VoxEU Column COVID-19 Labour Markets

Covid, hysteresis, and the future of work

Has Covid accelerated the future of work? This column argues that Covid has changed the future of work via four shocks: massive job losses, massive digital transformations, massive debt burdens, and massive costs of socially distanced office space. These matter in two ways. First, due to sunk cost hysteresis, re-hiring workers is very different than retaining workers. Second, the digital transformations, office-space costs, and debt burdens will push firms to replace domestic workers with ‘telemigrants’ or ‘white-collar robots’. The jobs that return will be those that require face-to-face interactions and involve tasks that AI cannot handle. 

Things have changed so much with Covid-19 that not even the future is what it used to be – or so it seems from all the predictions that analysts are making about the post-Covid world. But why should things change permanently? Things are the way they are for good reasons. If the reasons don’t change permanently, the outcomes won’t change permanently. 

Actually, it is more complicated than that. Some socio-economic-political outcomes are governed by what might be called ‘rubber-band dynamics’, others by ‘paper-clip dynamics.’ 

  • When you shock a rubber band – say, pull on it – the shape deforms, but the shape returns when you stop pulling. Temporary shocks have temporary effects.

Most economic phenomena are like this – the standard supply-and-demand mechanism, for instance. 

  • When you pull on a paper clip, it deforms but the shape doesn’t return when you stop pulling. Temporary shocks have permanent effects.

Technically, the two types of dynamics are called non-hysteretic (rubber band) and hysteretic (paper clip).  

When thinking about how the Covid shock will change the future, we must focus on hysteretic processes. Generally speaking, hysteresis happens in systems where the level of some capital stock alters marginal costs or benefits. Here ‘capital’ includes simple versions like human, physical, and knowledge capital, or more complex forms like organisational capital, social trust, aspirations, and expectations. Different sources of hysteresis work with different types of shocks. 

Four Covid shocks

Covid-19 has profoundly shocked the economy in many ways. Four are particularly pertinent to the future of work.

  • An epic number of people have lost their jobs.

In America the figure is almost 40 million. In Europe, fewer have been laid off, but millions have been dislodged from their usual employment relationships (Boeri et al 2020). Many of these separations will last for a few quarters at least.

This matters. Due to sunk costs, a firm’s decision to hire a worker is very different than its decision to retain a worker, and that’s why employment is a hysteretic process. The firings also matter since they boost the attractiveness of office-place automation – what I call ‘white collar robots’ in my 2019 book, The Globotics Upheaval. Automating service-tasks with, for example, robotic process automation (RPA) systems, chatbots, virtual assistants, or higher-end AI-systems is easier when you’ve already laid off the workers they are replacing. 

  • A large share of workers in Europe and America have learned to work remotely. 

Estimates are that from 40-70% of workers in the US and Europe worked from home due to the pandemic (Berg et al 2020, Dingel and Nieman 2020, Brehan 2020). To make this possible, firms have invested in digital transformations – especially in the service sector. Companies have invested in hardware and bandwidth. Workers have learned how to use collaborative software, access remote databases, and participate in virtual meetings. Managers have learned how to manage virtual teams. 

This matters since the human, physical, and organisational investments are not going away. And anything that makes it easier to telework domestically also makes it easier to telework internationally (what I called ‘telemigration’ in my book). 

Now that telework is feasible for a hefty slice of their workforce, firms will soon realise that they can get some of the remote work done at much lower cost by hiring workers sitting in low-wage countries. While it is certainly true that foreign talent working online is unlikely to be as good as domestic talent working in the office, the foreign talent may be a whole lot cheaper. This is just service-sector globalisation, so the deeper point is Covid-linked changes are lowering the technical barriers to this type of globalisation. 

  • Office space in Europe and America got more expensive due to social distancing and other anti-contagion requirements. 

This won’t last forever, but it is likely to be true during much of the recovery phase. 

This matters since firms will face the cost-raising aspects while they are deciding whether to rehire workers in-house, to sink the costs of outsourcing the jobs (domestically or internationally), or to automate the positions using white-collar robots. 

  • Corporate and government balance sheets have already been hit hard by the Covid concussion; more red ink is predicted for many quarters to come.

This matters since firms will be under pressure to cut costs in the short and medium term. Using ‘remote intelligence’ (RI) like telemigrants and artificial intelligence (AI) like white-collar robots will be one way of reducing expenses. 

The post-Covid employment landscape: What can’t RI and AI do?

As the world returns to normal, private and public employers will have to decide which sorts of workers they want, and how many of them – starting from something of a tabula rasa. Some may decide to just go back to doing things the old way. But on the margin, the investment in remote work capacity will surely result in fewer workers in offices. 

So, what jobs won’t be replaced by RI or AI? Applying the logic I developed in The Globotics Upheaval, the jobs that will return are those that are intensive in the sorts of tasks that (1) can’t be done remotely, and (2) can’t be done by white-collar robots. In other words, the way to think about the future of work post-Coivd-19 is to use a process of elimination. The question then is: What can’t RI and AI do? I have a couple of chapters on this in my book, but here’s the gist. 

As it turns out, humans have unique advantages over white-collar robots in things like judgement, empathy, intuition, creativity, ethics, curiosity, and comprehension of complex interactions among teams of humans. Psychologist call this ‘social cognition’ and we have it for very specific, very deep-seated reasons. Michael Tomasello’s book, The Cultural Origin of Human Cognition, asserts that social cognition allowed humans to live in large groups where survival turned on the ability of individuals to cooperate in a complex web of relationships involving trust, kinship, and dominance. The equipment for this is hardwired into the human brain. 

Machine learning has problems with social cognition, as I point out at length in my book. First, even today’s computers aren’t powerful enough given the combinatoric aspects of social interactions. Second, we don’t have the right kind of data to train the white-collar robots. Third, machine learning techniques are a shallow imitation of the biology of human learning, so it may be that a whole new computer-science approach is needed if machines are to rival humans in the most human skills (Lake et al. 2017). 

What kind of jobs can’t be done by telemigrants? Clearly, work that requires local or culture-specific knowledge will be hard to do from overseas. More generally, the key is to think hard about why face-to-face communication matters. Since verbal communication is almost costless these days, a good place to start is nonverbal communication – a topic that is widely studied by psychologists. 

When people are face-to-face in the same room, psychological experiments suggest that less than 30% of the information exchanged stems from the words spoken – some communication researchers put the number as low as 7%. The rest is nonverbal. One clear upshot of this is that forming trust is much easier with face-to-face communication since it is easier to lie with words than it is to lie face-to-face. This is exactly why we trust people more when they say things straight to our face. 

Almost ten years ago, Alan Blinder classified jobs according to offshore-ability, finding that something like 40% could be done remotely. The sectors where less than 20% of the jobs were offshorable – that is, could be done by telemigrants – were mostly of the have-to-be-there type: jobs in hotels and restaurants, transportation and warehousing, construction, leisure industries, education, and health and social care. The most vulnerable sectors to telemigrants were professional, scientific and technical sectors, and the finance, industry, and media sectors. More recently, Dingel and Neiman (2020) came up with the figure that 37% of US jobs could be done remotely. 

Concluding remarks

Most service-sector and professional jobs have been sheltered from globalisation (i.e. remote intelligence) and white-collar robots (i.e. artificial intelligence) since face-to-face interactions were important and computers couldn’t think. That has been changing progressively as digital technology has advanced. 

Pushed by Covid-19, firms and workers have invested in, say, ten years’ worth of digital transformation in just a few months. My guess is that this will accelerate the trend towards more service sector jobs coming into competition with ‘globotics’, i.e. telemigrants working in our offices while sitting abroad (the globalisation part), and software robots replacing particular office-tasks (the robotics part). The jobs that survive competition from telemigrants will be those that require face-to-face interactions. The jobs that survive competition from white-collar robots will be those that stress humanity’s great advantages. 


Baldwin, R (1986), “Hysteresis in trade”, MIT mimeo prepared for 1986 NBER Summer Institute, April, published in Empirical Economics 15: 127–42.

Baldwin, R (1988), “Hysteresis in import prices: The beachhead effect”, American Economic Review 78(4): 773–85.

Baldwin, R (2019), The globotics upheaval: Globalization, robotics, and the future of work, London: Weidenfeld & Nicolson.

Baldwin, R and P Krugman (1989), “Persistent trade effects of large exchange rate shocks”, Quarterly Journal of Economics 104(4): 635–54.

Berg, J, F Bonnet, and S Soares (2020), Working from home: Estimating the worldwide potential,, 11 May. 

Boeri, T, A Caiumi, and M Paccagnella (2020), “Mitigating the work-security trade-off while rebooting the economy”,, 9 April.

Brehan, M (2020), “US Workers Discovering Affinity for Remote Work,” Gallup Poll.

Dingel, J and B Neiman (2020), “How many jobs can be done at home?”,, 7 April. 

Lake, B, T Ullman, J Tenenbaum, and S Gershman (2017), “Building machines that learn and think like people”, Behavioral And Brain Sciences

Tomasello, M (2000), The Cultural Origins of Human Cognition, Cambridge: Harvard University Press, 2000.


1 “Hysteresis in Trade”, was the title of my first sole-authored, post-PhD paper in 1986 (Baldwin 1986); it eventually led to “Hysteresis in import prices: The beachhead effect” (Baldwin 1988) and “Persistent trade effects of large exchange rate shocks” (Baldwin and Krugman 1989). 

15,223 Reads