Epistemic status: Fairly confident on the economics, less so on the political science points. This write-up is mainly the result of a class I took on automation and the labor market with professor Nir Jaimovich as well as the book ‘The Technology Trap’ by Carl Benedikt Frey. Section 3.2 on thinking about the future is (necessarily) more speculative and un-grounded in evidence.

Qualifications: I have finished the coursework for a Master's in Economics which gives me some ability to evaluate the arguments, though I have not taken the time to evaluate the methodology of everything cited.

The following is focused almost entirely on the downside risks from automation and is not meant to represent the entire space of work regarding the impacts of automation. The literature on this topic is very broad and the topics I am addressing here are largely those that economists are thinking about with some political science-related concerns thrown in as well. For example, I omit welfare effects and the meaning of work.

In section 1, I provide a narrative of how the industrial revolution may have shaped the society we live in today. In section 2, I cover how economists think automation is changing society today. I then touch on how these points may be concerning for our future in section 3.

1. A quick summary of important lessons from the industrial revolution:

In order to set the stage and guide thinking about how we might expect near-future technological change to impact the way our society works, let’s take a look at how the industrial revolution may have led to the creation of a democratic society in the West (but mostly Britain). Unless otherwise indicated, this section is basically a summary of Chapter 11 in “The Technology Trap”, the relevant sources can be found there.

A convincing case can be made that the reason we ended up with democratic societies in the West was due to technological change. While this is far from being widely accepted among historians (I spent an hour or two looking for history books which follow this line of reasoning and came up dry), my impression is that it is at least tacitly accepted by many economists and political scientists. I would recommend Dafoe’s paper on Technological Determinism to dive deeper into the general thinking on topics like these and inter-disciplinary differences.

Taking a fairly simplistic (and not-in-the-book) perspective we can think of pre-industrial workers (almost all farmers) as having very little ability to organize or strike. They lived in fairly sparse communities, had no ability to withhold labor (without starving that is, a seemingly unpopular strategy), and were limited in their ability to organize. When the industrial revolution happened, all of the sudden economic elites needed workers in their factories. At first this was unskilled labor (often children as the work was so simple and children so cheap to employ). At first this met with little benefit for the working classes as during a period called Engel's Pause (roughly the first half of the 19th century) there was little to no wage growth (Allen, 2009). It’s worth noting that it was during this period when Marx was coming up with his theories about the exploitation of the working class by capitalists, which seems to have been largely accurate at the time.

The labor demanded was so unskilled that if workers were to strike, the factory owners could simply fire them and find someone else, this led to little bargaining power for the working class. It probably didn’t help that many of the laborers were children under 13 years old. While there were riots and protests at this time (notably the Luddites and Captain Swing Riots) about automation and working conditions more generally, they were at a scale that the English military could suppress. This took quite a lot of effort on the government’s part and was the reason that many countries actively fought against the adoption of labor-saving technology in the past (Frey, 2019. Ch. 3). However, it was not enough pressure that factory owners or governments were forced to increase wages or working conditions.

Eventually the situation changed in the second half of the 19th century as the skill-level needed increased. Workers were actually hard to replace and could negotiate for wages. Eventually they were able to secure the ability to strike legally and to unionize and became more organized. There is a good case to be made here that eventually this led to them being able to push for democratic change. Once some classes of people were enfranchised there was a positive feedback loop among politicians to enfranchise more in order to capture the new voters.

Throughout the 20th century this situation held, with most technology being complementary to labor rather than substituting. Typewriters are a good example of a technology that, while displacing some professional writers, created many new jobs that had previously been too expensive to be feasible. Writing a lot of correspondence and keeping extensive files became economically feasible and new industries and jobs could arise around it. Furthermore, unions were powerful and were constantly negotiating when new technology was adopted to ensure a smooth transition.

I think the general takeaway here is a framing of Western elites as being dragged into the creation of democratic systems by workers. It paints a picture of governments as driven by incentives. According to this narrative, it just so happens that for the past ~150-200 years these incentives have aligned with democracy fairly well. As you read further I encourage you to think about whether or not governmental and elite incentives may be shifting away from providing workers with power in the form of votes.

2. What seems to be happening due to automation now:

It is worth noting that there is a lot of disagreement about what exactly is happening now, to what extent it is due to automation as opposed to trade, and whether we would expect the number of jobs to eventually diminish. Before proceeding I should note that the impacts of trade and automation are very similar. Both automation and trade impact laborers working in routine jobs (jobs involving repetitive and straightforward tasks, think assembly line or data entry). In the case of trade this happens due to the fact that, similar to automation, lower skilled factory work that can be specified well is easier to replace with cheaper workers in poorer countries. So when thinking about trends and expectations for the future, we can look at studies evaluating the effect of trade and essentially treat them as the same. We can then guess that trade is unlikely to increase much more while automation likely will (Frey & Osborne, 2017). What follows is a summary of the ideas circulating about the likely effects from automation (and trade) that are occurring now.

[Edit] The likely pace of automation is hotly contested and I have not yet taken the time to assess the different studies. To give you an idea of the range of estimates, here is an excerpt from the ILO’s literature review on the topic that should give you an idea of the debate (Balliester & Elsheikhi, 2018. Section 2.1.2):

…on the basis of detailed occupations data some observers estimate that 47 per cent of total U.S. employment is at high risk of being digitalised within 20 years [Daheim & Wintermann, 2017]. Globally, automation is estimated to affect 1.1 billion workers (49 per cent of jobs) and US$12.7 billion in wages [Chui et. al, 2017]. Furthermore, the World Bank (2016) estimates as much as 66.6 per cent of jobs susceptible to be made redundant in the developing world due to technology disruption ... In contrast, other studies produce much lower figures, such as Arntz et al. (2016) who find that only around 9 per cent of jobs are automatable in OECD counties

[end edit]

Of primary importance (and more certain) is that many routine jobs are being lost in the US (Jaimovich & Siu, 2012) and Europe (Fernandez-Macias & Hurley, 2017). While it is likely that much of this is due to trade, there has also been some effect from automation (Cortes et. al, 2016a). A high proportion of the people who are losing their jobs either leave the labor force or move to lower-paid non-routine jobs (Cortes et. al, 2016a). It is worth noting that while unemployment is very low at the moment, the measures widely reported by the media only count those who have looked for work in the past 2 weeks. Once someone gives up looking for work they are categorized as NLF (Not in the Labor Force). This is reflected in a declining labor force participation rate in the US, though apparently not in Europe. In the US, the people leaving the labor force are often living on disability.

Many of the lost routine jobs happen to be middle-income jobs. This has led to wage polarization (Acemoglu & Autor, 2011), meaning that the middle is dropping out and we are seeing a greater concentration of workers at lower and upper income levels. This is a likely contributor to rising inequality. Many jobs at the lower end of the income spectrum are either not worth automating or are non-routine and therefore difficult to automate. For example, janitors are poorly paid but they do many tasks that are exceedingly complex for machines to accomplish (f.ex, non-routine tasks like cleaning a toilet).

It is broadly agreed among economists that inflation and purchasing power-adjusted wages have not grown (much) in the past 30-40 years (known as wage stagnation), though this is not without its dissenters. Automation likely contributes to wage stagnation but is unlikely to have been the primary driving force (Acemoglu & Restrepo, 2017). Some models suggest that wage stagnation has been largely the result of decreased worker bargaining power. There is some evidence to back this up (Benmelech et. al, 2019; Azar et. al, 2017).

Automation has further been linked to the movement in the US toward political polarization (Autor et. al, 2016a) and in particular the rise of Trump (Autor et. al, 2016b; Frey et. al, 2018), though it is unclear exactly why this is happening. Given that the studies are largely correlational and do not have convincing causal models, I would encourage a bit of skepticism. Here are a few (of many) theories for how automation might be driving these effects:

• Frey asserts that an important factor was Trump’s promises to rust belt voters to bring back manufacturing jobs, which they liked because they feared automation was taking their jobs (Frey et. al, 2018).
• Rodrick proposes that when economic conditions deteriorate people turn to salient narratives such as, ‘the immigrants are taking your jobs’ which far-right politicians are happy to provide (Rodrik, 2018).
• Dixit and Wiebull (2007) argue that deteriorating conditions lead people to become more extreme in whatever beliefs they have, which would lead to more votes for politicians on both extremes.

3. Reasons we might worry about the consequences of these trends:

The technological determinist narrative laid out in part 1 argues that technological developments during the industrial revolution led to democracy being established. If you buy into this argument, it seems reasonable to worry about what might happen should the conditions that led to democracy no longer hold. New technology (often AI-enabled) is often labor-replacing (Frey, 2019) rather than complementing, unions have lost power, there is a solid case to be made that wage stagnation is due to decreasing worker bargaining power, and even the democratic influence of voters may be waning (though I have not looked into the evidence on this) (Schlosman et. al, 2012). If the changing technological conditions that the industrial revolution brought with it really did shape society to the extent put forward above, what effect might more capable AI systems have on society? How might they shape our institutions and in what direction will they push?

3.1 Potential harms of inequality on democracy & stability

Most of the more tangible near-term (next 10-20 years) negative effects of automation on society seem to run through its effect on inequality (though I hold this view weakly). Assuming this is the case, let’s explore some likely consequences of increasing inequality as well as some related factors. It should be noted that there is likely a significant amount of literature on this topic which I am largely unfamiliar with and the points below are mostly drawn from Bruce Schneier’s blog and Frey’s, ‘The Technology Trap’.

The increasing cost of campaigns means that politicians rely increasingly on wealthier individuals rather than the average voter, making them more beholden to these people (Frey - Tech trap). Inequality worsens this trend by increasing the disparity between top donors and the average ones. There is another possible driver of this trend, better advertising and/or improved manipulation methods. The more these methods improve, the greater role money will play in winning an election as opposed to policy positions and accountability.

Although I do not know of evidence that automation has caused a decline in unions, I deem it fairly likely. If the trend of declining unionization continues, and truly represents a weakened ability of the lower classes to mobilize power, we should see a decline in the ability of the lower-class to influence political decisions. There is some belief among political scientists that the lower-class already has a weakened influence on policy (Gilens & Page, 2014). Though many argue that the studies showing this are not valid. Frey uses the example of minimum wage. Whether or not it would actually be good to increase the minimum wage (a hotly debated issue among economists), the policy is heavily favored by the public (95% of democrats, 75% of low-income Republicans, and 45% of Republicans earning over $150,000/yr) (Frey, 2019). In this case it would seem that organizations like the National Restaurant Association have significant efforts in place to avoid raising the minimum wage while workers seem to have no such coordinated action. We would expect this disparity to worsen as inequality increases.

This is in part because the lower class is generally not very active in politics (Kraus et. al, 2015). It therefore helps if the middle class is interested in passing the legislation that would help the lower class (i.e. welfare legislation). As disparities increase we may see less identification with other groups in society (apparently middle-class and lower-class used to mix much more than they do now (Frey, 2019). Further, Frey characterizes the new partisan divide to be along the lines of intellectuals (Democrats) and the wealthy (Republicans) rather than along rich and poor lines. Looking at how few of the Democratic candidate platforms take the welfare of the lower classes as a top priority, I think this is a fairly reasonable assertion. The implication is that there is not a party looking out for the interests of the lower-class as the Democrats used to. We would expect this to become more of a problem as inequality increased.

A further cause for divide is the perception by whites in the US that African-Americans are getting ahead (which African-Americans do seem to feel) while they get left behind (Frey, 2019). This relative power/wealth loss is unsettling to white Americans which causes some unrest and increases the likelihood of racial backlash. It is possible that increasing inequality continues to have this racially disproportionate effect which could further drive a wedge into an already divided society.

More generally, if Americans look to the future and expect little improvement (which they largely seem to already), they will be unhappy about the current state of affairs and want to change it. There are an abundance of theories as to how this might happen and what they might want to change. It seems clear though that an unhappy lower class could be a driver for further instability. Trump is an obvious outgrowth of this and I would wager that, should current trends continue, we should expect worse down the road.

Many people appear to be in favor of heavy regulation for automation technology (58% think there should be restrictions on how many jobs can be replaced by machines), perhaps due to perception that automation was the cause of their economic situation (Frey, 2018). It may be that, given a fairly populist leader, the US could impose heavy regulation on automation. In fact, I think there is an argument to be made that we should expect the public perception of AI to be closely tied to perceptions of automation. If this is the case, could we see a purposeful slowing down of AI research in the US if people perceive automation to be largely harmful?

There is a strong case to be made, and some evidence (Carbonero et. al, 2018), for greater international inequality due to automation as well. Most countries that have grown rich in the past several decades did so because of a comparatively cheap labor force. As we automate more processes however and the high-skill engineer running an automated factory is the primary source of labor needed for production, we would expect industries to move back to the developed world (closer to market). This would cut off one of the few ways we know for poor countries to catch up economically. It is unclear whether there will be other ways for these countries to grow. The consequence is a more unequal world and perhaps a ‘developing’ world that is no longer developing, but entirely dependent on wealthier nations. The effects of this will no doubt be far reaching, though it is unclear what effect this would have on the developed world.

Finally, it is worth mentioning that information warfare will continue to advance (though defenses will hopefully advance in step). Bruce Schneier’s excellent piece, “Democracy’s Dilemma” sheds some useful light here. He points out three major factors:

• Different sections of the population have fundamentally different ideas about reality. Republicans and Democrats cannot even agree on whether base conditions which a policy seeks to address are true (think anthropogenic climate change). Therefore the argument, which would ideally be about which policy will work better, is more about whether there should be a policy at all. This leaves little room for the classic elements of debate between parties to play out and incentivize finding an optimal solution.
• Trust between parties has declined. As each side perceives the other to be acting in bad faith they will increasingly see it as justifiable to break democratic norms in order to lock-in power while they have it. This could lead to a self-sustaining cycle of deteriorating democratic institutions.
• The public perception of elections, hearings, trials, etc are extremely important in that people need to trust them in order to respect the government's legitimacy. The perception of these can be tampered with fairly easily by outside actors in order to harm the legitimacy of democratic institutions

3.2 Thinking about the long-term future:

Above I have listed many of the more grounded fears based off of some, often weak, evidence. While, taken together, these are concerning, they do not capture the risks we might face if we were to see true mass unemployment (>20%) which could follow more along the lines of Harare’s ‘Useless Class’ idea (Harari, 2017). It is also worth noting that the impacts of automation correlate highly with development of AI capabilities. It seems likely then that if mass unemployment is to occur, it will happen before we develop AGI (possibly shortly before), therefore in any world where we reach AGI, we will also have had to deal with the consequences of mass unemployment.

While I believe it is entirely possible that humanity could benefit greatly from high levels of automation, the example of the industrial revolution gives me pause in thinking we will be able to distribute the benefits of what is coming. If it really is the case that the only reason we ended up with welfare-promoting policies is due to the bargaining power of workers, why would we expect future elites to distribute the benefits of increased productivity in the case that they no longer rely on the labor of the lower-class? While Democracy provides a safeguard by providing workers with voting power that should theoretically be able to safeguard their rights, the threats to democracy detailed above coupled with the possibility of an authoritarian advantage from technological growth (something I will write about more in the future) should be cause for concern. The extent to which this should be a priority cause area when thinking about the long-term future however is still far from clear.

If you're interested in this topic and/or have research ideas, get in touch at alex.l.lintz(at)gmail.com

4. Bibliography:

Allen, R. (2009). “Engels’ pause: Technical change, capital accumulation, and inequality in the british industrial revolution”. Explorations in Economic History. Volume 46, Issue 4, Pages 418-435

Acemoglu, D. & Autor, D. (2011). "Skills, Tasks and Technologies: Implications for Employment and Earnings". Handbook of Labor Economics, Volume 4b

Acemoglu, D. & Restrepo. P (2017). "Robots and Jobs: Evidence from US Labor
Markets". NBER Working Paper No. 23285

Arntz, M. et al. 2016. The Risk of Automation for Jobs in OECD Countries, OECD Social, Employment and Migration Working Papers (Paris, OECD Publishing).

Autor, D., and Dorn, D. (2013), ‘The Growth of Low-skill Service Jobs and the Polarization of the US Labor Market’, American Economic Review, 103(5), 1553–97.

Autor, D., Dorn, D., Hanson, G., and Majlesi, K. (2016a), ‘Importing Political Polarization? The Electoral Consequences of Rising Trade Exposure’, technical report, National Bureau of Economic Research.

Autor, D., Dorn, D., Hanson, G., and Majlesi, K. (2016b), ‘A Note on the Effect of Rising Trade Exposure on the 2016 Presidential Election’, technical report, mimeo, MIT

Azar, J., Marinescu, I. and Steinbaum M., “Labor Market Concentration,” NBER Working Paper 24147, 2017.

Balliester, T. and Elsheikhi, A. (2018). The Future of Work: A Literature Review. International Labor Organization

Benmelech, E., Bergman, N and Kim, H. (2019). “Strong Employers and Weak Employees: How Does Employer Concentration Affect Wages?” https://ssrn.com/abstract=3146679 or http://dx.doi.org/10.2139/ssrn.3146679

Carbonero, F., Ernst, E., and Weber, E. (2018). Robots worldwide: The impact of automation on employment and trade. International Labour Organization

Cortes, G. M., Jaimovich, N., and Siu, H. E. (2016a). Disappearing routine jobs: Who, how, and why? Technical report, National Bureau of Economic Research.

Dafoe, A. (2015). On Technological Determinism: A Typology, Scope Conditions, and a Mechanism. Science, Technology, & Human Values, 40(6), 1047–1076

Daheim, C.; Wintermann, O. 2017. 2050: Die Zukunft Der Arbeit, (Gutersloh, Bertelsmann Foundation).

Dixit, A. K., & Weibull, J. W. (2007). Political polarization. Proceedings of the National Academy of Sciences, 104(18), 7351-7356.

Fernandez-Macias, E, Hurley, J (2017) Routine-biased technical change and job polarization in Europe. Socio-Economic Review 15(3): 563–585.

Frey, C. & Osborne, M. (2017) “The future of employment: How susceptible are jobs to computerisation?”. Technological Forecasting and Social Change, Volume 114

Frey, C., Berger, T., Chen, C. (2018). “Political machinery: did robots swing the 2016 US presidential election?”, Oxford Review of Economic Policy

Frey, C. (2019). “The Technology Trap: Capital, Labor, and Power in the Age of Automation.” Princeton University Press.

Gilens, M., & Page, B. (2014). Testing Theories of American Politics: Elites, Interest Groups, and Average Citizens. Perspectives on Politics, 12(3)

Harari, Noah (2017). Homo Deus: A Brief History of Tomorrow. London: Vintage

Jaimovich, N. and Siu, H. (2012). Job Polarization and Jobless Recoveries. National Bureau of Economic Research. Working Paper

Kraus, M., Anderson, C., and Callaghan, B. (2015), The Inequality of Politics: Social Class Rank and Political Participation

Rodrik, D. (2018). Populism and the Economics of Globalization. Journal of International Business Policy, 1-22.

Schlosman, K., Verba, S., & Brady, H. (2012). The Unheavenly Chorus: Unequal Political Voice and the Broken Promise of American Democracy. Princeton University Press.

World Bank. 2016. World Development Report 2016: Digital Dividends, (Washington, D.C.).