AIQ speaks to Oxford economist Carl Benedikt Frey about his pioneering research into labour markets and automation.
6 minute read
When Oxford University economists Carl Benedikt Frey and Michael Osborne published The Future of Employment (2013), a study of technological automation, they could scarcely have imagined the response.
The paper has become one of the most-cited in history. Its key finding – that 47 per cent of all US jobs are vulnerable to computerisation – remains a staple of news reports on technology. The study’s methodology has been replicated by Barack Obama’s Council of Economic Advisors, the Bank of England and the World Bank. It has even been the subject of a segment on the satirical news programme Last Week Tonight With John Oliver.
Frey explores how societies can gain from technological progress while limiting the negative effects
In 2019, Frey – who currently directs a programme on technology and employment at the Oxford Martin School – followed up this study with a book, The Technology Trap: Capital, Labour and Power in the Age of Automation. Ranging through history, from ancient Rome to the Industrial Revolution to the advent of AI, Frey explores how societies can gain from technological progress while limiting the negative effects. He spoke to AIQ about his findings.
Could you explain the title of your book?
The “technology trap” refers to the period up until the Industrial Revolution, when the political economy of technological change was such that it was frequently blocked and banned by the ruling class. This slowed down the pace of technology adoption and led to slow economic growth as a consequence. A simple explanation for this is that the craft guilds had strong political influence and wouldn’t put up with anything that threatened their jobs. Monarchs sided with the guilds rather than pioneers of industry and innovators. What changed in Britain during the Industrial Revolution was that the government for the first time began to side with merchant manufacturers and innovators rather than the people doing the rioting.
What does Britain’s experience during the Industrial Revolution teach us about tech-fuelled automation today?
What that period shows is that during periods of rapid technological change, a lot of people can be left behind. The Luddites are often portrayed as irrational enemies of progress, but they were not the ones who stood to benefit from mechanisation, so their opposition to it made sense. These were middle-income craftsmen whose jobs were effectively replaced by children working in factories; while they might have benefited from cheaper textiles, that wasn’t nearly enough to make up for the loss in income.
The Industrial Revolution shows the ruling classes were right to fear technological progress
The Industrial Revolution shows the ruling classes were right to fear technological progress for so long, because it brought a lot of social unrest. The era that began with the mechanisation of factory production and ended with the construction of the railroads also paved the way for [Karl Marx and Friedrich Engels’] The Communist Manifesto.
Clearly we are not undergoing a re-run of the Industrial Revolution, but we are in a period in which middle-income jobs have disappeared and we are seeing growing economic inequality as a result. This is translating into political polarisation. The people whose manufacturing jobs are drying up due to globalisation and automation are the most discontented with their lives and they are more likely to vote for populist policies.
You write that for most of the 20th century, labour and technology enjoyed a more harmonious relationship. Is that because technologies in this period enabled labour rather than replaced it?
In the 20th century you still saw lamplighters being replaced, elevator operators disappearing and so on. But a couple of things changed in the early 20th century. People had the experience of rising wages after the Industrial Revolution, and for the first time they saw that technology can work in their interests. Even most economists during the Industrial Revolution didn’t believe that technology could improve the human lot.
Most economists during the Industrial Revolution didn’t believe that technology could improve the human lot
There were also a lot of product innovations during this time which gave people access to things they couldn’t have dreamed of before: personalised transportation in the form of automobiles, the electrification of the home. Gadgets that relieved people from tedious work in the household were available very cheaply. And while mass production involved machines, it required a lot of labour and complemented people’s skills. From the early 20th century in the US to the manufacturing peak (in terms of absolute employment) in 1979, the working class were doing so well that many of them became firmly middle class.
What has happened since then is that industrial robots, and automation more broadly, has cut down the number of jobs for machine operators and replaced many of those middle-income jobs that provided an elevated middle-class status for people without a college education. It is a period of reversal.
You wrote a famous piece of research about the potential effect of new AI-driven technologies on jobs. Which kinds of roles are most vulnerable?
Before artificial intelligence and machine-learning technology became more pervasive, automation was very much confined to routine, rule-based tasks that could be easily specified in computer code.
Ask the question the other way around: In which domains do computers still perform very poorly?
When my colleague Michael Osborne and I started to look at this in late 2011, we realised that there were a lot of examples of automation that had started to go beyond routine rule-based activities: translation work, driving a car, writing short news stories, medical diagnosis. We tried to think about how the division of labour between humans and computers had evolved and decided to ask the question the other way around: In which domains do computers still perform very poorly? The answer was things like complex social interactions, creativity, the manipulation of irregular objects. Jobs that involve these types of tasks are relatively safe from automation; those could be everyone from artists to software engineers to nurses.
On the other hand, a lot of low-skilled jobs that had been safe havens for workers before – such as truck driving, working in a call centre or warehouse – are now exposed.
Most commentators have attributed the rise of populism to a backlash against globalisation rather than automation. What is the connection between globalisation and automation and are the two trends affecting the same kinds of communities?
Both globalisation and automation are driven by technology – without computers, companies wouldn’t have been able to restructure supply chains to take advantage of cheap labour in places like China. Outsourcing was made possible with ICT. And Chinese import competition and robot adoption affect the same geographic areas, which are more likely to opt for Trump, in the US case. The far-right Swedish Democrats have also done better among voters whose jobs are at risk of automation. But automation and Chinese import competition are distinct phenomena and automation is more broad-based; in the US the output share of manufacturing has been constant over the last 50 years, even as the employment share has fallen.
What steps can governments take to help those affected by automation?
There is a tendency to think that because we are facing this big challenge we need a big solution. But what it comes down to is a lot of small things that make a big difference.
The automation dilemma is very much a dilemma of geography
If you look at the automation dilemma, it’s very much a dilemma of geography. Jobs have been disappearing in places like the rustbelt, and those manufacturing jobs supported lots of other jobs in the local service economy. New jobs are emerging, but in different places, as tech industries tend to be highly clustered in cities. That has driven up house prices in those prime locations, which has made it more difficult for people on lower incomes to move into those markets. Building more houses in those cities, or creating better connections to declining places through smart infrastructure investment, is a very important part of the story.
I grew up in southern Sweden, near Malmö, which used to specialise in building ships. But the shipyards closed down in the early 1990s and Malmö did very poorly for a long time. The revival came with the construction of the bridge to Copenhagen, which meant people could live in Malmö, where housing was cheap, and commute to Copenhagen, where there were better-paid jobs. But they were still spending money locally which gave a boost to the service economy and created this virtuous cycle. Now it’s one of the most dynamic labour markets in Europe.
How about training and education?
Early childhood education matters hugely. We know that children who do poorly in math and reading early in life, perhaps because their parents don’t spend much time with them, do worse in further learning. By investing in early childhood education you are giving people from more-deprived communities the opportunity to get better grades at school and go to college later on. So that needs to be part of it. Income tax credits at the bottom end of the income distribution, to ensure that low-skilled work still pays, could be important; I also discuss measures such as wage insurance for people who are forced out of middle-income jobs and have to take lower-paying jobs.
Who pays for these measures? Should we tax the big tech companies that are rolling out AI?
I’m all for getting these companies to pay more tax. What I don’t think is a good way forward is to say: tax the robots, tax the algorithms, tax the things that make societies more productive. While clearly we need to tax corporations, and maybe we need to tax capital and housing more broadly, the point is that many of these policies will easily pay for themselves.
Most people don’t have the ability to live where jobs and growth happen, which is a huge drag on economic growth
If you think about the huge distortions that occur in the economy, most people don’t have the ability to live where most jobs and growth happen, and that has been a huge drag on economic growth. But if you give relocation vouchers to someone who is unemployed, enabling them to move somewhere else to get a job, that means that person will no longer require unemployment benefits.
Similarly, if you look at the economic returns on childhood education it runs at between seven and ten per cent. Most of these policy proposals are not very expensive. And at a time of negative interest rates they are a no-brainer. Sooner or later, policymakers will have to come up with credible solutions to these problems if they want to win elections. I am reasonably optimistic the right policies will come, but I have no idea how long that is going to take.