The question of automation replacing jobs has been especially debated in recent times. This was initially because of COVID-19, which substantially increased investment into developing new technology due to skyrocketing demand for digital goods. Then, with the breakthroughs in artificial intelligence (AI) research, such as the invention of ChatGPT, the public has begun to worry about higher skilled jobs being at risk too. This essay will examine the theory behind decreasing wages and the history of automation and productivity and it will apply that to what we are seeing in today’s world. It will ultimately conclude that while automation can improve living standards for society, it may negatively affect those of low skill workers.
Theoretically, robots can reduce wages. Wages decrease when growth is low, unemployment is high, and worker bargaining power is low. The idea is that robots end up taking over many jobs because they are less costly and more productive. Instead of wages, robots require an initial purchase cost and further maintenance. As technology develops, the supply curve for automation shifts rightwards, driving down the associated costs in comparison to those with workers. In terms of productivity, robots can do more work as they are more dextrous than humans; they are not limited by slow neurological processes, they do not succumb to illnesses or old age, and can be easily maintained. Moreover, humans are fallible to error, which robots will not have through precise programming. As a result, robots may seem the more favourable option, and replace workers. Assuming the demand for output remains the same, more workers become unemployed. Meanwhile, those in work have less bargaining power and cannot negotiate higher wages. Indeed, some research seems to back this up; in the US, for every robot added per 1,000 workers between 1990 and 2007, employment decreased by 0.2% and wages by 0.42% (Acemoglu, Restrepo, 2020).
However, historically automation tends to improve the productivity of workers rather than replace them. Automation has been feared since as early as the 19th century, when a group known as the Luddites emerged in Britain and objected to the usage of automated textile equipment; it made the job less difficult and rewarded low skill workers at the expense of higher skill ones (Smithsonian, 2011). Others also protested that they would be made redundant. Yet time proved them wrong – throughout the Industrial Revolution, unemployment fell and living standards dramatically increased for everyone where technology was improving. Data from the UK, where the Industrial Revolution’s effects were greatest, shows that the share of manufacturing output grew between 1800 and 1900 from 5% to around 20% (Kennedy, 1987). Over the same period, unemployment decreased from around 10% (Spielvogel, 2011) to an average of just over 5%. (National Archives, 2016).
The question therefore lies as to why history and theory differ so greatly in terms of wages and employment figures. This is because the past instances of automation are very different from this one. In the past, mechanisation did not completely substitute labour. Robots eliminated the need for many workers working on the same task by increasing productivity, but still employed several people to operate them. Workers who had lost their jobs to robots were also able to work in other sectors of the economy. This transition of professions was mostly from agriculture towards manufacturing and was not limited by occupational immobility since the skills needed were similar. However, automation was also used in manufacturing, which theoretically should still have reduced the number of aggregate workers. Therefore, a more crucial factor accounting for the positive employment figures was that demand in Britain significantly increased during the 19th century. Between 1801 and 1841, the population in England and Wales doubled to almost 16 million, due to developments in healthcare (Independent, 2006). This thus meant that the derived demand for labour also increased, keeping unemployment relatively low.
To see if people’s fears are justified in today’s context, we can compare them against history. First, in the last 40 years, the UK’s population has increased by 20% from 56.5 million to 67.7 million, which pales in comparison to the 100% increase during the initial stages of the industrial revolution (Macrotrends, 2022). With less of an increase in demand for goods and services, the demand for labour has not increased as much as during the 19th century, and therefore employment prospects may seem bleaker in the face of robots.
The bigger difference in modern day automation is its capabilities. Businesses’ investments into automation have only increased over time. The COVID-19 pandemic accelerated this; while 58% of organisations were using automation in 2019, by late 2020, this was up to 73% (Deloitte, 2023). Robots have become better than humans at numerous tasks, and this is only set to continue as the focus on developing AI sharpens; since 2012, the compute used to train AI has increased by a factor of one hundred million (Financial Times, 2023). Creating artificial general intelligence (AGI), computers that can generate new scientific knowledge and perform any task that humans can, has become the explicit goal of AI companies (Vox, 2023). These ‘superior’ robots can have both positive and negative effects. On the one hand, they could complement labour (Graetz, Michaels, 2018). Firms may become more profitable and be able to expand, creating more jobs for higher skilled workers and being more able to pay them higher wages. Consumers would also benefit greatly from the firms’ increased productivity, as goods and services would come by cheaper. On the other hand, robots may outperform humans, and fit the aforementioned theoretical model, to the detriment of lower skilled workers.
Unsurprisingly, the effects of robots on wages and employment are currently concentrated in manufacturing and simple services such as retail – the electronics sector employs 15% of all existing robots (MIT Sloan, 2020). Robots in these sectors are advanced enough to replace previous workers entirely. Jobs in these sectors on average are low paying; the average blue-collar salary in the UK is £13 an hour, just over minimum wage (Economic Research Institute, 2023). Moreover, these workers tend to have a lower quality of education. Indeed, the pay gap between those who have college degrees and don’t has grown over the last few decades (Economics Observatory, 2021). Therefore, when their jobs are taken, they are disproportionately harmed as they cannot switch to other jobs since they need to undergo new training, which they may not be able to afford.
These results are scarier when we look at an international scale. Traditionally, manufacturing has been offshored to these countries due to their abundance of cheap low-skill labour. However, reports say that two-thirds of jobs in developing countries may be susceptible to automation (World Economic Forum, 2016). This is because the global cost of labour has increased to such an extent that developing countries have been required to produce at standards of firms from rich countries to be able to sell their products, which implies using the more developed technology of the Western world (Rodrik, 2018). These developing countries are already on the back foot in terms of government schemes to re-training their workers due to the lack of tax revenue and states’ inability to borrow at low rates of interest. The pace of automation therefore creates more worry, restricting the time that these economies have to reform their workforce. Another big conundrum is that robots reduce the comparative advantage in manufacturing that developing countries have. Robots in developing countries tend to have a steeper cost curve due to the scarcity of complementary inputs such as skills and infrastructure, meaning that after adopting them, firms are limited in how much they can increase manufacturing output. As a result, there may be lower growth opportunities in developing countries as their developed counterparts choose to manufacture at home. Tanzania is an example of a country which automation is adversely affecting (Diao, Ellis, McMillan, Rodrik, 2021).
Government policy isn’t making the situation any better. In the US, labour has been taxed at an average rate of 25% over the last four decades, whereas capital investment tax has fallen from 15% to 5% (Acemoglu, Manera, Restrepo, 2020). Cost incentives could therefore persuade firms to favour robots over low skilled labour. However, policy could also be used for good. Education programs would be useful to alleviate the pressure caused by automation and spread it more evenly across the population. These programs could be encouraged via conditional cash transfers, such as in Singapore (International Federation for Robotics, 2019). They should also be made as easy as possible to access, nudging redundant workers to take advantage of them. An example of a failing policy would be the Apprenticeship Levy in the UK, which firms complained was restrictive and a mistake that squandered £3.5 billion (Guardian, 2023). But even the more successful schemes, such as Trade Adjustment Assistance in the US, which has helped more than 2.5 million American workers find jobs since 1974, are limited in scale (US Department of Labour, 2022). Government budgets, especially in developing countries, are not large enough to fund school-like programs for all adults, and cheaper alternatives such as MOOCs have low completion rates of around 13% (Warwick, 2014). Therefore, more policy analysis is needed to help workers obtain the skills they need as robot technology advances.
If the results of this analysis hold true, then they justify the reconsideration of how we employ robots. Although robots and AI in the future can improve overall productivity and living standards within the global economy, this effect isn’t spread equally throughout society. Rather, the most vulnerable actors are likely to suffer decreasing wages and will be unable to access the benefits of technological advancements. More investment into educating low skill workers may mitigate this effect to some degree. However, given this is difficult, policymakers should consider restricting growth to maintain equality.
This post is part of a debate on the effects of robots on wages. Click here for the other side.
Stimulating and relevant article Zihan Tian. However, I would argue that the population increase has been greater in the last 40 years (11m) and between 1801 and 1841 (8m). It’s the headcount which matters when assessing the impact on demand rather than the percentage. Do you agree? When you make a comparison you need to carefully define your variables otherwise the commentary can be misleading.
Hi Bill, Thanks for this! I originally argued in my essay that measuring population growth as a percentage is important because it accounts for the infrastructure already set up for a given population; it is easier for a larger population to adjust to a smaller relative change than a smaller population to adjust to a large change.
Point well made – but issue could be discussed and elaborated in a longer essay