Will self-learning robots replace paid labour?

Introduction

The fear of unemployment due to automation is nothing new. In the 19th century, protesters destroyed textile industry machines (Thompson, 1977). Today, we are regularly warned about unemployment due to automation (Casper, 2016; Dijkstra, 2019). In the 1960s, President Lyndon B. Johnson set up a commission to investigate the supposed danger of automation (Autor, 2015). The commission concluded that while automation would replace some jobs, it would not eliminate work entirely (Bowen, 1966). However, many people believe that this time will be different. They expect that self-learning robots will soon lead to a lack of paid work for everyone. This expectation is based on two assumptions. The first is that our need for goods and services produced by human labour will grow more slowly than the increase in productivity from self-learning robots. The second assumption is that self-learning robots can take over the jobs currently performed by humans. In the following paragraphs, I will provide arguments for and against these assumptions. A conclusion is given in the last section. 

Why many of today’s jobs are vulnerable to automation

According to estimates from CEDA (2015) and Schwab (2017), 35-50% of jobs in Western countries are vulnerable to automation. This is partly due to the increased computing power and falling costs of computers (Nordhaus, 2001), which have allowed for the development of self-learning robots that can perform tasks previously only possible for humans (CEDA, 2015). For example, robots may soon be able to pack different products by navigating in three-dimensional space and sensing the size and quantity of the products. However, the current limitations of robots, such as their limited computing power, may prevent them from entirely replacing human workers in certain tasks. Moore’s law predicts that the computing power of computers will increase exponentially (Moore, 1965), which may enable robots to become more advanced in the near future. In addition, advances in networking technology have allowed for the centralised control of multiple robots, reducing their costs and increasing their effectiveness (CEDA, 2015). It is often assumed that humans and computers compete with each other in terms of absolute effectiveness. In practice, however, this is not the case. A robot that is ten times slower than a person but 100 times cheaper is still profitable. In addition, a robot does not have to work flawlessly; it just has to be better than the person it replaces. These principles are important in valuing the competitive position of the robot compared to humans. 

Will self learning robots replace paid labour?

In the previous paragraph, three reasons were given why intelligent robots may take over people’s jobs. Yet there is no historical indication of a reduction in the total number of jobs due to automation. Labour participation has increased over the past hundred years despite technological developments. The demand for goods and services has shifted, but has not fallen (Autor, 2015). Besides the fact that there are no indications of a reduction in the demand for goods and services despite increased productivity, the question remains whether intelligent robots will be able to replace humans in the labour market in the first place. There are many skills that people possess, but we cannot describe how we perform those skills. An example of this is the human trait of empathy; almost all people possess empathy, but we cannot accurately describe how we show empathy. Because we cannot accurately describe how we display this property, we can only teach these skills to machines to a limited extent.

Ways of self learning capacity

There are several ways in which self-learning capacity can be developed (Nilsson, 1983). A widely used technique is the ‘brute force’ method, in which a lot of data allows the computer to make connections (Alpaydin, 2016). This is currently happening mainly with image recognition. The computer is presented with a large number of images and then asked to determine which images show a particular object. After the computer has selected the images, it shows which predictions are correct and which are incorrect. The computer then adjusts the algorithm to make a better prediction. This process is repeated until the computer arrives at a correct prediction. In practice, the computer often gives a good answer with this method, but also makes many mistakes because the computer cannot reason (EconTalk, 2020). For example, it turned out that a computer could distinguish chairs from other objects based on the height and size of the seat. But the moment the image is disturbed (for example, because someone sits on the chair), the computer can no longer determine that it is a chair (Murphy, 2000). The self-learning computer can therefore be an important tool to support professionals, but not a replacement, because the computer cannot reason enough to independently fulfil a function. Recognising images is already successfully applied in recognising a disease in X-rays. In some instances, the self-learning robot is even more accurate than the human radiologist (Hosny, Parmar, Quackenbush, Schwartz, & Aerts, Hugo JWL, 2018). Nevertheless, the radiologist remains of great importance in interpreting the analysis results in the patient’s context. 

In addition to jobs that consist of some degree of standardised work, such as the radiologist as mentioned earlier, there are also jobs that take place in a constantly changing environment. An example is the work of an organisational consultant. The consultant must weigh different interests for each assignment in an environment highly subject to change. As a result, the consultant’s work cannot be translated into a collection of rules (if x then y). If these rules cannot be created by a programmer or a self-learning algorithm, an intelligent robot cannot take over the work. 

Are self-learning robots harmful to society?

All in all, it can be said that the self-learning robot will not take over all paid work, but probably a significant part of it. The question that remains is, to what extent is this development harmful? This question can best be answered by looking at the effects of self-learning robots entering the labour market. In addition to positive effects such as cheaper products and a smarter healthcare sector, a number of negative effects can have consequences for a large group of people. An effect that can already be observed is the effect of orbital polarisation. The self-learning robot currently mainly affects jobs in the middle segment of the labour market. Here, routine activities are carried out that can easily be taken over by intelligent technology. An example of this is the position of a shop or counter clerk. These functions are increasingly being taken over by online services and machines such as ATMS and collection machines. In addition, administrative tasks have been further automated or outsourced to the customer, so that fewer employees are needed. Employees who cannot work their way up to the top of the labour market are pushed to the bottom of the labour market. The demand for labour has increased faster at the top of the labour market than at the bottom, widening the gap between high and low incomes (Autor, 2015). To deal with this problem, training opportunities are important to realise sufficient growth opportunities, even with a shrinking middle segment. Learning analytical and interactive skills is particularly important here because these skills are complementary to intelligent automation systems (CPB, 2015). In addition to job polarisation, there is a second way in which the self-learning robot promotes inequality. The robot’s owner is the only one who earns from the productivity increase that the robot brings. The employee who is more productive or, in the worst case, is replaced by a robot, does not share in the profit. As a result, the return on capital is significantly greater than the return on labour. Which further increases inequality. A possible solution would be for companies to be obliged to provide some form of profit sharing or payment in shares to their employees. This allows both investors and employees to benefit from increased productivity (Went, Kremer, & Knottnerus, 2015). In addition to the fact that automation affects the segments of the labour market in different ways, it also influences the sectors of the labour market in different ways. Productivity gains due to automation are smaller in the services and public sectors because services are more difficult to automate than manufacturing products. This makes services relatively expensive compared to products. In addition, the demand for services is more flexible than the demand for products. As prosperity has increased, people have started spending more of their income on services. Because the public sector mainly consists of services, it will form an increasingly larger share of the gross national product. As a result, economic growth levels off and the public sector becomes unaffordable (Baumol & Bowen, 1966). A solution to the problem would be further privatisation of the public sector. But as with the effects as mentioned earlier, this will increase inequality because not everyone will be able to afford the same level of services. 

Conclusion

The self-learning robot will likely have a major impact on the labour market as we know it today. The extent to which this impact will be harmful depends on how we deal with the entry of the self-learning robot into the labour market. The biggest challenge will be to set up the system in such a way that this new technology serves the interests of the owner and the employees who will become unemployed due to the arrival of the self-learning robot. In addition, it must be ensured that the costs of the public sector do not become so great that economic growth is curbed.