Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity, by Daron Acemoglu and Simon Johnson (PublicAffairs, 560 pp., $32)
Is progress good? Despite the subjective nature of the term, it does seem generally so. We prefer to think of ourselves as moving forward rather than backward, improving over time, and the word embodies this idea.
MIT professors Daron Acemoglu and Simon Johnson are less sure. In their new book, Power and Progress, they argue that society’s elites are biased toward progress that largely benefits them. The elites are techno-optimists, who believe that progress is inherently good and ignore the costs it imposes on those unable to keep up or who lack a say in how new technology is deployed. Thus, shared prosperity from technological innovations is not inevitable. Instead, society must make a conscious choice to share broadly the fruits of progress. Making such a choice requires certain policies and institutions to balance the power between elites and everyone else. Shared prosperity requires, for example, that workers have enough bargaining power to go toe-to-toe with employers. Otherwise, industry elites, especially in tech, will tilt technology in their favor and keep all or most of the benefits of innovation for themselves:
What we are witnessing today is not inexorable progress toward the common good but an influential shared vision among the most powerful technology leaders. This vision is focused on automation, surveillance, and mass-scale data collection, undermining shared prosperity and weakening democracies. Not coincidentally, it also amplifies the wealth and power of this narrow elite, at the expense of most ordinary people.
The authors disdain the “monopoly” that elites have over national agenda-setting. They advocate for including more people and ideas in the national discussion about how technological progress should be pursued.
They provide several wide-ranging historical examples to support their claim that progress often benefits elites at the expense of workers. These include medieval feudalism, antebellum cotton production in the American South, and the construction of the Suez and Panama canals. The authors downplay how, in all these examples, it was government coercion that enforced the labor arrangements benefiting elites.
As the authors report, construction of the Suez Canal relied on forced labor. Officials on the project had to fill recruiting quotas, and the Egyptian military was charged with bringing workers to the canal site and supervising them. The institution of medieval feudalism, in which peasants were tied to a certain lord, was enforced by the king’s court. The same holds for the local monopoly power that medieval monasteries had over the location and use of mills. The example of cotton production in the antebellum American South needs no explanation.
Acemoglu and Johnson portray organized labor as a key institution that can level the playing field for workers and generate shared prosperity. “Technological change is never enough by itself to raise wages, however,” they maintain. “Workers also need to get more bargaining power vis-à-vis employers.”
They credit the rise of unions in the late nineteenth century as a key reason that wages rose for the average worker in the second half of the Industrial Revolution (after 1840). But their analysis of labor unions makes a simple but significant mistake: workers compete with other workers, not employers. It is not the employer but the worker willing to do the same job for a lower wage or under worse conditions who prevents another worker demanding more comfortable accommodations from getting a job. The idea that workers are all on the same team, competing against employers, is a macro view of the economy with little basis in reality. Workers—like employers—are not a homogeneous group, but rather a collection of individuals with distinct goals, preferences, and skills. Thinking in terms of groups leads to poor economic analysis.
Take the authors’ discussion of productivity. They state their primary argument for how wages increase: “When businesses become more productive, they want to expand their output. For this, they need more workers, so they get busy with hiring. And when many firms attempt to do so at the same time, they collectively bid up wages.” This is only half the story, at best.
First, it is not clear what it means for a business or industry to become more productive. A person or team must invent a new process, tool, or software that boosts productivity. How frequently this occurs depends on the benefits that accrue to innovators, which are shaped by regulations, the tax code, and culture. The authors ignore this context, leaving the reader to assume that productivity growth just happens.
Second, their stated mechanism omits important micro processes. For example, a high school graduate working in a service-sector job can take night classes, earn a welding credential, and start earning significantly more money as a welder. This does not require that an entire business or industry become more productive. Similarly, a person with a bachelor’s degree in economics working as an entry-level banking analyst can earn a master’s in applied statistics or data analytics to get a new job that pays a higher wage, commensurate with his new skills.
Individuals pursuing higher wages can raise economy-wide productivity by improving their own skills. Businesses play a role in this process through their demand for workers, but they are not the only force driving it. Explaining economy-wide productivity boosts solely or even primarily through independent, somewhat random increases in business productivity ignores all the “upskilling” that individual workers routinely do to improve their own situations. The authors’ explanation also makes it easier to support higher taxes and more regulation—which they do—since it makes it seem as if progress occurs regardless of incentives.
The authors try to distinguish between technologies that replace workers (“automation”) and technologies that complement workers or “create new uses for human labor.” They support the second while harshly criticizing the first, but their attempt to draw the distinction fails—largely because the distinction is dubious to begin with.
Consider their example contrasting textile manufacturing (automation) with railroads (new uses). They blame new spinning and weaving machines in the early part of the Industrial Revolution for putting home or small-scale spinners and weavers out of work, while doing nothing to improve the economy for workers. Meantime, railroads put some people out of work but also created opportunities, both in the rail industry directly and in other industries that used trains to ship their products.
But it is not obvious that the two industries are much different regarding their overall economic impacts. New spinning and weaving machines drastically increased the amount of cloth and clothing available, which should have expanded the number of merchants and employees selling clothes. Similarly, the new machines were built somewhere and needed to be maintained, creating new jobs for factory workers and service technicians. More textiles were also shipped internationally, raising demand for dockworkers, sailors, and other shipping occupations. It is just as easy to tell a story about employment spillovers from more automation in textiles as for railroads.
The authors use the textile story to support their assertion that, during the first part of the British industrial revolution (roughly 1760–1840), wages remained stagnant and living standards declined for many people, especially the working poor. But other studies find that income per person increased between 1760 and 1830 and did so even more rapidly afterward. Moreover, England’s population doubled from 1760 to 1830, meaning that much of the growth during this period was extensive (more people) rather than intensive (more income per person). The authors downplay population growth, despite the real economic and cultural benefits that come from it.
Other research finds that the mass automation of the early Industrial Revolution, which Acemoglu and Johnson criticize for creating more wealth inequality, did not lead to a significant decline in labor’s share of total national income—undermining their claim. This is important, since the authors use the example of the early Industrial Revolution as a cautionary tale about what new technologies like artificial intelligence will do to today’s workers.
AI and its proponents are the authors’ primary target. They spend most of chapter nine criticizing past claims that AI was on the verge of replacing human labor. Much of what they say makes sense: so far, replacing humans with AI is proving harder than the most optimistic AI champions have claimed. AI, at least in its current form, is not the end of human labor. This gives me hope, anyway (a hope that Acemoglu and Johnson do not share), that humans will continue to outshine AI at many tasks—especially those involving social intelligence—and that demand for human workers will remain strong.
Citing China as an example, the authors worry that AI technology will erode democracy by allowing the state to monitor, control, and, when needed, punish citizens. They also discuss misinformation concerns in the 2016 and 2020 elections but reveal their bias by focusing on Donald Trump and right-of-center voters, while ignoring such issues on the left, such as mainstream media suppression of the story about Hunter Biden’s laptop.
The authors are rightly wary of government’s use of AI and its potential to abuse it, but their healthy skepticism of technocrats does not extend to policy proposals. Nearly all their suggestions—including more redistribution, wealth taxes, breaking up “big tech,” and targeted government subsidies to technologies that benefit “all of society”—would empower the same government officials they worry will abuse their authority. They also fail to mention that their tax proposals would slow economic growth, harming the very workers they want to help. Instead of enforcing our constitutional rules to limit government power, Acemoglu and Johnson would give the government more resources to exercise discretionary power, while hoping that only the “right” people wield it.
This returns us to the author’s complaint about elites dominating the national agenda. Their solution to constrain the techno-optimist elites is for “society at large” to decide whether to approve technologies that affect the general population. “When a company decides to develop face-recognition technology to track the faces in a crowd, to better market products to them or to make sure that people do not participate in protests, their engineers are best placed to decide how to design the software. But it should be society at large that should have a voice in whether such software should be designed and deployed.”
This is no solution. The problem is not with the idea that elites should determine one agenda for all of society, but rather with the idea that society should have only one agenda. We need lots of bottom-up agendas pursued by individuals operating within the law and in accordance with key principles such as dignity, mutual benefit, and openness. One national agenda, regardless of how it is determined, will inevitably suffer from groupthink and hostility to innovative ideas. Stagnation, not shared prosperity, will result.
Power and Progress is better viewed as a polemic against AI and big tech than as a work of academic scholarship, such as The Bourgeois Trilogy by Deirdre McCloskey, Culture of Growth and The Lever of Riches by Joel Mokyr, or even Acemoglu’s earlier work, Why Nations Fail. Its policy prescriptions are the same big-government solutions that progressives have been pushing since Woodrow Wilson, and its economic analysis is incomplete and borderline sloppy in some cases.
Innovation drives economic growth and better living standards. It should be pursued vigorously, though it sometimes causes short-term disruptions. Birthrates are falling around the globe, and in a few decades many developed countries will not have enough workers to support their elderly populations. Rather than posing a threat, AI systems that can do many of the things people do now are our best hope for maintaining the high standard of living we have come to expect. Power and Progress gets this fundamental point wrong.