Automation and The Limits of Contemporary Economics

The rise of artificial intelligence has prominent economists eager to opine on the downsides of automation. But their solutions are of limited value.

The reality of increasing automation in the American economy has become unavoidable over the last few years. Grocery stores are employing fewer cashiers thanks to self-checkout machines; headlines warn that factories will transition to “lights-out” production, where human activity is rendered unnecessary; and AI models are proving themselves capable of producing creative output, numerical analyses, and computer code as well as any decent secretary or programmer. Every week, you can find a new article about yet another industry where AI is being used to displace labor.¹ And, despite occasional setbacks, there are no signs it will stop soon. In just the next two years, investment in AI for manufacturing is expected to rise by more than fifty percent.²

To some extent, these trends are not new. After all, automation is reported to have played a role in eliminating up to 70% of American middle-class jobs lost since 1980.³ However, the remarkable achievements of AI and its encroachment into activities previously thought unlikely candidates for automation have increased anxieties around this topic and caused economists to question their profession’s typical techno-optimism, which has been ingrained in economic theory since Adam Smith wrote that:

In consequence of better machinery, of greater dexterity, and of a more proper division of work… a much smaller quantity of labor becomes requisite for any particular piece of work, and though, in consequence of the flourishing circumstances of the society, the real price of labor should rise very considerably…⁴

Smith’s analysis has provided the basic framework for economists thinking about automation, at least until recently. Following his line of thought, automation reduces firms’ costs and thus increases their profit margins, incentivizing them to produce more, hire more, and thereby drive wages upward. For much of American history, these correlations at least ran the right way, even if an empirical basis for causation could never quite be established. While countless job categories have been eliminated because of automation, countless more sprang up in their place, undergirded by a fast rate of economic growth and domestic demand for new products. In postwar America —a period idealized by many economists— sharecroppers displaced by agricultural machinery and workers pushed off the assembly line could find alternate employment arrangements paving the way for suburbanization and automobile-centric life, building the American war machine, participating in America’s growing white-collar professions, or taking up work in one of the many industries still growing and not yet automated.

However, since the 1970s, secular stagnation has limited the number of new jobs arising to replace old automated ones, and several economists are now drawing attention to the fact that the position of workers is not necessarily improved by automation. In a recent paper, MIT’s Daron Acemoglu and Boston University’s Pascual Restrepo conclude that “technological improvements that increase productivity can actually reduce the wage of all workers.”⁵ Several papers now back this conclusion, finding that automation increases inequality, reduces the share of income received by labor, and particularly harms the prospects of low-skilled workers.⁶

But what precisely can be done about these harms? This question was taken up by MIT economist David Autor in an op-ed that perfectly characterizes his profession’s typical view:

My thesis is not a forecast but a claim about what is attainable. AI will not decide how AI is used, and its constructive and destructive applications are boundless. The erroneous assumption that the future is determined by technological inevitabilities — what Shoshana Zuboff terms inevitabilism — deprives citizens of agency in making, or even recognizing, the collective decisions that will shape the future…AI offers vast tools for augmenting workers and enhancing work. We must master those tools and make them work for us⁷

Stating that the direction of technological development is not inevitable is a truism. Instead of repeating the obvious, economists ought to be estimating the probabilistic direction of technological development given the structure of markets, institutions, and American politics. Autor sidesteps this topic by merely insisting that “We [italics added] must master these tools and make them work for us.” He follows up, insisting that  “AI — used well — can assist with restoring the middle-skill, middle-class heart of the U.S. labor market that has been hollowed out by automation and globalization.” 

But how are we supposed to ensure that AI is “used well?” The development of AI is not dictated by us, or workers, or even society at large, but by a narrow segment of capitalists who are utilizing it for their own social and economic goals— primarily the economic imperative to maximize profits at whatever social cost.

Sure, it isn’t inevitable that automation won’t be used to restore the middle class. But under our current economic system, what are the incentives for businessmen to expend their resources on that? And if those incentives already exist, why aren’t they being acted on?

An attempt to address this conundrum is the factually challenged⁸ bestseller by Daron Acemoglu and Simon Johnson, Power and Progress. They recognize that:

The concentrated power of business undercuts shared prosperity because it limits the sharing of gains from technological change. But its most pernicious impact is via the direction of technology, which is moving excessively toward automation, surveillance, data collection, and advertising. To regain shared prosperity, we must redirect technology, and this means activating a version of the same approach that worked more than a century ago for the Progressives.⁹

Acemoglu and Johnson are surely correct in highlighting the importance of both the direction of technological development and the distribution of the benefits from it, but their solutions leave a lot to be desired.

They begin by stating that we must decide how AI is used: 

[Change] can only start by altering the narrative and the norms…Debate on new technology ought to center not just on the brilliance of new products and algorithms but also on whether they are working for the people or against the people…Whether digital technologies should be used for automating work and empowering large companies and nondemocratic governments must not be the sole decision of a handful of entrepreneurs and engineers. 

This statement is obviously true on its face. Normal people should have a democratic say in how technology is developed, but the authors never mention how they might do so. Instead, they briefly mention the importance of investors having a voice in this process: “These are not decisions investors should care about only because of the profits they generate. A two-tiered society with a small elite and dwindling middle class is not a foundation for prosperity or democracy.” On the contrary, investors have been to happy to fund the trend toward labor-displacing automation in pursuit of short-term profits, though it is likely true, as Paul Mattick once noted, that this sort of labor displacement “undermines the very structure of capitalist society.”¹⁰

Acemoglu and Johnson also list several concrete policy prescriptions that they believe can address the issues posed by AI. These ideas generally fall into three sometimes-overlapping camps:

1. Policies that aim to slow the displacement of workers by automation and nudge firms toward hiring humans (A tax on automation (i.e. a “robot tax”), tax reform, new regulations regarding privacy protections and data ownership, breaking up Big Tech)

2. Policies that redistribute the wealth created by automation (supporting the growth of the labor movement and civil society, investing in education, embracing an expanded welfare state and/or UBI)

3. Policies that aim to change the direction in which AI is being developed (“Government subsidies for developing more socially beneficial technologies”)

I support many of these policies, but it’s hard for me to believe they’ll seriously impede the development of automation in its current direction. Taxing robots, for instance, may slow their adoption, but those taxes would have to be inordinately high to confiscate all of the labor costs they’d save firms. The adoption of strong regulations regarding what automation may not do has some promise, at least in the realm of privacy and consumer protection, but it’s hard to imagine it will play a major role in protecting labor. 

The redistributive policies listed here are the most ineffectual. How is the labor movement supposed to amass strength as AI erodes its membership and political power? And what makes anybody think that firms are going to be responsive to the anti-automation cries of civil society? An expanded welfare state could soften the blow from losses dealt by AI, but most Western countries appear to have little political appetite for expanding their welfare states with many more apt to contract them. As for UBI, its welfare effects remain ambiguous, so long as it may end up as a “substitute, not as an addition, to present social policy.”¹¹ Even if adopted alongside the existing welfare state, the benefits of UBI are limited. As Aaron Benanav summarized it: “UBI would empower workers without disempowering capital, providing people more autonomy in the fulfillment of their “animal functions” but no greater role in shaping the wider social conditions under which they do so.”¹² 

The third set of policies, centered around government subsidies for beneficial technologies is the most promising, though its potential seems limited to introducing some investments beneficial to labor alongside the existing labor-displacing ones. The authors’ ideas of subsidizing technologies that increase the labor share and provide additional jobs to humans are good ones so long as they don’t fall victim to graft and administrators are able to find some way to efficiently distinguish desirable investments from undesirable ones. But one questions how far these ideas could be taken before succumbing to business opposition. Perhaps the business class could tolerate some mild government innovations that could be appropriated by private enterprise for profit-making, but it surely will balk at sustained public investment in activities that put the government (or the recipients of its largesse) in direct competition with private businesses. The closer the government gets to socializing the process of automation (i.e. the more control it can exert directly over the innovative process), the closer it would likely get to remedying the wrongs highlighted by Acemoglu and Johnson. However, they are not calling for anything so drastic.

Overall, the remedies provided in Power and Progress consist of a variety of incremental nudgings that are unlikely to seriously curtail the incentives of firms to automate in order to maximize their profits. This, however, is not a unique observation; less apologetic explanations for the way technology has developed have been around for a while, and fit today’s society even more than their own. Nearly fifty years ago, Harry Braverman systematically analyzed the role of automation in degrading work and segmenting the labor market into a small minority of skilled laborers and a large majority of workers confined to increasingly mindless tasks.¹³ Even as early as 1972, a design consultant at Case Western Reserve could note that “We may have created too many dumb jobs for the number of dumb people to fill them.”¹⁴ This trend did not develop by happenstance nor even because it was the only way to maximize the productivity of labor. The deskilling of labor helped maximize profits not only by increasing worker productivity but also by making every worker increasingly replaceable and driving down competition between firms for workers.¹⁵

These analyses also explain why John Maynard Keynes’s premature prediction of “technological unemployment” or his imagined reality in which his grandchildren worked only fifteen hours weekly never came to pass.¹⁶ Why on Earth, after all, would employers voluntarily shrink the length of the working day, employ additional workers, and thus have to deal with a greater administrative burden and what in aggregate would be a tighter labor market? As Mattick explained:

to cut down working hours generally and maintain the wages bill would defeat the capitalist’s purpose in introducing technological change and make automation a senseless affair. The point of automation is precisely to reduce wage costs relative to overall costs of the “factors of production” and to recoup the higher capital costs by greater productivity.¹⁷

Countering these great economic forces is an enormous challenge— one that requires questioning the economic assumptions of the past and the economics profession’s typical deference toward capitalists’ investment decisions and their right to endlessly recoup gains from them. But Autor is right when he says that we shouldn’t embrace what Zuboff called inevitablism. As she so well explained it:

The relentless drumbeat of inevitably messages presents the new apparatus of ubiquity as the product of technological forces that operate beyond human agency and the choices of communities, an implacable movement that originates outside history and exerts a momentum that in some vague way drives toward the perfection of the species and the planet. The image of technology as an autonomous force with unavoidable actions and consequences has been employed across the centuries to erase the fingerprints of power and absolve it of responsibility.¹⁸

A better world is in fact possible, but it will require entertaining ideas that undermine the power of capitalists— ideas like full employment, collective and public ownership, and even socialism. Contrary to Autor, Acemoglu, and Jonhson, there is no democratic debate shaping the direction of technological development, and under our current economic system there can’t be, leaving aside some relatively niche regulatory issues. Asking capitalists not to develop and employ profit-maximizing technology is asking them not to be capitalists. Such an approach is a dead end that mainstream economists will remain stuck in until they venture beyond the current ideological horizons of their profession.

1. This week’s industry was video game development. See Merchant, Brian. “AI Is Already Taking Jobs in the Video Game Industry,” WIRED, July 23, 2024. https://www.wired.com/story/ai-is-already-taking-jobs-in-the-video-game-industry/

2. “How one company is using AI to transform manufacturing,” World Economic Forum, January 9, 2024. https://www.weforum.org/agenda/2024/01/company-using-ai-transform-manufacturing-business/

3.  Winck, Ben. “Automation helped kill up to 70% of US’s middle-class jobs since 1980, study says,” Business Insider, June 22, 2021. https://www.businessinsider.com/automation-labor-market-wage-inequality-middle-class-jobs-study-2021-6

4. Adam Smith, Quoted in Johnson, Simon, and Daron Acemoglu. Power and progress: Our thousand-year struggle over technology and prosperity. Hachette UK, 2023, 4.

5. Acemoglu, Daron, and Pascual Restrepo. "Artificial intelligence, automation, and work." In The economics of artificial intelligence: An agenda, pp. 197-236. University of Chicago Press, 2018. https://www.nber.org/system/files/chapters/c14027/c14027.pdf Quote is sourced from Frick, Walter. “AI Making Some Economists Rethink the Story of Automation,” Harvard Business Review, May 27, 2024 https://hbr.org/2024/05/ai-is-making-economists-rethink-the-story-of-automation

6. See Prettner, Klaus, and Holger Strulik. "Innovation, automation, and inequality: Policy challenges in the race against the machine." Journal of Monetary Economics 116 (2020): 249-265. https://www.sciencedirect.com/science/article/pii/S0304393219301965 and Moll, Benjamin, Lukasz Rachel, and Pascual Restrepo. "Uneven growth: automation's impact on income and wealth inequality." Econometrica 90, no. 6 (2022): 2645-2683. https://www.nber.org/system/files/working_papers/w28440/w28440.pdf

7. Autor, David. “AI Could Actually Help Rebuild The Middle Class.” Noema Magazine. February 12, 2024. https://www.noemamag.com/how-ai-could-help-rebuild-the-middle-class/

8. For several examples, see: Smith, Noah. “Book Review: “Power and Progress”,” Noahpinion Substack, February 21, 2024.

9. Johnson, Simon, and Daron Acemoglu. Power and progress: Our thousand-year struggle over technology and prosperity. Hachette UK, 2023, 393.

10. Mattick, Paul. Marx and Keynes: the limits of the mixed economy. Extending Horizon Books, 1969, 192-193

11.  Milton Friedman quoted in Burns, Jennifer. Milton Friedman: the last conservative. Farrar, Straus and Giroux, 2023, 178

12. Benanav, Aaron. Automation and the Future of Work. Verso Books, 2020, 78

13. Braverman, Harry. Labor and monopoly capital: The degradation of work in the twentieth century. NYU Press, 1998.

14. Cited in Ibid, 24.

15. For several examples of this phenomenon see: Noble, David. Forces of production: A social history of industrial automation. 1984.

16. Keynes, John Maynard. "Economic possibilities for our grandchildren." In Essays in persuasion, pp. 321-332. London: Palgrave Macmillan UK, 1930.

17. Mattick, Paul. Marx and Keynes: the limits of the mixed economy. Extending Horizon Books, 1969, 203.

18. Zuboff, Shoshana. The age of surveillance capitalism. Profile Books, 2018, 224