"They'll take our jobs". The future of labour in the light of automation due to artificial intelligence and robotics

Table of contents

Table of contents

1. Introduction

2. Which jobs are automatable and why?

3. Possible economic problems of automation

4. Solution approaches to the arising problems of automation

5. Conclusion

References

ITM Checklist

They’ll take our jobs: The future of labour in the light of automation due to artificial intelligence and robotics

1. Introduction

German newsletter articles in 2018 fanned fear in many employees with headlines like ״Siri, machst du uns bald alle arbeitslos?“ (Kramer, 2018) or ״Wird jeder Zehnte ar­beitslos?“ (Rudžio, 2018). The worry over joblessness due to industrialisation is not a new occurrence (Frey & Osborne, 2013, pp. 5-14). Although vast technical and indus­trial development should be pursued recklessly in order to make bigger profits that should increase wealth for all (Smith, 1776), there have been movements against this development throughout history (Schumpeter, 1962). Reasons for these dissenting voices can be for example: social (Ricardo, 1819), ecological (Foster, 2002) or even economic to maintain the status quo (Acemoglu & Robinson, 2012, pp. 182-185). The destruction of mechanical looms from 1811 by Ned Ludd and his followers show, how people have been aware of the effects of technology on employment (Brynjolfsson & McAfee, 2011, p. 72). Even though there are a plenty of reasons to work against an in­dustrial revolution, e.g. pollution or change of the environment, a lot of them are rooted in a fear of unemployment (Aghion & Howitt, 1994). It has been shown, that unem­ployment leads to an increase of criminality (Raphael & Winter-Ebmer, 1998), decrea­ses mental and physical health (Linn, Sandifier, & Stein, 1985) and has a negative im­pact on individual happiness (Clark & Oswald, 1994). If the assumption that “[...] about 47 percent of total US employment is at risk.” (Frey & Osborne, 2013, p. 1) is correct, employments and therefore society is facing a massive disruption.

This assignment is focused on social and macroeconomic outcomes from the latest technological trend, and highly discussed topic in media: Artificial Intelligence based Automation. Following the introduction it is primarily discussed, which kinds of jobs are potentially threatened by automation. Second, this assignment illustrate the econo­mie problems arising with automation. The third part is a brief summarisation of ap­proaches on how to cope with the problems caused by automation in the near future.

2. Which jobs are automatable and why?

Autor, Levy and Mumane (2003) argue that computers can be programmed to follow a certain set of rules to substitute human activities.

"A task is ‘routine‘ if it can be accomplished by machines following explicit programmed rules. Many manual tasks that workers used to perform, such as monitoring the temperature of a steel finishing line or moving a windshield into place on an assembly line, fit this description. Because these tasks require me­thodical repetition of an unwavering procedure, they can be exhaustively speci- tied with programmed instructions and performed by machines.“ (Autor et ak, 2003, p.1283)

These findings are in line with those by Goos, Manning and Salomons (2009, p. 16) who define routine tasks as "[...] those which computers can perform with relative ease, such as jobs that require the input of repetitive physical strength or motion, as well as jobs requiring repetitive and non-complex cognitive skills.“ Autor et al. (2003) assumed it would be easier for a computer to substitute routine tasks than non-routine tasks. Even though the study is already 15 years old the author’s assumption is still correct: Every program, including Artificial Intelligences, need some sort of an input to start working. This input has to be given by an operator who has to have an idea of what kind of task the computer should be performing (Brynjolfsson & McAfee, 2011, p. 48). Since Ma­chine Learning Algorithms need a lot of data, that was simply not utilisable until the past few years, computerisation was used to fulfil routine tasks that can be described by a finite set of rules (Autor & Dom, 2012). In order to be efficient it only made sense to write a program to fulfil a repetitive task so the computer can outplay a human. Folio­wing this logic and the stated definitions of Autor et al. (2003) and Goos et al. (2009) leads to the assumption: As soon it is possible to automate every part of a certain job, this job will become obsolete, if the automation process is cheaper.

Most recently, Machine Learning and other sub-fields of Artificial Intelligence like Deep Learning or Data Mining allow intellectual task to be computerised as well (Bryn- jolfsson & McAfee, 2011). Therefore it is possible for machines to also perform non- routine tasks. Still these algorithms follow a certain set of rules and can only be descri­bed as a so called weak AI (Russel & Norvig, 2016, p. 1020-1033). Thus they are not able to think or act humanly yet. On the other hand, a lot of jobs do not require the full potential of human acting or thinking even if they are non-routine jobs and demand a high education. Following Frey and Osborne (2013, p. 16-22) an extensive variety of non-routine tasks is getting computerised in the near future. They see two main reasons for this:

1. Big data makes a lot of information available that has been concealed years ago, ma­king decisions made by a Machine Learning Algorithm much more sophisticated. These algorithms can therefore make better decisions as they take much more information into account than a human brain could do. Assuming a human would try to take big data into account there is a lot of potential for careless mistakes.

2. Decisions made by an Artificial Intelligence are not biased like human ones. While human decision-making is influenced by a variety of environmental factors like fatigue (Danziger, Levav, & Avnaim-Pesso, 2011), a computer makes a decision regardless of its form of the day.

Given these two advantages that Artificial Intelligence has over humans, it is only rea­sonable that a lot of tasks that used to be done by professionals got automated. In health care millions of patient records can be taken into account to develop an optimal treat­ment plan (Cohn, 2013). In legal and financial services algorithms are used to scan thousands of documents within a few hours to assist in research (Markoff, 2011). There are others examples where computers are able to perform human labour like in trading (Mims, 2010) or education (Simonite, 2013). Not only cognitive but also tasks that are associated with physical labour has been partly substituted by computers. Industrial ro­bots maintaining wind turbines, operating logistics and agriculture already have a great impact upon employment (Frey & Osborne, 2013). Following the study of Frey & Os­borne (2013) in which they examined 702 occupations in the United States the most endangered jobs are those in transportation and logistics, together with some office and administrative support and production jobs.

3. Possible economic problems of automation

A meta-analysis of the most acknowledged papers showed that there are two positions regarding to the possibility of big joblessness caused by automation. Those who claim, joblessness is going to be high and those who think automation will only marginally effect employment. The following section is going to explain both sides and their argu­mentation.

Following Aghion and Howitt (1994), from a macroeconomic view, technological pro­gress effects employment in two ways:

1. Technology replaces labour, which means that those affected have to look for a new area of employment.

2. A progress in technology leads to more productivity and more market entries. This would have a positive effect on employment.

However, both of the mentioned ways follow the assumption, that employees are capa­ble of performing different tasks as soon as they get replaced by a machine or а сотри- ter program. This could be achieved by developing new skills and continue one’s educa­tion in order to meet expectation of another assignment (Goldin & Katz, 2009). These expectations become higher, since low-skilled Workers are of less need as the decrea­sing wages in this fields of employment indicate (Acemoglu & Autor, 2011). Brynjolfs- son and McAfee (2011) state that most of the routine jobs that require little skills have already been substituted by industrial machines, ft has been found, that advances in technology has generally lead to a higher demand for high-skilled labour (Autor et ak, 2003, p. 5; Autor, Katz, & Kearney, 2004, pp. 1-2). This implies, that low-skilled workers have to get a better education in order to fulfil the requirement of high-skilled labour. The access to higher education is limited though. First, education is costly even if education is basically offered for free like it is in Germany, the opportunity costs for not working full-time for three or even five years are enormous (Becker, 1994). Second, academic success is highly connected with cognitive capabilities like working memory (Alloway & Alloway, 2010). These limitations make it challenging for people who do not have the money or the cognitive capability to attend a university where they could be prepared for more sophisticated tasks.

Furthermore, following the current state of research, the demand for high-skilled workers is also decreasing (Beaudry, Green, & Sand, 2013). Beaudry et al. argue that around the year 2000 technological advances diminish the need of high-skilled workers and cognitive tasks that are associated with a high level of educations. Their findings indicate that digitalisation and artificial intelligence are entering the domain of cognitive labour, replacing even jobs which require a higher level of education. This assumption is in line with Frey and Osborne (2013). According to them, “[...] about 47 percent of total US employment is at risk.” (Frey & Osborne, 2013, p. 1). This includes physical as well as cognitive labour.

Scientists are discordant whether automation is the source of great joblessness in the next few years or not. Some claim that 35 percent to 59 percent of all jobs are going to get lost due to automation (Brzeski & Burk, 2015; Frey & Osborne, 2013; McKinsey Global Institute, 2011, 2013). This is a task based approach which concentrates on all the tasks that jobs include and discuss whether this specific tasks are automatable. Other studies that have an occupation based point of view differentiate more and claim that automation will take over some specific tasks but unlikely destroy large numbers of jobs (Amtz, Gregory, & Zierahn, 2016; Autor & Handel, 2013; Autor 2014, 2015). They ar­gue that most occupations consist of several tasks, some are automatable and some are not. Therefore not nearly as many jobs will become obsolete as Frey and Osborne (2013) and others are assuming. Amtz et al. (2016) found that only about 9 percent of all jobs in 21 OECD countries are likely to get completely automated in the near future. They explain the huge difference between their findings and Frey and Osborne (2013) "[...] is driven by the fact that even in occupations that Frey and Osborne considered to be in the high risk category, workers at least to some extent also perform tasks that are difficult to automate such as tasks involving face-to-face interaction.“ (Amtz et ah, 2016, p. 8). Another argument against massive joblessness is found in the possibility of shifting employment.

[...]