We are not the first to worry about technological unemployment. Indeed, concerns over it have been with us for at least two centuries. It dates back to the early ages of the Industrial Revolution in Britain with, for instance, machinery breakthroughs aimed at automating weaving and spinning. The industrial production process based on standardised and repetitive tasks distributed through a complex division of labour rendered factory floors constantly susceptible to replacement by machines depending on developments in engineering science. But an apparent question arises: if technologies were to take over the tasks performed by labour, why did labour demand and labour share remain broadly strong and stable over the last two centuries?
This question has been clearly addressed by Acemoglu and Restrepo, using economic modelling that emphasizes the productivity gains brought about by automation technologies. This is straightforward to understand in that for automation to be economically worthwhile technologies must be cheaper than labour in performing the same tasks. It is this cost-saving feature of automation that stabilizes labour demand because industries can therefore expand and hire more workers in performing non-automated tasks. A clear example of this is the wide adoption of Automated Telling Machines (ATMs) among banks. These machines, as documented by Bessen, actually increased the employment of bank tellers who specialized in non-automated tasks, such as customer relationship management, whilst banks were able to open more branches at a cheaper price.
But productivity growth did not fully explain why the labour share of national income remained roughly constant. Again straightforwardly, if automation were to reduce labour share by displacing workers from their jobs, the opposite state of affairs would mean the creation of new labour-intensive tasks that reinstates labour back into the production process. Historically, the creation of products, services, and even industries often by technological advances has successfully generated new tasks in which humans have comparative advantages over machines. The episode of agricultural mechanisation in the 19th century is a good example of this pattern. Workers displaced from the agriculture industry were largely re-employed to fulfil emerging demand in both manufacturing and service industries.
A general pattern discernible in history is that technological developments do not just automate tasks and automation does not necessarily reduce labour demand. Instead, technological breakthroughs trigger major structural transformations within occupations and industries by changing tasks in which humans have comparative advantages over machines. These processes, in the past, manifested in positive results in the sphere of employment. In light of these experiences, many economists argue that even with terrifying advancements in artificial intelligence (AI) that hold the potential to automate many non-routine and cognitive tasks, things will not be much different this time around too. This argument is justified primarily on the basis that the potential for task creation is estimated to be substantial in many occupations. For instance, Vermeulen and his colleagues documented a large increase in the number of jobs in occupations such as computing, mathematics, architecture, engineering, robotic technology and so on. Vermeulen, therefore, argue that the development of employment and labour markets in the context of advances in AI and robotics technologies is likely to represent just the ‘usual structural change’.
So, will this time be any different? By taking a closer look at the process of structural transformation, we can see that workers are always pushed to adjust and update their skill toolkits to respond to the major shifts in task requirements for labour. It could be that this adjustment process is much more challenging than in the past and will be rather ‘unusual’ insofar as technological developments now raise labour demand mostly in the high-skill occupation category that requires a substantial level of training and educational attainments. For instance, new developments in AI and robotics technology, create complementary tasks, such as programming and training, for highly educated workers to perform. Similar patterns of new demand for skills are expected to unfold in Big Data analysis, software engineering, automotive, 3D printing and so on, which again favour more educated workers. This happens at the same time as declining traditional middle-skill jobs in manufacturing, clerical and administrative support, sales and so on due to computerisation, with which displaced workers should find a way to be re-employed in other occupations preferably with new labour demand. This is the reason why economists advocate the urgent need for more up-skilling and re-education programmes to accelerate the skill adjustment process.
But, can we really manage to accomplish such an upgrade in skills? Isn’t it an illusionary wish? Many professions labelled as ‘high-skilled’ above require college and graduate degrees – in other words, a requirement for new entrants to undergone five to ten years of relevant training and education at least. A more realistic view on future structural transformation is that displaced workers will increasingly tend to enter low-skilled and low-paid service occupations. A consequence of this is downward pressure on the wages of jobs in these occupations, with the declining power of unions, and eventually low-quality jobs. This, together with the increasing demand for the high-skill occupations and a limited supply of qualified workers, may render society subject to huge wage inequality.
Perhaps the good news is that tasks in low-skill occupations are somehow out of the reach of automation due to both technical issues and cost considerations. Hence, remaining labour demand may well absorb the displaced workforce.
However, the Covid-19 pandemic may unfortunately change this picture in at least two ways. First, health concerns and social distancing restrictions on human labour created more economic incentives for firms to automate tasks. Therefore, if job opportunities in some sectors are doomed to decline, Covid-19 acts to shorten the period in which this will take place, putting even harder pressure on the re-employment process.
Second, the pandemic helps firms and workers recognize the substantial indirect costs of being physically present in offices, such as the cost of renting offices and daily commuting etc., which promotes ‘telepresence’ as a new means of relating to the workplace. This has obvious consequences for many low-skilled service occupations as labour demand shrinks for building cleaning, travelling, hotel, restaurant, security, and taxi industries and so forth. If these job opportunities were to permanently disappear, displaced workers would face significant hardship in being re-employed again.
In these respects, this time may well be different. We may well face a much more challenging and painful adjustment process in the labour market. Raising unemployment, wage inequality, low-quality jobs, declining labour unions, a flourishing ‘gig’ economy, will continue to be the outcomes if we fail to respond properly to the requirements imposed by the new technologies we have created.
Qingkun (Eric) Deng is a Research Assistant on the Bristol Model project at the University of Bristol.
Image credit: Lenny Kuhne on Unsplash