Carl Benedikt Frey on Technequality:
“The Pandemic Has Increased Anxiety Towards Automation”
Fears around automation are centuries old, but advances in artificial intelligence have created fresh concerns. Robert Scammell hears from Dr Carl Benedikt Frey, director of future of work at Oxford Martin School, about automation’s impact on the labour market in the age of coronavirus
Is artificial intelligence (AI) going to steal your job? Contrary to the alarming headlines, it’s not a binary answer. In recent years, a consensus has emerged that AI and automation will continue to replace simple, rules-based and repetitive jobs. This will eventually see workers take on new roles that could prove more fulfilling and for fewer hours. To some, this is the wheel of progress.
But advances in machine learning – AI that is able to adapt without following explicit instructions – will see more complex roles become automated over time, causing displacement in the job market. What will happen to the labour market during the transition, and what does that mean for equality?
Dr Carl Benedikt Frey, an economist and director of future of work at Oxford Martin School believes that there are similarities between the current wave of disruptive automation and previous waves of technological change.
“What we do know is that historically automation has had both winners and losers,” he says, speaking at the World AI Summit, a virtual event.
The key question – and one that he addressed in his 2019 book The Technology Trap – is whether we should “feel reassured if the future of automation mirrors the past”.
Take the Industrial Revolution, which took place between 1760 and 1840. As labour-replacing machines entered factories – driving up production and causing a surge in GDP – there was fierce opposition from the displaced workers.
“What we do know is that historically automation has had both winners and losers.”
Chief among them were the Luddites, a group of textile workers that destroyed the machinery they saw as threatening their jobs – actions that seems futile and senseless today.
“Luddite efforts to avoid short-term disruption in the labour market meant that people were also denied some of the long-term benefits that had the potential to occur,” says Frey.
He believes the reason why resistance continued throughout the first Industrial Revolution was because it took a while for the benefits of automation to be felt by most in society.
Such resistance to technological change affecting labour markets has been “the historical norm, rather than the exception,” he says.
So what does this teach us about the latest period of labour-replacing technological change – one driven by advances in machine learning?
Covid accelerates automation shift
First, Frey believes that the coronavirus pandemic is accelerating the use of automation technologies. Alongside this, Covid-19 is shifting people’s attitudes.
“What the pandemic has done, if anything, is to increase anxiety towards automation,” he says.
He points to research that shows people in China – a country that has seen great economic benefits from becoming the world’s manufacturing hub – have changed their views on automation.
Before the pandemic, just 27% of respondents in China said they favoured limits and restrictions on automation that might threaten jobs. Since Covid-19, that figure has jumped to 50%.
“And this is not surprising, because during a recession people face worsening outside job options,” explains Frey.
“Losing your job to automation or to other sources of dislocation is worse during an economic downturn because you have fewer job options to choose from, and that's why we saw greater automation anxiety during the Great Depression, more automation anxiety during the three post-Korean War recessions, and also after the great recession [in 2008].”
The Trump factor
The short-term impact of automation on labour markets can create ripple effects that extend beyond their local geographies and help reshape the political landscape.
Areas where industries have “deindustrialised particularly rapidly” are more likely to have introduced automation into production, says Frey.
He gives the examples of the US states of Michigan, Wisconsin and Pennsylvania, which all voted for a Democratic candidate in every election since 1992.
“Those communities have felt a lot of the sort of social ills from automation.”
They are also the three states that now have the greatest number of robots and automation in manufacturing jobs.
“All of a sudden, they ended up being won by the Republican candidate President Trump in 2016 – and not without reasons,” says Frey. “Those communities have felt a lot of the sort of social ills from automation, in terms of persistent unemployment, deteriorating public services, increases in crime, increases in opioid abuse, and so on and so forth.”
Where the disenfranchised Luddites were unable to express their concerns at the voting booth, Americans today can and have made their feelings known.
“I think the economist Wassily Leontief was on to something when he suggested that ‘if horses could have joined the Democratic Party and voted, what happened on the farms might have turned out differently’,” he says.
“They could have used their political voice to express concerns over automation, despite the obvious benefits to the economy as a whole.”
“We’ve seen the rise of China already, but we haven't seen the end of the automation story.”
Yet much of the discourse around the problems facing these communities has tended to pin the blame on America’s exposure to China’s exports – in large part thanks to Trump’s rhetoric and the ongoing trade war.
“Automation and China's input competition has actually had very similar effects on the labour markets,” says Frey. “And while we’ve seen the rise of China already, we haven't seen the end of the automation story so far, because the potential scope of automation has expanded very rapidly only in the past few years.
“It used to be the case that automation was confined to routine rule-based activities that can be easily specified in computer code... But what we’re seeing now is that, with advances in machine learning, the potential scope of automation is much greater, and a lot of tasks that we used to deem non-automatable, like driving a car, medical diagnostics, translation works, those things are looking increasingly automatable.
“And many of those jobs, like driving a truck, can't be offshored to places like China. But they can be automated. And I think it's a real concern that attitudes towards automation might shift as it progresses, and it becomes clearer that restrictions on imports aren’t going to solve that problem.”
AI threatens multiple sectors
According to Frey’s research estimates, a “very significant” share of jobs in transportation, construction, food preparation and retail are now exposed to automation due to recent advanced in AI.
For some jobs, such as cashiers, it has been clear for some time that they are at high risk for replacement by automation. But advances in generative adversarial networks, which can create realistic and original images of humans, mean that even fashion models could see their work replaced by automation. Already, companies such as Dior having made use of the technology.
“Automation has historically not been a steady process, it's often come in spurts. And one thing that we do see is that Covid-19 has already accelerated automation. Jobs that are more risk to automation, according to our estimates, have disappeared more rapidly and so far at least they have failed to rebound.”
Looking back, the 2008 recession saw businesses shift to using automation for simple rules-based tasks in a bid to cut back on costs.
So what does automation mean for the distribution of wealth? Frey believes that levels of income inequality have been “approaching levels not seen since the first industrial revolution”.
“The reason for this is obviously not spinning machines and steam powered factories, it is the declining cost of computational power, which allows businesses on the one hand to offshore production to low-cost destinations like china and coordinate production at distance, and it also allows increasingly businesses to automate routine, repetitive tasks. And this something which we see quite clearly when we look at the distribution of income wealth over time.”
“It's important to remember that these technologies are likely to bring enormous benefits over the long run”
And as history has demonstrated, during periods of labour-replacing technological advances there has been more social and political unrest.
Even before the pandemic, there were signs of a backlash against automation. In 2019, dock workers in Los Angeles went on strike to protest the introduction of autonomous cargo trucks.
Research has also shown that broadly, Americans favoured restrictions on automation to protect jobs (with the exception of automation for hazardous jobs).
But as with the Industrial Revolution, the net-benefit over time will likely outweigh the short-term pain caused by job displacement. However, it is vital to take measures to ensure a smooth transition, says Frey.
“I think it's important to remember that these technologies are likely to bring enormous benefits over the long run, just like the industrial revolution did, it can also bring significant dislocations over the short term.
“And to ensure acceptance for automation we need to think about policies to smoothen the transition or we might face similar resistance, which may deny us the long-term benefits of these technologies.”
Back to top