2002, steel production rose from 75 million tons to 120 million tons, while the number of steelworkers declined from 289,000 to 74,000.
Traditionalists believe this pattern indicates that continued advances in AI and automation will have an even more substantial impact on the nature of work and jobs in the longer-term future. Many more jobs will be at risk, and finding a way to prepare for and deal with that dislocation is critical.
The analytical tool in vogue for the traditionalists is to estimate what fraction of a set of job descriptions is amenable to automation and what the economic impact would be if those jobs were automated away. Their findings (at a headline level) aren't encouraging.
A 2013 Frey and Osborne study1 dove deep into the skillsets that have a high likelihood of being computerized. Dentists are safe, as are recreational therapists. Telemarketers and taxi drivers, on the other hand, not so much (Figure 1). Similarly, a 2016 World Bank study2 contends that between 35-77% of all jobs in various countries could be at risk from the rise of AI and automation (Figure 2).
Considering the numbers, it's pretty clear that AI and automation will have a material impact on the future of jobs, or the jobs of the future. But will they really? What are we missing, and why aren’t we as worried as the headline numbers suggest we should be?
To begin with, the devil is in the details. The Frey and Osborne report, a 72-page oeuvre with lots of charts and data and (more importantly) caveats, usually gets summarized as some version of robots are coming for 47% of your jobs in the next 3/5/7/10/20 [pick one] years. The echo chamber that is the blogosphere, populated by a mix of thoughtful people and pundits,3 regularly and continuously uses some variant of that headline to get you to read the latest contribution to the conversation (or more likely just to click on it).
But if you bother to read the report, you will find two (huge) caveats. The first one is the limitations of their methodology:
We make no attempt to estimate how many jobs will actually be automated. The actual extent and pace of computerization will depend on several additional factors which were left unaccounted for.
They estimated an upper bound on what the level of automation could be for a given job description—i.e., how bad it could potentially get—but they did not forecast what that level of automation is going to be—i.e., how bad it will actually get. In other words, their work is an estimation of the worst case scenario.
It’s like weather forecasts in the New York Times:4 in addition to getting your forecasted high and low temperatures for each day, you’ll see the historically lowest and highest temperatures for that day. What’s the likelihood that on any given year, a record will be broken?
The second caveat is even more telling. Somewhere in the text, buried among tables and figures, we find the following nugget:
We do not capture any within-occupation variation resulting from the computerization of tasks that simply free-up time for human labour to perform other tasks.
What does this mean? In plain English, it says that some of the tasks within 47% of job descriptions can potentially be automated. But just because some task in a job is automated, it doesn’t mean the job is going away.5 What about all the other tasks that make up that job?
A telling example is that they assign a 65% or better chance to the automation of atmospheric and space scientist jobs in the future.6 Do we really think atmospheric and space scientist jobs are going to be automated and disappear? No, and neither do Frey and Osborne. All they are saying is that a number of the tasks carried out by atmospheric and space scientists will be automated rather than done by hand. Big whoop.
A 2016 OECD paper7 applies a different methodology: a “whole job” approach to automation. In other words, they investigate which jobs could be completely automated. They come up with 9% over the mid- to long-term. How does that compare to the natural cycle of job creation or destruction in a developed economy?
The US Census Business Dynamics Survey has been tracking job creation and destruction (both in absolute numbers and as a fraction of available jobs), and has found that over the last 40 years or so, about 10-15% of jobs in the US disappear on a yearly basis (Figure 3). Of course, on average, a higher number of jobs are created every year, so the net job creation rate has hovered in the low single digits. Given this rate of job churn (which is no different for other developed economies), the destruction of just 9% of jobs over the long term is not really material. Even if you buy into the misguided idea that 47% of all jobs will disappear in the next (say) 20 years, that’s only a 2.7% job destruction rate per year, which is a fraction of the natural 10-15% job destruction rate.
2015 and 2017 McKinsey reports8 had findings similar to those of the OECD report:
Our results to date suggest, first and foremost, that a focus on occupations is misleading. Very few occupations will be automated in their entirety in the near or medium term. Rather, certain activities are more likely to be automated, requiring entire business processes to be transformed, much like the bank teller’s job was redefined with the advent of ATMs.
All things said, even if the net impact is small, there is no question that at a granular level there could be major changes in the nature of employment and work in the future due to AI and automation. Unlike the techno-optimists, who believe this job reshuffling will be accompanied by employment and productivity growth, the traditionalists are more concerned with the societal and economic dislocations.9 Many reports make an attempt to understand the types of workers that will likely be affected and to what degree, but we really don’t know what the ultimate impact is going to be.
Small or large, the impact of AI and automation on jobs deserves careful thought. Not surprisingly, microeconomic policy responses (e.g., retraining, universal basic income, taxing the robots, upskilling incentives) are being hotly debated. But there is also a material (and maybe even transformational) longer-term macroeconomic impact that has not had a lot of air time. You’ll have to wait for the next blog post to find out what it is, so stay tuned.
1 Frey, C.B., Osborne, M.A., The future of employment: How susceptible are jobs to computerization?, Oxford Martin School, 2013.
2 Arntz, M., Gregory, T., and Zierahn, U., World Bank Development Report: Digital Dividends, World Bank, 2016.
3 Many of which, in the precise words of Rodney Brooks, are “more large hats than cattle.” Australia also has ranches and cattle, so perfectly appropriate.
4 At least in the print edition, which is the one I read. Call me old fashioned, and get off my lawn.
5 In fact, serendipitously, the Economist published an article yesterday quoting a conversation with Frey, one of the authors of this study, on the fact that the common summarization of the study’s findings is incorrect:
“Lots of people actually think I believe that half of all jobs are going to be automated in a decade or two,” he says, leaving half the population unemployed. That is, Mr Frey stresses, “definitely not what the paper says”.
6 This example comes from The Fourth Age, by Byron Reese. A good read overall, so I recommend it to you.
7 The risk of automation for jobs in OECD countries: A comparative analysis, OECD Social, Employment and Migration working paper 189, OECD, 2016.
8 Four fundamentals of workplace automation, McKinsey Global Institute, 2015; A future that works: Automation, employment, and productivity, McKinsey Global Institute, 2017.
9 For further reading: The future of jobs: Employment, skills, and workforce strategy for the fourth Industrial Revolution, World Economic Forum, January 2016.