i
The nature of automation and the future of jobs
Copenhagen, Denmark
12 June, 2015
2014 Tesla Model S production line
1914 Fort Model T production line
Master Thesis – Master BLC, Leadership and Management Studies (cand.merc.int) Copenhagen Business School
Master’s Thesis Author: Stefan Lund
Supervisor: Dr. Battista Severgnini Character count: 128,461
ii
1 Acknowledgements
I want to thank Sidsel Rasmussen and Kristian Thorup-Kristensen for proofreading and giving valuable feedback about the statistics and general structure of the thesis.
I am also grateful for the help and support by my mother Mette C. Lund.
Also I appreciate the feedback from my supervisor Battista Severgnini and I am thankful for the time he took to help me with my thesis.
iii
2 Abstract
I dag rettes et stort fokus mod automatisering og den afledte effekt på arbejdsmarkedet i form af kompetencer og behovet for arbejdskraft. Bekymringen blandt arbejdstagere er generelt en frygt for, at robotteknologi vil overflødiggøre en meget stor del af
arbejdsstyrken i fremtiden og derved skabe arbejdsløshed. Dette modsiges af tilhængere af ny teknologi og (robot-)automatisering. De fremfører, at modstandere af den teknologiske (it-)udvikling i flere århundreder har ytret det samme synspunkt og taget fejl i
århundreder.
Specialet beskriver, om der med (nutidens) teknologiske muligheder for automatisering er sket en forandring i efterspurgte jobkompetencer fra tidligere, og om de efterspurgte kompetencer i givet fald være forskellige fra tidligere.
Opgaven bruger kilder af forskellig art til at opbygge en plausibel forståelse af, hvordan teknologien påvirker arbejdsmarkedet.
I afsnittet, som redegør for valgt litteratur, drages (del)konklusioner på baggrund af relevante kilder. Teknologiens historiske betydning for arbejdsmarkedet, samt teknologi, som forventes at få stor betydning for arbejdsmarkedet på kort sigt, gennemgås. I opgaven vurderes den teknologiske udvikling over tid og de teser, der kan uddrages heraf.
Herudover beskrives det, hvordan arbejde generelt mere og mere frakobles fra økonomisk vækst, og hvordan andelen af output fra arbejdskraft er dalende, både med hensyn til udbetalt output og produceret output.
Derefter præsenteres forfatterens egen statistiske undersøgelse af, hvordan de kompetencer, som arbejdsmarkedet fordrer hos arbejdsudøveren, ændres.
Ved hjælp af førnævnte analyser sandsynliggøres en plausibel fortolkning:
Efterspørgslen på innovative og teknologirelaterede kompetencer hos arbejdsstyrken generelt er faldende samtidig med, at de personer som stadig arbejder i disse jobs, bliver bedre og bedre betalt. En plausibel forklaring kan være, at den stigende
internationalisering og derved virksomhedernes øgede rækkevidde bevirker, at en mindre
iv
elite inden for f.eks. it-teknologi i højere grad er i stand til at være jobaktive på flere globale markeder og derved påvirke aflønningen i opadgående retning.
Derudover ses det, at et stigende antal produktionsjob automatiseres. Nuværende og tidligere ansatte inden for fremstillingsvirksomhed bevæger sig i stigende grad hen imod job, hvor de efterspurgte kompetencer bredt betragtet er ”bløde”, såsom præsentation eller salg. Disse nye ”bløde” kompetencer aflønnes på et lavere niveau end de
maskinteknologirelaterede kvalifikationer, som produktionsvirksomhederne efterspurgte tidligere.
Dette kan plausibelt have bidraget til den stagnering i lønningerne vi i de sidste tre årtier har set i USA, som er det geografiske udgangspunkt i opgaven.
v
3 Contents
1 Acknowledgements ... ii
2 Abstract ... iii
3 Contents ... v
4 Introduction ... 1
5 Research question... 3
6 Delimitations ... 4
7 Outline of structure ... 5
8 Methodology ... 6
8.1 General methodology ... 6
8.1.1 Philosophies ... 6
8.1.2 Research approaches ... 7
8.1.3 Strategies ... 8
8.1.4 Choices ... 8
8.1.5 Time horizons ... 8
8.1.6 Techniques and procedures ... 9
8.2 Specific statistical methodology ... 9
9 Literature review ... 15
10 Technological progress’ historical impact on labor ... 24
11 Technology trends today ... 34
11.1 Electronics ... 34
11.2 Information processing ... 39
12 Would jobless growth be a problem? ... 40
13 Examining the evidence ... 44
13.1 Work-output relation ... 44
vi
13.2 The recent effects of automation on the job market ... 48
13.3 The relative decline in effectiveness of labor ... 50
13.4 The relative decline in income flow to labor ... 53
13.5 Offshoring - an alternative explanation? ... 55
14 Changes in demanded work skills ... 55
14.1 Interpretation ... 60
15 Further extrapolating these trends ... 62
16 Conclusion... 65
17 Perspective ... 66
18 Bibliography ... 68
19 Appendix ... 74
19.1 Skill slopes ... 74
19.2 Price per mega pixel data ... 89
1
4 Introduction
Ken Jennings, a soft spoken software engineer competed in Jeopardy for the first time in 2004. He not only won his first contest but he crushed his opponents by a considerable margin. And as is tradition on the Jeopardy show, he returned to the next show as the champion, where again he crushed his competition. In fact he kept returning and winning to such an extent that the Jeopardy team had to change some of the rules of the game to the benefit of new players. New players were allowed to practice longer before show start and the buzzing mechanism was changed so that returning champions like Jennings would not have the opportunity to get too familiar with the timing. Still, Jennings won 74
consecutive games before losing his first match. He is the most winning Jeopardy player ever and is considered a grandmaster of Jeopardy (Jennings, 2015).
But on a cold February day in 2011 Jennings competed in a very special match. It was against another grand champion, Brad Rutter, and a computer program named Watson developed by IBM computer scientists. Watson was not developed with any knowledge or answers hardcoded into its software, but rather with the ability to scrape text and extract factual knowledge from it. Prior to the Jeopardy show, Watson had scraped Wikipedia and other online sources in their entirety.
And Watson won the game, in an event that was reminiscent of the 1996 event when IBM’s Deep Blue computer system beat the legendary chess master Kasparov at his own game. And just as the Deep Blue event marked a milestone in computational capabilities, so does Watson. Today in virtually any chess program you can play against an automated opponent and in many case set the difficulty settings so high that it can beat even the best of human players. Imagine what happens when better and faster versions of Watson are available on computers and smartphones everywhere. What happens when both the most knowledgeable doctor and the most knowledgeable lawyer is an app? It is technologies like this Watson that might very well bring about massive unemployment in the future.
2
On the other hand, the fear of technological unemployment is not a new phenomenon.
Queen Elizabeth I famously said the following to inventor William Lee when he applied for a patent for the first stocking frame knitting machine in 1589:
“Thou aimest high, Master Lee. Consider thou what the invention could do to my poor subjects. It would assuredly bring to them ruin by depriving them of employment, thus making them beggars.” (Acemoglu & Robinson, 2012)
William Lee nonetheless went on to put his invention to work, and as we now know, the job market ultimately evolved and provided us with other lines of professions to pursue.
Later these same concerns were voiced by the people of the Luddite movement, a 19th century English artisan uprising that caused the destruction of industrial textile equipment all over England. In fact some economists have gone as far as to term this whole concern
“the Luddite fallacy” (Tabarrok, 2003).
Even Maynard Keynes touched on the subject and coined the term “technological unemployment”, a phenomenon that he thought would have decreased the work week to about 15 hours by about this time (Keynes, 1930).
None of these predictions have come true so far, so why should they now? C.G.P. Grey gave a powerful rebuttal:
“Imagine a pair of horses in the early 1900s talking about technology. One worries all these new mechanical muscles will make horses unnecessary. The other reminds him that everything so far has made their lives easier -- remember all that farm work? Remember running coast-to-coast delivering mail? Remember riding into battle? All terrible. These city jobs are pretty cushy -- and with so many humans in the cities there are more jobs for horses than ever.
Even if this car thingy takes off you might say, there will be new jobs for horses we can't imagine.“ - (Grey, 2014)
As is alluded to here, the number of working horses in society peaked in the early 1900s and as most people are aware, it has been declining ever since. The notion that technology will always create new jobs for horses sounds absurd. Yet well respected economists will
3
still defend the notion of “the Luddite fallacy”, i.e. that new technology will always create new jobs for humans. Horses were made unnecessary because technology provided
cheaper sources of muscle power. Man is still necessary because we have no cheaper source of brainpower. Yet, like Ken Jennings on Jeopardy, humans may soon take second place to machines on the job market. And it is hard to feel content just assuming that technology will, as if by magic, always create new jobs that we now “cannot imagine”.
Let us try to imagine, or logically analyze even, what competitive skills, if any, humans of the future will have left to offer.
5 Research question
Given the current environment of seemingly ever increasing automation and the subsequent changes for the job market, this thesis will attempt to answer the following question:
How will automation shape the skills required on the job market?
In addition to the above research question, I would like to further explore the issue and operationalize my research question by examining these related sub questions:
a. Would economic growth without new jobs cause problems, and if so, why?
b. Automation has been destroying certain types of jobs and creating new ones for centuries. Are there any signs that it will be different now?
4
6 Delimitations
This chapter to give a short overview of the limits in scope of this thesis. First of all this thesis will look at the effects of automation in the short to medium term, but the longer into the future your predictions are, the more uncertain they become. It could perhaps be possible to make reasonable arguments for the effects of automation when looking into the far future. For such a study, one could perhaps look at what is theoretically possible to automate with technology in the long term and from that deduce that because these things can be automated, the basic laws of economy will ensure that they eventually will be. But this would require a fundamentally different approach to the one used in this thesis. This thesis will follow the trends that are visible already now and use these to extrapolate how technology and the job market will evolve in the near future.
A second delimitation is that this thesis will concern itself with the technology that has an effect on the job market. There are all sorts of interesting technologies that could arguably affect our lives in the near future, such as for example virtual reality glasses, but as these technologies do not obviously affect the job market in a substantial way, examining them will be outside the scope of this thesis.
Also this thesis will only look at how the job market will evolve with regards to technological advancements. There are certainly many other possible factors, such as increasing internationalization, global warming or demographic challenges that may or may not affect the future of how we work. But these are also not within the scope of this thesis.
This thesis will focus only on the developed world. Arguably, technology could have an equally large or perhaps even larger impact on the developing world. But ultimately developing countries are attempting to catch up with the developed world and so it is mainly in the developed world we find the technological cutting edge from where this thesis is will attempt to extrapolate into the future.
Additionally, no value judgements will be made about the impact of technology for society. This thesis will endeavor to describe how work will be impacted by technology,
5
and will only give a brief overview of other potential societal problems that may arise in the light of unemployment.
Finally, policy suggestions will not be made. (Brynjolfsson & Mcafee, 2014) and others have made what ample sound policy suggestions that cover the subject well.
7 Outline of structure
In the literature review will start by giving an overview of the different influential contributions to the topic, more or less in historical order. Next an overview of
technological progress will be given, starting at the beginning of the enlightenment and forwards, as there are important macro societal trends that can only be observed through a historical perspective. Then a short overview of which specific technological areas that seem poised to have impact on the job market and whose beginning influence and potential can be observed already today will follow. That same chapter is also the foundation for establishing a sense of why the impact of the technologies of the near future might be very different from that of the past, in terms of the amount of jobs it could render obsolete. Then it will be discussed whether technological unemployment is a problem, given continued overall economic growth. It seems rather obvious that massive unemployment could cause major societal difficulties, but this thesis will mainly assess the economic effects from this potential jobless growth.
When arguments for whether or not jobless growth would pose a problem have been made, different evidence that this economic reality looks increasingly likely in a few different ways will be examined. More specifically the following evidence will be examined:
1. Recent decoupling of labor and output
2. The relative decline in the effectiveness of labor when compared to capital 3. The decline in income paid out as wages
4. The changes in skills demanded by the job market
6
Then, this thesis will explore what happens if we extrapolate the trends established and finally end with a conclusion of the thesis.
8 Methodology
8.1 General methodology
In order to describe my general research approach, I have chosen to use the framework known as the “research onion” theory (Saunders, Lewis, & Thornhill, 2009). The image below shows an overview of the structure of the research onion.
Figure 1- A visual model of the research onion framework used in this thesis.
I will begin by describing my approach to the outermost layers of this framework and describe my thinking about each layer until I have “peeled” my way to the core.
8.1.1 Philosophies
Based on my fundamental understanding of how reality works, my philosophical approach will have to be realism, or more specifically critical realism. I believe that there is an objective truth, regardless of human thought or perception. I also believe careful
7
observation can reveal these truths, but inadequate data or misguided analysis can lead to wrongful conclusions or misconceptions. Furthermore, I believe that any researcher or person will necessarily have to be a product of his or her environment, at least to some degree. Any person will therefore always have some amount of cultural bias, which can play a role in misinterpreting experiment when the data is incomplete or inconclusive.
Moreover, I view both qualitative and quantitative research methods as having their appropriate usefulness for data gathering of different nature.
8.1.2 Research approaches
I see the inductive approach as more of an explorative one, that is very useful for building an overview of a given research inquiry. The inductive approach does not give any
certainty for its findings and no way of ruling out competing explanations. But when confronted with a novel and scientifically poorly understood phenomenon, the inductive approach is nonetheless often the best to pursue. Given for example the task of explaining different outcomes that are the result of a complex social interaction, the deductive
approach is simply often not practical. It might not even be obvious which variables to measure, or how to go about measuring them in the first place. Therefore, the inductive approach is often necessary for initial exploration and theory building of a certain
phenomenon. Then, at a later stage the deductive approach can be applied to quantifiably prove or disprove connections between variables. Additionally, I believe that while the inductive approach can be used to demonstrate the plausibility of a given theory, nothing can ultimately be concluded without some form of verifiable measurement. Of course the collection of sample data for deductive tests can often be time consuming and tedious.
This is why I do not view choosing the deductive or the inductive approach as an either/or proposition, I believe both have their place of usefulness. Often in social sciences, the amount of variables involved is so great that the inductive approach is needed initially, to provide a plausible framework of understanding. Later, a few of the likely countless variables, those with highest presumed importance or those which are available for measure, can then be quantifiably measured and in this way used to further lend
credibility (or the opposite) to the explanation in question. Also, I do not believe that any
8
serious theory can be fully outside the scope of measure. Data taking may be difficult or done in a more or a less reliable manner, but ultimately a theory that relies on
immeasurable variables is the same as a non-falsifiable theory and is therefore by its nature outside the field of science, something that is best left for philosophers and theologists to ponder.
In conclusion, I will use a mix of inductive reasoning and deductive proof to highlight the plausible causes of the phenomena examined in this thesis.
8.1.3 Strategies
Saunders, Lewis and Thronhill (2009) in the research onion framework, put forth several different strategies, but also emphasize that the list is not exhaustive and that the different strategies overlap. Of the strategies that they outline, I feel that “grounded theory” and to some extend “archival studies” best describe my research strategy. I use grounded theory in the sense that I examine evidence and attempt to build a plausible explanation from it, where after I will examine this explanation, using data. The dataset examined covers 14 years, so although the term “contemporary” fits the data better than “historical”, using this data is considered “archival studies”, as this strategy is defined in the book.
8.1.4 Choices
I view both the deductive and inductive method as useful approaches. I will therefore use a combination of these two. But at the same time, as also explained above, I do not
subscribe to the idea that firm results can be derived from pure qualitative data. Therefore my approach will ultimately be that of a “multi-method quantities study”.
8.1.5 Time horizons
This is a horizontal study . Certain trends will be examined as they have unfolded over time and from these trends I will attempt to probe how our future is likely to evolve.
9
8.1.6 Techniques and procedures
I will collect and use data from different statistical sources. Some of this data will be statistically analyzed, as further discussed in the statistical methodology chapter. Other data will simply be presented as proof or evidence of certain trends or assumptions. Some non numerical observations will also be used, again to examine the credibility of the theories I propose as explanations for the phenomena discussed in this thesis.
The theoretical frameworks I will use in this thesis will mainly be of economic nature, such as for example, macro or micro economical theory.
8.2 Specific statistical methodology
In order to examine the current trend of how professions are changing I have used data from the Occupational Information Network (O*NET), which is the United States’
primary source of occupation information, conducted by the US government (The
Occupational Information Network, 2015). The data are based on surveys of large samples of the working population of the US. The reasoning behind choosing US data over data from another country will be explained later in this thesis.
In the O*NET survey, data are collected about many different aspects of each occupation.
For example, data is collected about tasks, tools and technology, knowledge required, skills required, abilities, work activities, work context, education and more. Further datasets collected from earlier years are publically available for download. Unfortunately, some of the categories, such as “tools and technology” are for some reason not available in the databases from previous years and the authors of the database did not respond to inquiries regarding this aspect of the data categorization. I analyzed the nature of how the survey of each category has been conducted and it turns out that some categories that at first glance seem very interesting for the topic of this thesis are not that useful after all. As an example, it seems interesting to study how responses to a survey category such as
“Tasks” has evolved over time, given that the focus of this thesis is to explore the nature of change on the job market. However, the response categories for each profession are
10
different in nature. As an example, for the “New accounts clerks” profession, there is a score for how important it is to “Process loan applications”, where other professions have different data points, making it difficult to analyze data across different professions. Based on this examination I found that the best category to examine was “Skills needed”, where all professions had scores for the same general data points, such as “Mathematics” or
“Negotiation”.
O*NET databases can be found from all the way back to 1998, although the older database sets are not provided in MySQL database format. In fact, the only database format that was used consistently for all the older databases is the Visual FoxPro type that was discontinued long ago and very cumbersome to work with by modern standards. A lot of work went into converting all databases into MySQL format so the data points could be compared. When all the data had been imported it turned out that the first database from 1998, was very different from later databases as its occupational categories were not comparable (newer databases used the “SOC” occupational title system1 and this older one the “DOT” system2). In databases from after 1998 the definitions of professions have changed slightly over the years, so that a few professions were split up, merged, added or removed each year. I chose to use data from only the 625 professions that were consistent throughout the databases using the SOC system (newer databases). In addition, the skills measured had also changed slightly over the years, so again I chose only to look at the 31 that remained consistent. Finally, I discovered that “skills required” had not been updated in all databases, and I therefore discarded a few databases that did not include any new and updated skill information. In the end, I was left with data from 13 database files with date measured from the year 2000 to 2014.
Each skill has an “importance score” between 1-5 and a “level score” between 0-7. These two measurements at first glance seem similar, but the O*NET documentation explains their difference with an example:
1 United States Department of Labor, http://www.oalj.dol.gov/libdot.htm
2 US Bureau of Labor Statistics, http://www.bls.gov/soc/
11
‘The skill "speaking" is important for both lawyers and paralegals. However, lawyers (who frequently argue cases before judges and juries) are required to have a higher Level of speaking skill, while paralegals only need an average Level of this skill.’
To simplify my data I multiplied these two scores to get a “full score” that fell between 0- 35. The reason for multiplying instead of for example adding these two scores, is that every unit of “level score” is affected by the “importance score”. In this way a level 2 score with 5 importance should get a score double that of a level 1 score with the same importance. Obviously you could have done this in a multitude of other ways.
I then wrote a small program that calculated the average “full score” of each skill for all professions for every database, giving average “total score” for mathematics, negotiation and so on for each database. The computational time was more than 24 hours for running through the more than 20 million records in the skill categories for all the 13 relevant databases. As each database represents a time point, I was in this way able to establish how the need for a skill such as “Reading comprehension” has changed over time.
Figure 2- The importance of reading comprehension in jobs over time. The data is collected by survey and the score range is between 0-35.
0,0 2,0 4,0 6,0 8,0 10,0 12,0 14,0 16,0 18,0 20,0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Total Score
Year
12
As can be seen from the graph in Figure 2, there is a strange “bump” in the middle of the graph. In fact this feature turned out for most of the skills. The average curve for all the skills looks like this:
Figure 3- The average importance of all measured skills used in jobs over time. The data is collected by survey and the score range is between 0-35.
At first glance it seems strange that such a bump should appear on these curves, when no big changes in survey methods were introduced. But looking at the dates, it is clear that the “crash” of the curve corresponds to the beginning of the financial crisis. In fact, the last data point from the top of the curve corresponds to the database released in June 2008, three months before the crash of the stock market. The curve then looks to decline slower than you would expect if it had any relation to the mood of the world economy, but this is somewhat misleading, because the next data point is actually not before June 2010. So if there was in fact a sudden crash in the skill curve, it would not be recorded here. One likely explanation for this graph could then be that people’s perception of how important their job and skills are, corresponds to the general financial mood of the time.
Investigating that proposition is of course outside the scope of this thesis, but could perhaps make for a research question all by itself. Regardless of the underlying cause of this effect, I normalized the data, by weighing the skill averages such that the value for each dataset was adjusted for the overall average at that time.
0 2 4 6 8 10 12
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Total Score
Year
13
I then tried different regression methods to see which one best fit these adjusted curves.
Although one might expect such societal macro trends to develop in an exponential way, the resulting best fitting curves simply did not justify picking an exponential regression over a linear one. Perhaps the 14 year span of the data is not long enough to show whether the underlying relationship is linear or exponential in nature. I then did a significance test to check which slopes where significant at an at least 95% confidence level (P < 0.05), finding that 23 of the 31 skill slopes were significant. These results will be presented later in the thesis.
After having calculated the slopes for the changes in skill needs I wanted to see how well these data could predict actual employment data. Luckily, the US Bureau of Labor
Statistics (BLS) uses the same classification of jobs (SOC) as the O*NET surveys. The oldest (SOC classified) employment data I was able to find at BLS was from 2003 and the newest from 2013. First I compared the 2003 and 2013 employment database with the 2013 skill database to see which employment categories were consistent across all of them. For each of the resulting consistent 473 job titles, I computed how important each skill was as compared to the overall amount of skills needed for that job. For example, lawyers might need 20% speaking, 10% presentation, 10% listening etc. and all
percentages would add up to 100% (example data used here, not actual data). The
importance for each of the skills of the professions could then be used to weigh the impact of each of the overall skill slopes for this profession. For example, if speaking makes up 20% in importance of the total required skills for a lawyer and the overall trend or slope of speaking is such that the importance of speaking is set to halve over a certain period, then it would lawyers 20% importance of speaking will be reduced by 10% points, making their total importance reduce to 90% points, accounting for this one skill alone. Of course all skills must be accounted for in each profession. If the final result after accounting for all the skills is below 100, then it would signify an overall decrease in the total
“importance score”, of this profession. Conversely, if the final result is above 100 it would signify an increase in importance score (since the initial score is always 100).
14
Assigning each of the 31 skills a number (i), this can be expressed mathematically for each profession as such:
𝑖𝑚𝑝𝑜𝑟𝑡𝑎𝑛𝑐𝑒𝑛𝑒𝑤 = ∑(𝑤𝑒𝑖𝑔ℎ𝑡𝑖∗ 𝑠𝑙𝑜𝑝𝑒𝑖∗ ∆𝑦)
31
𝑖=1
Here ∆𝑦 denotes change in years (since the slope was calculated in terms of years).
𝑤𝑒𝑖𝑔ℎ𝑡𝑖 denotes the weight of skill 𝑖, as calculated by how many percent skill 𝑖 made up of the total initial importance score. 𝑠𝑙𝑜𝑝𝑒𝑖 would signify the best fitting slope of skill 𝑖.
Note that 𝑤𝑒𝑖𝑔ℎ𝑡𝑖 here is different for each profession, but that 𝑠𝑙𝑜𝑝𝑒𝑖 is the same across professions. Also all 31 skill categories are used. The logic behind this was that even though 8 of these slopes were not significant, the slope that was calculated for them is still the best guess at a slope, better than guessing their slope as being 0, which would have been the effect of omitting them.
Naturally it would be interesting to see if this resulting 𝑖𝑚𝑝𝑜𝑟𝑡𝑎𝑛𝑐𝑒𝑛𝑒𝑤 had any
predictive capabilities, and to do so, a correlation with actual numbers would have to be shown.
To do this, the numbers for the 2003 and 2013 employment databases were first compared to determine how much the number of employed people and the wage had changed for each job over these 10 years for each of the 473 job titles. I then calculated the expected skill importance change over the same period, using the formula and method described above as well as data from the 2013 database.
I then calculated the correlation between first employment change and skill importance change and then wage change and skill importance change.
Table 1- Calculated importance change correlated with skill change Correlation coefficient from
comparing with skills
P-Value
Employment 0.25 5*10-8
Wage 0.10 0.03
15
As can be seen here, both correlations were significant at a P < 0.05 level. Of course causality cannot be assumed either way, but these results increase the likelihood that changing skill needs can in part be a reliable predictor for the future of jobs needed.
It is interesting, and maybe a bit surprising, to note how much of employment can be explained by using the skill data alone (25%). But regardless, it is also important to note that a large part of employment and wage must be explained by other factors (75% and 90% respectively).
In carrying out my statistical analysis, I have been forced to make several decisions that certainly have the potential to influence the results. Based on time constraints I have discarded occupational data that was inconsistent because it used a different classification system (the 1998 DOT database) and data that was inconsistent because of the minor occupational redefining that occurred over the years. I also made a choice in how to calculate a “full score” from level and importance and to use a linear regression instead of an exponential one. I also chose to use slopes that were insignificant instead of discarding them, because they represented the best estimation available nonetheless. Common for all decisions were that I made them based on what I saw as the best solution before
calculating and not retrospectively and then selecting the solution yielding the most interesting result. Doing that would have introduced a selection bias towards interesting results which would have constituted a major flaw in the study.
9 Literature review
In this literature review, I will discuss major contributions to the debate about the destruction of jobs by automation, in a more or less chronological order and attempt to show what each paper or book contributed to this particular field. More recent works will receive a more detailed description as their material is of course more up to date.
16
As mentioned in the introduction, as far back as in the paper “In Economic Possibilities for our Grandchildren” Keynes (1930) describes technological unemployment:
“We are being afflicted with a new disease of which some readers may not yet have heard the name, but of which they will hear a great deal in the years to come – namely,
technological unemployment.”
The great depression that was already ravaging the US job market in 1930 when Keynes wrote his paper might have influenced his view, although he correctly predicted that the overall crisis was temporary. All said, Keynes in this paper makes a surprising large number of predictions s that have come true, from the continued exponential growth of wealth to the end of the struggle for subsistence in the rich world. But he also predicted that by 2030, work would be rarer to get and that people will share what work is left until we see 15 hour work weeks. The 100 years have not yet passed, and although the work week shortened drastically in the US for a few decades after Keynes wrote this paper, it has since then remained virtually unchanged and there are no signs that this is set to change during the next decade and a half, so this prediction is unlikely to become reality.
Fisher (1939) among other influential economic works of the time is widely attributed with first formulating the three-sector economic theory of growth. The theory divides production into three sectors based on how far down the production chain its output is.
The model of how growth affects these three sectors with technological development (being primary/raw materials, secondary/manufacturing and tertiary/services) was based on empirical observations. Clark (1940) suggested a fourth sector (quaternary/innovation services) and while this more in-depth model is not as widely used today, it could become more dominant in the future if its implications are true:
17
Figure 4 – Idealized version of Clark’s quaternary model of sector employment over time.
As the figure shows, employment in raw materials/agriculture, is dominant in primitive societies, but then drops as countries industrialize and manufacturing rises. Later manufacturing peaks and start dropping as societies transition to one dominated by services, some of which could increasingly be innovation services (quaternary). The validity of this notion shall later be probed.
Much later Shaiken (1984) contributes an analysis of how automation leads to
“deskilling” of the workforce. He talks about how new technology has taken “redesign”
out of the hands of shop floor machinists at a General Motors plant. Just as with the skills of woodworkers, blacksmiths and other artisans before them, the machinists must accept to have their individual creative control be superseded by top down designs. He also describes how computerized automation helped break the US traffic controllers’ strike of 1981 and ultimately made 12.000 people redundant. His work is important in emphasizing that it is not always just the low skilled jobs that are made redundant by automation, as the common narrative in our modern media and public debate would otherwise have you believe.
18
Rifkin (Rifkin, 1995) is one of the first more serious authors to write about technological unemployment in a much more urgent tone. At this point global unemployment is at its highest since the great depression. Also the emergence of computers and
telecommunications promise to disrupt manufacturing, retail and transportation, just as it does today. Rifkin talks about how society is facing a fundamental shift and a challenge in redefining the role of individuals in a society without work.
In retrospect, it is easy to dismiss Rifkin’s predictions as premature, but many factors could potentially have dampened the effect he predicted that we would see by now, such as the growth spurred by the inclusion of new regions into the world economy, like Eastern Europe’s integration into the Eurozone, or Chinas acceptance into the World Trade Organization.
Murnane and Levy (2004) discuss “information labor”, the type of job that is susceptible to automation by computers. The authors discuss two types of these jobs, one type that is likely to be fully automated by computers, creating “deskilling” and one type that will likely merely augment the information worker and increases his productivity with tools and thereby “upskills” him. The authors go on to discuss which skills can and cannot be automated and describes how the effect of the previously mentioned two types of jobs will create a “digital divide” that can only be addressed by public policy and educational reform. The authors also claim that the rapid change in the job market creates a higher demand for verbal and quantitative literacy. They back these claims up with historical data showing that a 30 year old college educated man in the US today earns 50% more than a 30 year old man with only a high school diploma. 25 years prior the difference was only 17%.
One major critique which is now becoming more obvious is that the authors’
presumptions of what a computer can and cannot do is flawed. They claim, presumably correctly, that for a computer to carry out an action it must be possible to break the action down into small tasks of the sort that a computer programmer can instruct it to do. But from this they wrongly deduct that computers therefore cannot carry out optical, tactile or highly complex tasks. Now in retrospect, we know that very complex computer programs
19
with hundreds of thousands or millions of lines of code have been developed, programs with capabilities that even some researchers, such as the authors of this book, thought impossible.
A good example of this is when the authors state that:
“But executing a left turn against oncoming traffic involves so many factors that it is hard to imagine discovering the set of rules that can replicate a driver’s behavior”. It was only 6 years later, in 2010 that Google introduced their prototype of a self-driving car that after thousands of kilometers of driving was shown statistically to be a much safer driver than its human counterparts.
Cowen (Cowen, 2011) gives a much more pessimistic outlook in the book “The great stagnation”. One of the many interesting observations in the book is that there are two kinds of economic growth. One is what I would call “catch up growth”, which is when poorer countries grow by adapting already invented technology imported from more advanced countries. The other kind is actual innovation driven growth.
The rich world has been doing the second type of innovation for the most part of the last century and lately they have not been doing it too well. The author claims that the rich world has been growing lately mainly by picking other low hanging fruits, but that these low hanging fruits have now mostly been picked. For example in the US, a much higher percentage of people receive higher education today than was true in the past, which presumably has led to some of the growth we have seen. Of course, levels of education in the workforce can be increased even more and arguably should, but the author argues, that it will get increasingly harder to get marginal students through collage, which nearly half of all young people already attend.
Historically, America has been able to harvest a lot of low hanging fruit from exploiting the vast amount of undeveloped land, and by accepting droves of European immigrants.
Today land is no longer as abundant. If you want to develop a plot of land, chances are that you have to convert land that already serves a productive function.
20
But the largest setback for the rich world according to the author, is the end of extra gains to be extracted from some of the major innovations of the past - innovations in industry, chemistry and electricity. People born in the 30s or 40s will have experienced drastic changes to their lifestyle as technology in the household has changed completely. This generation has for example enjoyed the advent of the consumer electrical appliances such as refrigerators, dishwashers, laundry machines, electric lights, radios, air conditions, etc.
But other than perhaps the introduction of the computer, today’s home looks about the same as it did 40 years ago. So as technology has improved, radical new technologies are much rarer and technological progress therefore seems to have stalled, is the argument.
And while the internet has had great and positive impact on society, it has not produced much in terms of economic growth or jobs. Many of the uses and services offered by the internet are free and therefore do not contribute to GDP, while the internet has made a lot of workers redundant.
While this book has many good insights and is certainly a good contribution to the discussion, I disagree with it in a few ways. Firstly, I do not agree that technology has stagnated and I do not agree to his arguments about this. Regarding his examples with household appliances, I see it more as the time spent doing housework has been
exponentially decreasing with technological progress. Take the robotic vacuum cleaner for example; sure the move from a regular vacuum cleaner to this is not as big a time saver as going from no technical help to the vacuum cleaner in the first place. But if we expected a linear reduction in time spent on household tasks, we should already have hit negative numbers. Therefore the nature of how the time spent doing housework decreases, would more resemble an over time exponentially decreasing graph, approaching, but never quite reaching zero. Additionally, the most impacting advancements in technology in modern times are likely found elsewhere, as for example in the computer that he also mentions.
Also, it is ironic to how the author can acknowledge the positive utility of the internet and all the great services it has brought with it, but belittle its importance solely because it is not captured in GDP growth. It seems that this is more an example of the shortcomings of the GDP measure. If technology makes a product or service 10 times cheaper, this
21
development is captured in inflation and therefore adjusted for in real GDP numbers. But when technology invents a new service or product that is free, it is not accounted for in real GDP, regardless of how big or small its utility might be.
The diametrically opposite view of technological change is argued for in “Race Against the Machine” (McAfee & Brynjolfsson, 2011). They talk about the nature of change, especially in the information and communication technologies area. They explain that like electricity, these information and communication technologies are general purpose tools, with the ability to increase production across just about all sectors of the economy. They also make a point of the incredible pace at which this sort of technology develops. As implied by Moore’s law, technological progress has been compounding. The authors point out how compound growth does not seem too overwhelming before it seemingly suddenly grows out of control. This phenomenon is well exemplified in the book with the story of the king who wants to reward his loyal subject for inventing chess and agrees to give him 1 grain of wheat for the first square on the chessboard, 2 grains of wheat for the next, then 4 and so on. This seems very underwhelming on the first half of the chessboard (up to 232), but the amount of grain grows dramatically from there. By the time you reach the 64th and last square, the total number of grains exceeds the number of grains ever produced in human history.
The implications of this exponential growth example are of course that we could be on the verge of a large disruption of technological change. As a result, society may enter a future of chronic unemployment, more so perhaps for some type of laborers than others. The authors therefore recommend education reform and increased educational investment as well as focusing on organizational innovation and making it easier for entrepreneurs to succeed. Entrepreneurs are needed to exploit the new possibilities that technological advances create. This could, for example, mean creating sensible immigration laws for skilled entrepreneurs, clearing legal hurdles and reforming patent laws.
22
Like this thesis, Frey and Osborne (2013) examineO*NET data and use this to estimate how great a percentage of US jobs are at risk of being automated. They divide each job task into three categories: “social intelligence”, “creativity” and “perception and
manipulation” and link these to different 2010 O*NET tasks, skills and abilities. They subjectively rate these tasks, skills and abilities relation with those three categories. They then look at the connection between each job and its connection to those tasks, skills and abilities, and using the before mentioned ratings they calculate a “risk of automation”
score. Finally they overwrite the scores for 70 out of their 702 occupations with subjective scores for jobs which future they feel they have a firm understanding of. The paper
concludes that about 47% of US jobs are at risk of being automated. Although this paper has been cited widely and featured in many articles and other media, the study seems fundamental flawed in that it relies so heavily on subjective opinion. Therefore this paper is perhaps better understood as more of an expert opinion than a scientific study, the study cannot be reliably reproduced.
Additionally, calculating a “risk of automation” score implies that the outcome is
necessarily binary. Either automation renders a certain job obsolete or it does not. I would argue that this premise is flawed since technology and tools will often automate a certain subset of the tasks or processes of a profession, rather than always automating everything completely or nothing at all. Take construction engineers as an example. In contrast to construction engineers 50 years ago, their modern counterparts have 3D modelling software with earthquake and aerodynamic simulations and so on available. Even the invention of something like the computerized spreadsheet probably automated a lot of their calculation work. Given all these tools, modern engineers are much more productive than earlier ones, which means that we would have needed a lot more engineers in today’s society had these tools not been invented. In other words, even if the engineering
profession is very much alive and well today, large parts of what it historically meant to be an engineer has already been automated. The other side of this is of course that since every engineer is now much more productive, their pay has likely also increased. Further, in the particular example of the engineer, new technology has of course also opened up new tasks for the engineer to work on.
23
Pikkety (Piketty, Capital in the Twenty-First Century, 2014) sparked much debate in economic circles with the book “Capital in the twenty-first century”. The book itself became a bestseller and is highly regarded by many economists. Even the influential magazine “The Economist”, has proclaimed Piketty “the modern Carl Marx”. Some think the book will herald a shift in focus for economists towards inequality, others see flaws in its logic.
The topic of the book itself is on the relationship between capital and inequality and is based on a massive amount of data that Piketty and other economists have compiled over the last decade.
Piketty argues that throughout recent history the ratio of private wealth to national income has been very high. Historically, a few rich upper class individuals or families owned most of the wealth and society was very unequal. This changed due to the chaos of the first and second world war as well as the great depression, and society became for a time much more equal, and wealth as a fraction of national income plummeted. But now this sort of inequality is on the rise again, maybe on its way back to its historic precedence. In general Piketty also claims that interest on wealth (r) will generally be greater than general economic growth (g):
𝑟 > 𝑔
But Piketty also states that periods with high growth can close the gap between the two somewhat.
In any case, as much praise there is for his work, there is also a lot of criticism of it.
I hope I have here given a satisfying overview of the most influential works on this topic and clarified how each work has contributed to the ongoing debates and investigations. To
24
my knowledge there is no literature that specifically addresses the subject of this thesis, that is, examining what specific skills, if any, workers of the future can rely on to earn a wage.
10 Technological progress’ historical impact on labor
In 1405 the Chinese admiral Zheng He left Suzhou with what was likely the largest fleet the world had ever seen. The largest ship of the several hundred ship large fleet was 134 meters long, almost the size of a modern aircraft carrier (Tamura, 1997). The fleet went on several voyages reaching as far as the Middle East and east Africa.
But after a regime change and only 7 voyages, the fleet was dismantled and the
knowledge of how to construct these huge ships was forgotten. This backs the argument Spinney (2012) makes that technology, seen over a long time period, will not necessarily follow an upward sloping exponential curve. She mentions other examples to back her claim, for example that during the excavation of Pompeii in the 18th century, an advanced aqueduct system was discovered that was superior to anything made at the time. And there was evidence from this phenomenon even in the very distant past, as new archeological findings at Howieson's Poort Shelter in South Africa showed evidence of. People there seemed to be making very sophisticated stone tools until about 60.000 years ago, when the people living in the area for unknown reasons abruptly stopped and went back to making much simpler tools.
But even Spinney ultimately confesses that she thinks we at this point are unlikely to lose the knowledge we have accumulated and return to previous levels of more primitive technology.
Another and, perhaps more influential view of pre-industrialization technology and
growth was first outlined by the British economist Thomas Malthus around the turn of the 19th century (Malthus, 1798). Malthus argued that the increase of population is limited
25
primarily by the means of subsistence and that given the means to increase, populations would inevitably do so. Because of this, technological progress only served to increase population, not wages.
Additionally, this theory implies that there is always bound to be what I would, for the lack of a better term, call a “marginal population”, living in poverty and subsistence. It was ironic, however, that Malthus wrote about this observation just as the foundation for it was about to change. Today the observations Malthus made are the basis for what is now known as the “Malthusian Trap” and the “Malthusian era” that ended just around the time when he made his observations. Perhaps the one graph that best captures the essence of early automation and production history is shown in Figure 5 below. Notice how wages had almost been stagnant for a century when Malthus wrote his observations (around 1800).
26
Figure 5 – The relationship between real wages and population in England 1280s-1860s.
(Clark G. , The Condition of the Working-Class in England, 1209-2003, 2004)
Historians do not quite agree, but the (first) industrial revolution is generally thought to have taken place in the time period 1760 to 1830, but as can be seen in the graph above, something seems to have changed fundamentally, starting around 1590. This corresponds well with when, William Lee invented the stocking frame knitting machine in 1589. When he was denied a patent on his invention by Queen Elizabeth I, he improved his machine so that it had 20 needles per inch instead of 8 and therefore made much finer thread
(Acemoglu & Robinson, 2012). He went back to the queen who then refused him again, citing the fears of the textile workers losing their jobs as explained in the introduction.
But William Lee’s invention marks the start of an era of continuous improvements in textile production, a type of production that made Britain very rich, but at the same time made a lot of skilled artisans obsolete. Other important industrial disruptions were made in the production of iron beginning in the late 1600s. Production methods were invented
27
that utilized coal instead of charcoal. This was important because it moved iron production out of the woods, from where it was very costly to make and to transport.
Official opinion on the importance of industry had changed so drastically during the roughly two centuries since Queen Elizabeth I denied William Lee this patent, that the parliament (having gained supremacy over the crown in 1688) passed a bill in 1769 making the destruction of machinery punishable by death (Mokyr, 1990).
By the late 1700s, Britain’s agricultural revolution had begun. One of the big factors to this was the use of crop rotations, planting turnips and clover in between harvest instead of leaving the ground fallow (Sturgess, 1966). As turnips can grow in winter and have deep roots, they could gather deep minerals that are not reachable for other types of crops.
Additionally, clover pulls nitrogen from the air, in effect acting as an early form of
fertilizer. This allowed higher yields and enabled the cultivation of otherwise poorer soils.
Together with land conversions and drainage, an increase in enclosed farm lands and improved roadways meant that productivity and food production were rising fast. This in turn made food production increase faster than population and more and more people started moving into the cities. Many scholars believe this is likely one of the most important factors that helped Britain increase its production fast enough that population growth could not catch up, and thereby escape the Malthusian trap.
Another innovation that completely revolutionized production in Britain towards the end of the nineteenth century was the steam engine. The power potential of coal powered steam engines was many times that of the animal, water and wind power that Britain had previously relied on. Of the many great advances in steam power, one of the most
impactful was made by the Scotsman James Watt (who, due to his scientific observations, has the unit of power named after him). He made a series of technical improvements, by for example introducing the use of a steam jacked and the introduction of the condenser chamber. The development and deployment of steam power had a profound effect on work in the late 1700s and onwards. It replaced a considerable amount of manual labor and made possible factories on a much larger scale.
28
The impact of coal can maybe best be understood by observing the change in the total energy consumption of England and Wales as shown below. Energy consumption seems to reach an inflection point around 1840, from where power consumption started
increasing at a rapid rate and which has not slowed down since.
Figure 6 - Energy production ocer time by source in England and Wales.
(Warde, 2007)
By the beginning of the nineteenth century, the skills of craftsmen were becoming obsolete as workers were moving into the factories.
The feeling about the conditions of the factories and their low wages as compared to skilled artisan work was later summed up by influential Victorian writer Charlotte Elizabeth Tonna, as such:
“The factory system is one of the worst and cruelest things ever invented to pamper the rich at the expense of the poor. It fattens them, and melts the flesh off our bones: it clothes them in grand raiment, and bids us shiver in rags: it brings all indulgences within their reach, and kills the industrious creatures whose toil provides them.” (Tonna, 1841)
0 2000 4000 6000 8000 10000 12000
1560 1580 1600 1620 1640 1660 1680 1700 1720 1740 1760 1780 1800 1820 1840 1860 1880 1900 1920 1940 1960 1980 2000
Energy (Peta Joules)
Year
Work Animals Firewood Wind Water Coal Oil Gas Primary Electricity
29
The tough economic conditions for workers and unemployed artisans ultimately caused riots by textile workers that were collectively known as the “Luddites”, and the riots spread throughout England between 1811 and 1816 (Beckett, 2015).
It might not be obvious today, but in the nineteenth century, textile manufacturing was a major industry.
The spark that first ignited the Luddite riots was Parliament revoking an old law
prohibiting the use of a certain type of mill used in the wool finishing trade. During the five years of Luddite riots and sabotage, they destroyed stocking frames, spinning frames and power looms. Parliament took a harsh stance against the uprising deploying as many as 12.000 soldiers (Mantoux, 2006).
Another popular worker uprising known as the “swing riots”, was sparked by
unemployment and the fifty years of gradual impoverishment of the agricultural workers leading up to 1830 (Harrison, 1984). The conditions of the English laborer were so poor that Lord Carnarvon said in parliament that:
“The English laborer is reduced to a plight more abject than that of any race in Europe, with their employers no longer able to feed and employ them” (Hammond & Hammond, 1912).
The rioters especially targeted and destroyed powered threshing machines, a type of machine that would eventually replace as much as 30% of the agricultural workforce over the course of the nineteenth century
(Brynjolfsson, 2013). Nineteenth century horse powered threshing machines.
30
By late 1800s we enter what some historians refer to as the second industrial revolution, mostly driven by improved steel production methods, chemicals and petroleum refining and distribution.
Petroleum had great implications for the transport industry and it has helped make transportation what it is today: one of the largest employers of the economy. Electricity further increased factory productivity and towards the end of the 1800s created a large array of consumer products. With the increasing move from artisan jobs being converted into factory jobs, the nineteenth century was also the century that saw the formation of the first labor unions, although these were not formally legalized in Britain until 1871. The early 20th century saw innovation in new management techniques which influenced the roles of workers. Especially important was the introduction of the assembly line
manufacturing, which divided workers’ tasks into a few very repetitive processes. This is well illustrated with the advent of the mass production of Ford Motors Model T,
commonly referred to as the first affordable motor car. By 1913 Ford Motors, with the new highly segmented division of labor and the world’s first moving assembly line at their Highland Park facility, was able to manufacture a Model T in only 93 minutes and sell it for an affordable $575. By the next year Ford had captured about half of the automobile industry, and by 1920 the factory churned out a car every minute, and half of the world’s cars were Model T’s. (Detroit National Park Service, 2015).
This highly monotonous form of labor was further spread by the outbreak of the First World War which brought about the need for efficient munitions production. The First World War had a profound effect on innovation and the labor markets in general. As many of the great nations of the western hemisphere mobilized and man power became scarce, the collective bargaining power of unions grew and women started entering the workforce. Industrial production also rapidly shot up to feed the growing demands of the war. For all its horrors, the First World War spurred lots of inventions in fields ranging from stainless steel to aircraft radio communication.
By the onset of the Second World War, the US had unambiguously taken the lead as the world’s largest superpower from the UK. It was the Second World War that lamented the importance of tanks and aircrafts, and the US was able to churn these out in much greater
31
numbers than any other country. To stir industry towards military purposes, production of goods like cars, home appliances and housing was banned until the end of the war, and meat, clothing and petrol were tightly rationed (Schneider & Schneider, 2003). The war and the need for manpower also further increased women’s move into the workforce and reduced unemployment to historic lows. And as with the First World War before it, the Second World War spurred investment in research and brought about a host of
innovations, some of which would greatly influence the rest of the century. Among the most influential technological advancements were penicillin, radio navigation, radar, the invention of rocketry, the jet engine, nuclear power and the first primitive computer.
And especially the computer´s influence on the second half of the 20th century, and indeed up until now, can hardly be overstated. It has
ushered in a period of increasingly sophisticated automation and a movement in the workforce towards the information industry. Indeed, the advent of the industrial robots and other computerized technology meant that for the first time in centuries, workers started moving out of factories. The graph below shows how in only 25 years US manufacturing jobs roughly halved while production has stayed roughly unchanged.
1941-1945 US war poster
32
Figure 7 – US Manufacturing employment and output comparison.
The graph plots manufacturing output against employment in the US. (Gavin, 2013)
The latter half of the 20th century also brought the liberalization and opening up of Eastern Europe and China, who with their low wages created the perfect vacuum to suck in
massive amounts of low skilled manufacturing jobs from the developed world.
My interpretations of the history of technological advances and its consequences for workers are twofold. Firstly, I think the narrative often presented by politicians that education and knowledge are the workers’ defense against automation might not be true.
At least this notion is not backed by historical evidence, as the unfortunate fate of the skilled woodworker, weaver and blacksmiths of the last few centuries illustrates. The industrialization and move into factories in the developed world destroyed and deskilled the whole artisan production economy, even as new technologies created opportunities for new types of skilled labor.
Secondly, I want to sum up the broad movements that happened over the past few
centuries. In 1750s England 53% of the workforce worked in agriculture and much of the
33
remaining population worked in what we would probably classify as artisan jobs today (Clark G. , 2002) as referencd in (Lindert; 1980).
Since then agricultural employment has continually fallen, and today it makes up less than 1% of the jobs in England and Wales. At the same time manufacturing was transformed by the progressively more advanced factory industrialization. These factors were so great at displacing labor that they caused serious unemployment and very low wages for the laborers of the early nineteenth century. These conditions led people to move to the cities, but also caused widespread riots throughout England.
Figure 8- Relative composition of workforce in Great Britain / England and Wales over time.
(UK Office for National Statistics, 2011)
The second half of the twentieth century has seen manufacturing decline, just as
agriculture did before it with the result that 80% of jobs today are in the service industry.
Using a common economic three-sector classification system as described in the literature review, we see the predicted pattern. Work in the “primary sector”, i.e. raw materials, agriculture and so on has slowly been decreasing for centuries. The “secondary sector”,
0,0 10,0 20,0 30,0 40,0 50,0 60,0 70,0 80,0 90,0
1841 1851 1861 1871 1881 1891 1901 1911 1921 1931 1941 1951 1961 1971 1981 1991 2001 2011
Percent
Decade
34
i.e. production, has fast been decreasing for half a century to a point where this too, might soon be largely irrelevant from a job market perspective. Now the modern workforce has moved into the “tertiary sector”, also known as the service sector. But certain technical developments are lurking on the horizon, promising to automate even these perhaps last bastions of human productivity competitiveness.
Examining these trends and their likely impact is the purpose of this thesis.
11 Technology trends today
Some would argue that the invention of for example steam power, chemistry and
electricity are the defining technologies of the past few centuries. But in terms of the job market, I would argue that these inventions are just the enablers of what is more
important; certain macro societal trends. For example the mechanization of physical labor, the automation of agriculture forcing large scale migration from rural areas to the cities and improvements in mass production, rendering the skills of craftsmen obsolete. The result of those trends and others are what has defined the labor market today. Looking forward, certain new job market defining macro trends are becoming more and more apparent, promising to perhaps once again revolutionize work as we have come to know it. Future historians may well come to view fields like nanotechnology or bioengineering as the defining enablers of countless new applications that revolutionize industrial
processes and consumer products. But what in my opinion looks especially likely to revolutionize work environment is information technology.
To discuss this topic I have divided this general area into the hardware and software part of information technology, also known as electronics and information processing.
11.1 Electronics
There is little question that technological progress is happening fast in electronics and that technological progress is compounding exponentially in this area. The notion that the
