Daron Acemoglu won the Nobel Prize in economics in 2024 and is the third most cited economist of all time. Thus, the paradox of his new book, What Happened to Liberal Democracy?, is that its most serious analytic failures are economic. He provides a sincere and serious consideration of modern liberalism’s estrangement from the working class. But when he moves from description to solutions, the basic building blocks of economic analysis disappear. Imperfect markets are compared with an idealized government. Expansive agencies are proposed while the risk of their capture—well known in public-choice theory—is ignored. And while Acemoglu is known for integrating institutions into economics, he sometimes celebrates democratic consensus without attention to the elite institutions that formed it.

Acemoglu “pitches his tent” on the left side of liberalism, arguing that it “has historically been an engine of material and social progress.” Today he believes tradition should give birth to “working-class liberalism,” which means, for him, creating a more level economic playing field and technological change that complements rather than displaces labor. The book is well-intentioned. He recognizes that elites cannot be trusted to direct culture. But he never explains why they can be trusted to direct technology, redistribution, and the administrative state.

Acemoglu argues that the postindustrial economy puts a premium on cognitive skills useful in fields such as elevating finance, health, and education, while reducing the importance of manufacturing. He claims that automation “severs the link between mass production and shared prosperity” and has resulted in “the disappearance of good jobs for workers without a college degree.” Acemoglu thus endorses the economic explanation for populist discontent popular among left-liberals.

What is original about his account is his willingness to also highlight the contribution of left-liberals to working-class alienation. As in my own book, Why Democracy Needs the Rich , he sees the class of educated liberals as attracted to social engineering because it “confers on them greater status and social influence.” Acemoglu recognizes that this class then abused its power by imposing cultural values without democratic buy-in, further alienating the working class. He rues the fact that many on the contemporary Left became “convinced that protecting minorities and vulnerable groups and building a fairer economy required silencing those with opposite viewpoints.” This is a welcome admission, but Acemoglu is selective in his political economy. His distrust of elites largely vanishes when they administer the economic policies he favors, and he also fails to consider the influence of other powerful interest groups.

His argument for more redistribution and a bigger state in economic matters flows from his principle of “nondomination,” which Acemoglu defines as “the capacity to undertake free actions and try new things without being subject to the arbitrary power of any person, norm, or institution.” That principle should be contrasted with classical liberalism’s principles for a relatively bounded government—one that prevents coercion and fraud, protects property and the right to contract, and provides public goods.

Nondomination, in contrast, “calls for some amount of redistribution and the provision of public services such as health and infrastructure to ensure a level playing field in the economy,” also, in his view, requires a more active and powerful government that can steer new technology to promote the principle by broadly distributing its occupational benefits.

Acemoglu does not address the problem of limiting principles. When does dependence on an employer or, indeed, on one’s family become domination? Which social norms constitute domination, and who decides? Why is discretionary state power not a strong form of domination? And perhaps most pointedly for Acemoglu, how can government emancipate citizens from allegedly oppressive norms without engaging in the social engineering that Acemoglu condemns? Acemoglu himself calls nondomination “an argument against social engineering—top-down efforts to change and police people’s beliefs and practices.” It is difficult to square that principle with his recommendations for activist government.

The vagueness of the nondomination mandate also raises well-known public-choice concerns. Public choice is the branch of economics that describes how politics works given interest groups’ incentives. Nondomination invites groups to characterize their own interests as claims against domination. Politicians, bureaucrats, unions, firms, and professional-class constituencies will define the concept in ways that protect them, even if doing so contributes to their domination of other groups.

Acemoglu himself warns that working-class liberalism “should never bypass the needy to curry favor with the middle class.” But entitlements and regulations that skew toward the better-off, particularly older people as a class, are not simply moral failures of government—they are predictable results of people voting in their own interests, with older people having more time and thus a greater propensity to vote. Moreover, the educated are better able to navigate administrative systems, and concentrated interests can wield far more influence than the working class because they do not face free-rider problems. When Acemoglu calls for a new agency to ensure AI aids the working class, he does not explain how he would create institutions or norms to alter the all-too-predictable incentives that will steer it in directions other than the ones he wants.

What protects pluralism when elites want to use courts or agencies to facilitate an elite consensus on the nation?

Acemoglu also does not make a persuasive economic case for what makes a pro-worker AI agency necessary. He says the agency should ensure AI is used to make workers more productive. He argues that neither digital advertising nor selling automation software “will naturally lead to a concerted effort to develop pro-worker AI.” But he never identifies the market failure that makes this mission necessary. If AI can help electricians diagnose equipment or nurses perform more sophisticated tasks, companies will have incentives to provide these AI tools. Incumbent firms do not exhaust the universe of potential suppliers. Startups have incentives to fill any gaps and increase workers’ know-how. The case for markets is not that they must be steered to provide people with the right tools, but that they have a natural incentive to do so.

Acemoglu also argues for a tax system that does not favor capital over labor. Moving toward a tax system more neutral as to source is at least a proposal that avoids market distortions and does not require a large bureaucracy. But it accentuates his failure to show an actual market failure. Why is neutrality not sufficient? Acemoglu thus has two problems. He does not sufficiently show why the market for upskilling workers through AI is imperfect, and even if it is, he compares it with a government agency whose imperfections he has not modeled.

Let’s provide a concrete example—one close to my heart—of the economic problems posed by the AI agency he proposes. For instance, he calls AI pro-teacher when it “would increase the demand for teachers—for helping with smaller groups after AI-generated group assignments and lesson plans are drawn,” rather than reduce the number of teachers needed. But why should increasing the demand for a particular occupational category be the focus of government rather than the market?

Occupations wax and wane. Just ask farmers. Labor-saving technologies can reduce prices, and teachers are also consumers and taxpayers. Such technologies have also historically created new kinds of jobs that we could not have imagined. To be sure, some people think we are on the cusp of artificial general intelligence, which will wipe out almost all human work, but Acemoglu is an avowed AGI skeptic.

Acemoglu himself quotes F. A. Hayek’s warning that the economic “data” “are never for the whole society ‘given’ to a single mind which could work out the implications and can never be so given.” His example illustrates the knowledge problem that occurs when one replaces the market with government steering. Which tasks should be automated? Which technologies will create new complementary jobs? Will an innovation that displaces labor create new jobs elsewhere?

Government agencies also provide opportunities for rent-seeking. The omission is especially striking because Acemoglu elsewhere acknowledges that “the speed and effectiveness of public service provision have diminished while the cost of these services has ballooned thanks to heavy regulation.” Teachers’ unions are notoriously powerful and are likely to use the agency to protect their members; other professional organizations, such as the Association of American Law Schools, will seek barriers to substitution. Even when national security is at stake, politicians steer funds to favored constituencies and regions. They will regard an agency with such a broad and indeterminate mission—promoting pro-worker AI—as a rudder to help them set a favorable course toward reelection.

The best part of Acemoglu’s book is his argument for broad democratic buy-in to government-led cultural change. He argues that “a practical social compact should leverage self-government and get buy-in from a significant fraction of the communities making up a society.” He is rightly worried that elites can seize government institutions and use them to impose their values on the rest of the citizenry.

But he fails the acid test of recognizing how elite institutions manufacture democratic consent. Amazingly, his example of consensus is the 2022 Respect for Marriage Act, which requires all states to recognize same-sex marriages and which he describes as “the end point of a similar journey.” The problem is his account of how the consensus was produced. Seven years before the statute, Obergefell v. Hodges required states to license same-sex marriages, overturning bans in thirty states. Thus, same-sex marriage cannot be fairly presented as simply the result of bottom-up democratic consensus. It is the kind of decision that alienates many citizens who see their preferences invalidated by a Court composed of elite lawyers.

Again, Acemoglu understands in the abstract that “community-level experimentation is vital for social adaptation in a changing world” and that “experimentation is powered by variation.” But he does not make sufficient use of federalism as the constitutional mechanism that enables such variation and, over time, helps build democratic consensus. Federalism permits different jurisdictions to have different policies. Our national guarantee of free speech then allows people to talk about the consequences, facilitating debate. That is how national consensus, if one is to be had, is best reached. When people have their say without it being preempted, they are much more likely to accept the result. Federalism is to political knowledge what competition is to economic knowledge—a process of decentralized discovery.

The book’s failures have some general lessons for left-liberals. Like classical liberals, they cannot avoid answering comparative institutional questions. Do ideas like nondomination actually empower state domination? What are the specific market failures that prevent the market from boosting welfare over the long run? What mechanisms prevent redistribution and new regulatory agencies from being captured by special interests? What protects pluralism when elites want to use courts or agencies to facilitate an elite consensus on the nation? Acemoglu is right to worry that elites can turn left-liberalism into an instrument of domination. But he never explains why the new AI agency he wants to empower would not become the next instrument of elites and interest groups.