Visions of Automation: A Comparative Discussion of Two Approaches
2. The Future of Automation: Two Approaches
2.1. Investigating Future Technological Potentials
2.1.1. Searching for Refuges of Human Labor
2.1.2. Utilizing Machine Learning to Learn about the Impacts of Machine Learning
- our assessment of the potential of contemporary and near-future automation technologies is correct (based on the identification of engineering bottlenecks and the reverse assumption that all activities not affected by these engineering bottlenecks are technically automatable);
- O*NET data adequately represents occupational reality;
- nothing went wrong in composing the training data set; and
- the machine learning algorithm we used on the data adequately generalized the training data set in order to assign its probabilistic assessments,
2.1.3. The Problem with Assumptions #1
2.2. The Past’s Future: Empiristic Prognostics
2.2.1. The (Dis-)Advantages of More Classical Macroeconomic Models
2.2.2. The Problem with Assumptions #2
3. Potentials, Projections, and Indeterminacy
Conflicts of Interest
- Frey, C.B.; Osborne, M.A. The Future of Employment: How Susceptible are Jobs to Computerisation? 2013. Available online: https://www.oxfordmartin.ox.ac.uk/downloads/academic/future-of-employment.pdf (accessed on 8 December 2020).
- Frey, C.B.; Osborne, M.A. The Future of Employment: How Susceptible are Jobs to Computerisation? Technol. Forecast. Soc. Chang. 2017, 114, 254–280. [Google Scholar] [CrossRef]
- Matuschek, I. Industrie 4.0, Arbeit 4.0—Gesellschaft 4.0: Eine Literaturstudie. Available online: https://www.rosalux.de/fileadmin/rls_uploads/pdfs/Studien/Studien_02-2016_Industrie_4.0.pdf (accessed on 10 January 2020).
- Kaltenborn, B. Auswirkungen der Digitalisierung auf die Erwerbstätigkeit in Deutschland: Literaturstudie; Working Paper Forschungsförderung No. 157; Hans-Böckler-Stiftung: Düsseldorf, Germany, 2019. [Google Scholar]
- Laukhuf, A.; Runschke, B.; Spies, S.; Stohr, D. Beschäftigungseffekte der Digitalisierung in Branchen: Ein Literaturüberblick; Working Paper Forschungsförderung No. 162, 2019. Available online: https://www.boeckler.de/pdf/p_fofoe_WP_162_2019.pdf (accessed on 10 January 2020).
- Brzeski, C.; Burk, I. Die Roboter kommen: Folgen der Automatisierung für den deutschen Arbeitsmarkt. INGDiBa Econ. Res. 2015, 30, 7p. [Google Scholar]
- Bonin, H.; Gregory, T.; Zierahn, U. Übertragung der Studie von Frey/Osborne (2013) auf Deutschland. ZEW Kurzexpertise 0174-4992 2015, FB455, 51. [Google Scholar]
- Manyika, J.; Chui, M.; Miremadi, M.; Bughin, J.; Georg, K.; Willmott, P.; Dewhurst, P. A Future that Works: AI, Automation, Employment and Productivity. Available online: https://www.mckinsey.com/~/media/McKinsey/Featured%20Insights/Digital%20Disruption/Harnessing%20automation%20for%20a%20future%20that%20works/MGI-A-future-that-works_Full-report.ashx (accessed on 10 January 2020).
- Betz, G. Fallacies in Scenario Reasoning; Karlsruher Institut für Technologie (KIT): Karlsruhe, Germany, 2016. [Google Scholar]
- EPTA. The Future of Labour in the Digital Era: Ubiquitous Computing, Virtual Platforms, and Real-Time Production. Available online: Epub.oeaw.ac.at/ita/ita-projektberichte/EPTA-2016-Digital-Labour.pdf (accessed on 15 August 2019).
- Frey, C.B.; Osborne, M. Automation and the Future of Work—Understanding the Numbers. Available online: https://www.oxfordmartin.ox.ac.uk/blog/automation-and-the-future-of-work-understanding-the-numbers/ (accessed on 15 November 2018).
- Pfeiffer, S.; Suphan, A. Der AV-Index. Lebendiges Arbeitsvermögen und Erfahrung als Ressourcen auf dem Weg zu Industrie 4.0. Available online: http://www.sabine-pfeiffer.de/files/downloads/2015 (accessed on 20 August 2016).
- Nuffield Council on Bioethics. Emerging Biotechnologies: Technology, Choice and the Public Good; Nuffield Council on Bioethics: London, UK, 2012; ISBN 9781904384274. [Google Scholar]
- Edwards, M.A.; Roy, S. Academic Research in the 21st Century: Maintaining Scientific Integrity in a Climate of Perverse Incentives and Hypercompetition. Environ. Eng. Sci. 2017, 34, 51–61. [Google Scholar] [CrossRef] [PubMed][Green Version]
- OECD. Getting Skills Right: Skills for Jobs Indicators; OECD Publishing: Paris, France, 2017; ISBN 978-92-64-27787-8. [Google Scholar]
- Autor, D. Polanyi’s Paradox and the Shape of Employment Growth. In Proceedings of the Polanyi’s Paradox and the Shape of Employment Growth, National Bureau of Economic Research. Cambridge, MA, USA; 2014; pp. 129–177. [Google Scholar]
- Frey, C.B. The Technology Trap: Capital, Labor, and Power in the Age of Automation; Princeton University Press: Princeton, NJ, USA, 2019; ISBN 9780691172798. [Google Scholar]
- Brownlee, J. How Much Training Data is Required for Machine Learning? Available online: https://machinelearningmastery.com/much-training-data-required-machine-learning/ (accessed on 22 December 2019).
- Brandes, P.; Wattenhofer, R. Opening the Frey/Osborne Black Box: Which Tasks of a Job are Susceptible to Computerization? Available online: http://arxiv.org/pdf/1604.08823v2 (accessed on 20 October 2020).
- Durán, J.M. Varying the Explanatory Span: Scientific Explanation for Computer Simulations. Int. Stud. Philos. Sci. 2017, 31, 27–45. [Google Scholar] [CrossRef][Green Version]
- Durán, J.M.; Formanek, N. Grounds for Trust: Essential Epistemic Opacity and Computational Reliabilism. Minds Mach. 2018, 28, 645–666. [Google Scholar] [CrossRef][Green Version]
- Krohs, U. How Digital Computer Simulations Explain Real-World Processes. Int. Stud. Philos. Sci. 2008, 22, 277–292. [Google Scholar] [CrossRef]
- Humphreys, P. Computational Science and Its Effects. In Science in the Context of Application; Carrier, M., Nordmann, A., Eds.; Springer Netherlands: Dordrecht, The Netherlands, 2011; pp. 131–142. ISBN 978-90-481-9050-8. [Google Scholar]
- Valenduc, G.; Vendramin, P. Work in the Digital Economy: Sorting the Old From the New; European Trade Union Institute Brussels: Brussels, Belgium, 2016. [Google Scholar]
- Wolter, M.I.; Mönnig, A.; Hummel, M.; Weber, E.; Zika, G.; Helmrich, R.; Maier, T.; Neuber-Pohl, C. Economy 4.0 and Its Labour Market and Economic Impacts: Scenario Calculations in Line with the BIBB-IAB Qualification and Occupational Field Projections; IAB Research Report 13/2016; Institut für Arbeitsmarkt- und Berufsforschung: Nürnberg, Germany, 2016. [Google Scholar]
- Frey, P.; Schaupp, S. Futures of digital industry: Techno-managerial or techno-political utopia? BEHEMOTH-A J. Civilis. 2020, 13. [Google Scholar] [CrossRef]
- Pfeiffer, S. The Vision of “Industrie 4.0” in the Making-a Case of Future Told, Tamed, and Traded. Nanoethics 2017, 11, 107–121. [Google Scholar] [CrossRef] [PubMed][Green Version]
- Das IAB/INFORGE-Modell: Ein Sektorales Makroökonometrisches Projektions- und Simulationsmodell zur Vorausschätzung des Längerfristigen Arbeitskräftebedarfs; Zika, G.; Schnur, P.; Bertelsmann, W. (Eds.) W. Bertelsmann: Bielefeld, Germany, 2009; ISBN 3763940057. [Google Scholar]
- Ahlert, G.; Distelkamp, M.; Lutz, C.; Meyer, B.; Mönnig, A.; Wolter, M.I. Das IAB/INFORGE-Modell. In Das IAB/INFORGE-Modell: Ein Sektorales Makroökonometrisches Projektions- und Simulationsmodell zur Vorausschätzung des Längerfristigen Arbeitskräftebedarfs; Zika, G., Schnur, P., Bertelsmann, W., Eds.; W. Bertelsmann: Bielefeld, Germany, 2009; pp. 15–170. ISBN 3763940057. [Google Scholar]
- Burrell, J. How the machine ‘thinks’: Understanding opacity in machine learning algorithms. Big Data Soc. 2016, 3, 205395171562251. [Google Scholar] [CrossRef]
- Tetens, H. Wissenschaftstheorie: Eine Einführung, Orig.-Ausg; Beck: München, Germany, 2013; ISBN 978-3-406-65331-5. [Google Scholar]
- Timcke, S. The One-Dimensionality of Econometric Data: The Frankfurt School and the Critique of Quantification. Triple C 2020, 18, 429–443. [Google Scholar] [CrossRef]
- Colander, D.; Goldberg, M.; Haas, A.; Juselius, K.; Kirman, A.; Lux, T.; Sloth, B. The Financial Crisis and the Systemic Failure of the Economics Profession. Crit. Rev. 2009, 21, 249–267. [Google Scholar] [CrossRef]
- Naidu, S.; Rodrik, D.; Zucman, G. Economics after Neoliberalism: Introducing the EfIP Project. AEA Pap. Proc. 2020, 110, 366–371. [Google Scholar] [CrossRef]
- Dengler, K.; Matthes, B. Folgen der Digitalisierung für die Arbeitswelt: Substituierbarkeitspotenziale von Berufen in Deutschland; IAB Research Report 11/2015; Institut für Arbeitsmarkt- und Berufsforschung: Nürnberg, Germany, 2015. [Google Scholar]
- Schiølin, K. Revolutionary dreams: Future essentialism and the sociotechnical imaginary of the fourth industrial revolution in Denmark. Soc. Stud. Sci. 2020, 50, 542–566. [Google Scholar] [CrossRef] [PubMed]
- Dieckhoff, C.; Appelrath, H.-J.; Fischedick, M.; Grunwald, A.; Höffler, F. Zur Interpretation von Energieszenarien; Stand: Mai 2014; Acatech—Deutsche Akademie der Technikwissenschaften: München, Germany, 2014; ISBN 978-3-9817048-1-5. [Google Scholar]
- Horkheimer, M. Traditional and Critical Theory. In Critical Theory: Selected Essays; Horkheimer, M., Ed.; A&C Black: Edinburgh, UK, 2002; pp. 188–243. ISBN 0826400833. [Google Scholar]
- Srnicek, N.; Williams, A. Inventing the Future: Postcapitalism and a World without Work; Verso: Brooklyn, NY, USA, 2015; ISBN 9781784780968. [Google Scholar]
- Weeks, K. Anti/Postwork Feminist Politics and A Case for Basic Income. Triple C Commun. Cap. Crit. Open Access J. Glob. Sustain. Inf. Soc. 2020, 575–594. [Google Scholar] [CrossRef]
- Grunwald, A. Technology Assessment in Practice and Theory; Routledge: London, NY, USA, 2019; ISBN 1138337080. [Google Scholar]
- Gransche, B. Vorausschauendes Denken: Philosophie und Zukunftsforschung jenseits von Statistik und Kalkül; Transcript: Bielefeld, Germany, 2015; ISBN 978-3-8376-3038-1. [Google Scholar]
- Adorno, T.W. Spengler nach dem Untergang. In Kulturkritik und Gesellschaft I/II; Adorno, T.W., Ed.; Suhrkamp: Frankfurt am Main, Germany, 1977; pp. 47–71. ISBN 3518572261. [Google Scholar]
- Adorno, T.W. Soziologie und empirische Forschung. In Soziologische Schriften I.; Adorno, T.W., Ed.; Suhrkamp: Frankfurt am Main, Germany, 1972; pp. 196–216. ISBN 3518572261. [Google Scholar]
- Kosow, H.; León, C.D. Die Szenariotechnik als Methode der Experten- und Stakeholdereinbindung. In Methoden der Experten- und Stakeholdereinbindung in der sozialwissenschaftlichen Forschung; Niederberger, M., Wassermann, S., Eds.; Springer Fachmedien Wiesbaden: Wiesbaden, Germany, 2015; pp. 217–242. ISBN 978-3-658-01686-9. [Google Scholar]
- Besley, T.; Hennessy, P. Letter to Her Majesty The Queen: The Global Financial Crisis—Why Didn’t Anybody Notice? The British Academy: London, UK, 2009. [Google Scholar]
- Solow, R.M. Written Statement. In Building a Science of Economics for the Real World: Hearing before the Subcommittee on Investigations and Oversight; Committee on Science and Technology: Washington, DC, USA, 2010; pp. 14–15. [Google Scholar]
- Aldred, J. This Pandemic Has Exposed the Uselessness of Orthodox Economics. Available online: https://www.theguardian.com/commentisfree/2020/jul/05/pandemic-orthodox-economics-covid-19 (accessed on 7 July 2020).
- Grunwald, A. Transformative Wissenschaft als honest broker? Das passt! GAIA—Ecol. Perspect. Sci. Soc. 2018, 27, 113–116. [Google Scholar] [CrossRef][Green Version]
- Frey, P.; Schneider, C.; Wadephul, C. Demokratisierung von Technik ohne Wirtschaftsdemokratie? TATuP 2020, 29, 30–35. [Google Scholar] [CrossRef]
- Frey, P.; Schaupp, S.; Wenten, K.-A. Towards Emancipatory Technology Studies. Nanoethics 2021, 15, 19–27. [Google Scholar] [CrossRef]
Their approach thereby also circumvents the distinction between manual and cognitive labor, acknowledging the fact that the implicit identification of manual labor with (automatable) routine labor and cognitive labor with (unautomatable) non-routine labor might hold less and less true over time, allowing more widespread automation in the service sector.
To verify the reliability of the hand-labelled classification, Frey and Osborne used Gaussian process classifiers based on the set of O*NET variables linked to the engineering bottlenecks. The algorithm accurately managed to reproduce the hand-labels of the experts, verifying “that our subjective judgements were systematically and consistently related to the O*NET variables” ( (p. 34)).
To be fair, this should not be interpreted simply as a sign of excessive enthusiasm or even personal conceit, but (at least in part) as an effect of a highly competitive scientific system in which any scientist is called upon, even forced, to highlight the great potentials of the respective field she is researching, lest the scarce funding go to the development of some other promising technology—or even worse, the humanities [13,14].
In composing the data training set, the machine learning experts were accordingly asked to consider “the possibility of task simplification” to the best of their knowledge ( (p. 30)).
In light of the immense volumes of data utilized in today’s machine learning, a training data set of 70 feature vectors, each containing only nine variables (the engineering bottleneck-related variables of O*NET, deemed relevant to the question of automatability), seems rather modest. Although the amount of data needed for machine learning depends on the specific use case, this concern seems particularly relevant in this case, as non-linear algorithms are known to require even bigger training data sets .
In a notable exception, two computer scientists of the Swiss Federal Institute of Technology in Zurich dedicated themselves to “Opening the Frey/Osborne Black Box” . Yet although they refer to the study as a black box, they do not engage in great detail with its workings. Rather, they build their own model to identify outliers in the results of Frey and Osborne in order to allow for a more detailed scrutiny of the study’s results.
The literature review on cross-country validity of O*NET scores of a recent OECD study concluded, however, “that occupational titles refer to very similar activities and skill demands across different countries”  (p. 42), implying that the claim that the findings could not be applied to other economies might owe less to actual differences in job realities and more to an implicit nationalist bias.
One might, of course, also criticize their study by claiming that they should have dealt with labor market impacts, rather than simply highlighting technological potentials. I will return to the “use value” of these studies at the end of this chapter. Thus far, I have focused on a form of immanent critique, reviewing the study in the light of the objectives it sets itself.
The term Economy 4.0 represents an extension of the Industry 4.0 term, popular in contemporary German debates to denote the current phase of technological development, to the whole of the economy, as the study does not limit itself to changes within industry and agriculture  (p. 9). For an introduction to the Industry 4.0 discourse see [26,27].
The QuBe was developed by the BIBB and focuses on modelling the general demography of Germany (by nationality, gender and age), labour supply (with factors including for instance levels of labour participation and qualification) and labour demand (with factors including occupational requirements and wage and price levels).
The study actually reads “95 percent of all households will have a 50 Mbit/s connection by 2018“  (p. 26). I would suggest interpreting this assumption as saying that they in principle could access broadband, rather than that they in fact will have such a connection, provided that there might be a number of reasons for households not to opt for more expensive broadband tariffs—unless the connection would be supplied by the public sector to all households free of charge as a public service. However, Wolter et al. give no indication that they had that in mind.
I would suggest that the reservations towards the (self-)assessment of practitioners that were raised above regarding AI experts should also be taken into account here. After all, within a societal context that is buzzing with high expectations and the normative pressure to endorse and enact innovation to attract investors, the assessment of technological potentials appears to be at very least skewed (regarding the normative power of the Industry 4.0 discourse, see [26,27].
See my discussion of O*NET above. The BERUFENET, for instance, also does not cover differences in occupational realities within job profiles. Nonetheless, it should be positively noted that using a German database bypasses issues resulting from applying assessments from the US labor market to the German one.
To be fair, in a more recent paper, published after the peak of the Industry 4.0 debates and unavailable in English, Wolter et al. addressed both these desiderata by moving towards a methodology much closer to the one developed by Frey and Osborne (which can be understood as a tacit vindication of their approach) and by modeling branch-specific utilization levels based on investment activities. Although the projected job losses due to accelerated technological development are much higher in comparison to the 2016 study (e.g., they projected that 100,000 jobs will be lost in 2030 compared to just 30,000 in the 2016 projection), they remain miniscule in comparison to the whole of the labor market. This is consistent with my earlier expositions regarding the socioeconomic determinacy of technological unemployment: Even if one assumes a higher technological dynamic and use, the development of unemployment ultimately depends strongly on demand for goods and services and the associated job creation, rather than technological development per se.
For instance, their projections of increased governmental consumer spending is limited to the areas of cyber crime and/or cyber warfare, with the state projected to hire 14,000 additional soldiers and boost the federal police force by 2000 employees  (p. 45)). The exclusive focus on additional military and police spending seems, for lack of a better term, odd. Another assumption—that domestic consumer demand will be boosted by rising wages as productivity increases—is normatively appealing and should, in my opinion, indeed be pursued as a policy goal, but is currently not as self-evident as Wolter et al. assume. After all, the erosion of the link between productivity and wage increases can be considered one of the key contributors to the increased social polarization of the last decades.
The findings should not be mistaken as direct “instructions” for policymaking, however. Not only normatively because of the relative autonomy of the political sphere, but also because the study seems to lack robust sensitivity analyses for individual factors that that might then inform policymaking  (p. 33). The approach to create a number of scenarios that build on each other, each linked to a more limited set of assumptions, could be charitably interpreted as serving as an “aggregate sensitivity analysis” of sorts, but even then we do not know whether specific changes in the scenarios are dependent on a certain exact assumption.
Since it is central to this Special Issue’s subject, I would only like to remind you of the exemplary fact that the assumption regarding the form and extent of automation in the future employed by Wolter et al. is informed by an outdated understanding of automatability and an additional ad hoc assumption (see above, also for a reference to the 2019 study improving on this assumption). It is also noteworthy that although the assumptions are discussed individually, there is no attempt to justify them in combination (i.e., is it possible that all these assumptions will come to pass at once?), although it seems likely to me that such a justification could be achieved. Regarding the need not only to justify individual assumptions in scenario modelling but also their combination, see  (p. 24).
This realization echoes earlier comments by Horkheimer, who pointed out that directions and goals of research “are not self-explanatory nor are they, in the last analysis, a matter of insight”  (p. 196). Rather, they should be understood as being shaped by social conditions.
The awareness of alternative futures constitutes a key epistemic advantage of scenario modelling in comparison to earlier prognostic models, as it owns up to the epistemic uncertainty linked to any attempt to ”look into the future“ .
Much in the same spirit, the Committee on Science and Technology of the US Congress convened a year later for a hearing committed to “Building a science of economics for the real world“ (note the delegitimization this title implies—after all, one should have expected economics to always have been about the real world, particularly in light of the prominence of economists in scientific advisory practices). Among the witnesses was Robert Solow, one of the most highly decorated and influential economists of the period after the Second World War (not only did Solow receive the Nobel Prize for Economics himself, but so did four former PhD students of his). In his statement, he echoes his British colleagues, pointing out that “the approach to macroeconomics that dominates serious thinking, certainly in our elite universities and in many central banks and other influential policy circles, seems to have absolutely nothing to say about the problem [of justifying their basic concepts, particularly in relation to (un-)employment]. Not only does it offer no guidance or insight, it really seems to have nothing useful to say”  (p. 14).
On a side note, the disproportionate scrutiny facing scientific critics of contemporary society was already reflected on by Horkheimer: “Although critical theory at no point proceeds arbitrarily and in chance fashion, it appears, to prevailing modes of thought, to be subjective and speculative, one-sided and useless. Since it runs counter to prevailing habits of thought, which contribute to the persistence of the past and carry on the business of an outdated order of things […], it appears to be biased and unjust”  (p. 218).
e.g., that Frey and Osborne were first, that the public outreach of Oxford University might be better than that of IAB and BIBB, or that statements about the US labour market are deemed more interesting internationally than those about the German labor market.
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.
© 2021 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Frey, P. Visions of Automation: A Comparative Discussion of Two Approaches. Societies 2021, 11, 63. https://doi.org/10.3390/soc11020063
Frey P. Visions of Automation: A Comparative Discussion of Two Approaches. Societies. 2021; 11(2):63. https://doi.org/10.3390/soc11020063Chicago/Turabian Style
Frey, Philipp. 2021. "Visions of Automation: A Comparative Discussion of Two Approaches" Societies 11, no. 2: 63. https://doi.org/10.3390/soc11020063