Ten years ago, the economist Carl Benedikt Frey made a prediction that changed the study of employment. With his co-author Michael Osborne, Frey warned that 47 per cent of jobs in the US labour market were “at risk” from computerisation. The paper, which has been cited almost 12,000 times – including in publications by the Bank of England, the World Bank and the US government – was Frey’s first published work as a postdoctoral researcher.
“I didn’t expect that it would receive the amount of attention that it did,” he told me when we met at Oxford University’s Martin School, a facility dedicated to interdisciplinary “frontier” research, where he is director of the Future of Work programme. In the years after his 2013 paper, political shifts occurred that were closely related to the loss of work: the votes for Brexit in the UK and Donald Trump in the US were strongest in deindustrialised areas where skilled jobs had been exported or automated.
In his latest paper, Frey and his co-author, Chinchih Chen, have studied the impact of two factors – Chinese imports and industrial robots – on employment in 352 areas across the UK over 16 years, from 1991 to 2007. They find that while globalisation has been widely blamed for the deindustrialisation of the West – particularly since China joined the World Trade Organisation in 2001 – industrial robots actually have a more pronounced effect on employment in the UK than Chinese imports. “On average, one new robot replaces nine workers in the economy,” he told me.
“Broadly speaking, what we find is that during the period of investigation, roughly 32,000 jobs have been replaced by robots, and roughly 21,000 were replaced by Chinese imports,” he explained.
This disparity is only likely to grow as robots and machine-learning technologies develop and become established across the economy. “Automation is likely to become a more important factor in the future,” said Frey. “The rise of China has already happened. Trading relationships between China and the US, in particular, but also Europe, are deteriorating. But automation is progressing, and it’s taking on new types of work.”
The paper also shows that the UK is, like the US, particularly vulnerable to job losses from automation. But that this doesn’t have to be the case: in some countries, robots are positive for employment. Britain is at a turning point at which it can continue to become more like America (where robots tend to remove jobs), or change to become more like Germany (where robots tend to create jobs). The key to widespread automation without mass unemployment, the research finds, may lie in something the political right tends to portray as inefficient and Luddite: organised labour.
Content from our partners
Because unions protect workers’ employment rights, Frey argues that where there is collective bargaining power, a job taken up by a robot doesn’t automatically put someone out of work. Instead, a new job – perhaps one more suited to a human – is found or created for the worker, and the company’s productivity increases. “If you can easily retrain people, or… reallocate [them] in the company, and they have skills that are still important, that can be very good for productivity,” Frey said. If automation means people work shorter hours or take more holiday, both of these outcomes have also been shown to increase productivity, while companies benefit from lower staff turnover.
[See also: Big Tech and the quest for eternal youth]
He clarified, however, that the benefits are most likely to appear where labour relations are stable. “There’s nothing magic about unions,” he said. “Clearly, when new automation technologies are being adopted, if that comes with labour unrest, that’s not good for productivity.” He pointed to the difference between Sweden, where about 65 per cent of workers are union members but there are relatively few strikes, and the UK, where less than 25 per cent of workers are in a union but industrial action has accounted for 1.6 million lost working days over the last six months. “I think having a tradition of stable labour relations, where the parties trust each other, which you have in places like Sweden… you see fewer strikes and less unrest. The UK has a different tradition. It hasn’t had very stable labour management relations in the past, generally. So, in my mind it is a reflection of that.”
Both the US and Germany are further ahead in terms of automating work: Germany has many more robots across its large industrial sectors, while the US is “much more advanced when it comes to automation technology” than the UK. In a 2018 paper, titled Political Machinery: Did Robots Swing the 2016 US Presidential Election?, Frey and his co-authors argued that America’s own industrial machinery was a factor in the vote. The point, he told me, was to “explain why three key swing states – Michigan, Wisconsin and Pennsylvania, which have been part of the Democratic base at every election since 1992 – ended up swinging in favour of Trump. We found that robots were a large part of that story.”
In the latter paper there are two colour-coded maps: one shows the electoral districts with the biggest swings to Trump, the other shows the districts where the most industrial robots had been adopted. They are almost the same picture. “There are more robots in Michigan alone than the entire American west,” Frey told me.
Frey’s new paper also examines whether industrial robots were a factor in the Brexit vote – they weren’t, as there weren’t enough of them to account for the result, although he observes that “deindustrialisation has clearly had an effect”.
The most worrying outcome for the UK is that a wave of automation does nothing to our chronically low productivity growth. Even for those who don’t benefit from the labour of robots, robot productivity can make goods cheaper. But if a robot simply replaces workers without increasing the amount of stuff produced, all that rises is unemployment and inequality: “If automation accelerates, and you don’t see a productivity offset, that’s a bad situation to be in.” Frey’s concern is that even as automation takes hold in the UK, he isn’t seeing a corresponding rise in productivity: “I do very much worry about the particular trajectory that the UK is on.”
In the week that we met, thousands of workers in technology-focused companies, from banks to Big Tech, were being laid off. These were companies that had scaled up at breakneck speed thanks to their innovations, but their workers were discovering that they could scale down equally fast. Does Frey think technology itself makes the labour market more volatile? “I do think that workers have become more dispensable as a consequence,” he agreed, making institutions such as unions all the more important.
It’s also important, Frey said, that such institutions work together at a national level. “Let’s say Amazon adopts a lot of warehouse robots. That could potentially be good for workers at Amazon, because maybe that makes Amazon more productive, they expand their operations, more people are hired at Amazon, the local union is very happy. But a lot of Amazon competitors are wiped out, and other workers in the industry suffer as a consequence. In that case, you know, bargaining at the firm level, doesn’t really do much for people more generally. It’s not something that you can solve with a few tweaks.”
Working out how to manage future changes can be informed by the past. “The trajectory we’ve been on since the computer revolution of the 1980s, in economic terms, looks quite similar to the first Industrial Revolution,” Frey explained: the share of income that went to workers was falling, wages were stagnant. “The [original] Luddites were right that they were not the ones who stood to gain from the progress during [that] period. They could not have known that their grandchildren would benefit from their misery. And so this was a period of great social unrest. I’m very much worried about the same effects now, in the context of advanced robotics and artificial intelligence.”
[See also: ChatGPT proves that AI still has a racism problem]