Daron Acemoglu on Liberalism, Automation, and the Educated Elite

Core argument: Liberal democracy in crisis

Daron Acemoglu argues that current liberal democracy is in crisis because modern liberalism relies too heavily on social-contract-style ideas that avoid confronting the need for genuine consensus around shared moral values and priorities.

He proposes an alternative framework grounded in:

1. Protecting meaningful individual freedom
2. Building community-level agreements rather than relying on overarching “absolute truths” or a “general will” that overrides plural moral convictions

Contrast with Rorty-style consensus liberalism

Acemoglu contrasts his approach with Rorty-style consensus liberalism:

• He does not endorse objectivism.
• He also rejects the idea of an external, decisive metric for settling moral conflicts.
• When values clash (e.g., nondomination vs. noninterference), he admits there is no perfect rule.

His practical emphasis is nondomination:

• Freedom requires not only being left alone by higher authority,
• but also protection from people and institutions with much greater power over others.
• Therefore, safeguards must involve community-level institutions, not only individual noninterference.

---

Working-class liberalism and the “educated elite”

Acemoglu discusses “working-class liberalism,” updating what “working class” means in a post-industrial economy.

• Instead of focusing only on industrial manufacturing employment or unionized factory work, he defines the working class broadly as those whose incomes come primarily from the labor market across diverse occupations (manual trades, office jobs, entrepreneurship, etc.).
• He argues liberalism has become dominated by an “educated elite” that is both more powerful and more numerous.
• Empowering elite institutions without addressing social hierarchies (education, income, race) undermines genuine freedom and social flourishing.

He also emphasizes using expertise without technocracy:

• Expertise must not become insular, self-confident, or contemptuous of ordinary people.

---

Automation and AI: not anti-technology, but anti-misaligned technology

A major exchange focuses on automation’s effects on shared prosperity.

Automation

• Cowen challenges Acemoglu’s claim that automation breaks mass production from shared prosperity.
• Acemoglu replies that at the task level, automation converts labor tasks into capital/algorithms, tending to reduce the labor share (even if wages don’t immediately fall due to other factors).
• Automation is not uniformly harmful; the key is whether societies pair it with:
  • the creation of new tasks and opportunities,
  • and mechanisms that give displaced workers bargaining power and wage growth.

AI

Acemoglu treats AI as not identical to automation:

• Its benefits depend on achieving a “balance” between automating tasks and enabling new ones.
• He is more optimistic when AI/automation respond to shortages and when adoption includes job redesign and new task creation.

---

Free speech and platform regulation (especially social media algorithms)

On free speech, Acemoglu says he is effectively an absolutist about the right to speak (he wouldn’t take anyone’s right away).

However, he supports regulating how powerful platforms and their algorithms use speech:

• He argues algorithmic amplification can “weaponize” vulnerabilities in ways not analogous to older printing technologies (which he suggests societies adjusted to).
• In his view, chronological, non-algorithmic feeds would be more tolerable—algorithms are the disruptive variable.

---

Immigration: pro skilled immigration, worried about “education outsourcing”

Acemoglu supports skilled immigration (including H-1B-type policies), but is critical of how the educated elite uses immigration as a substitute for improving domestic education.

• He argues influential groups have become less committed to training Americans for the “skills age.”
• Instead, they assume skill gaps can be filled by importing expertise rather than investing in broad-based education and skill development.

He also critiques teacher unions “in their current form”:

• he calls for “reconstitution” to serve children’s interests,
• while still supporting unions more generally for labor representation.

---

AI productivity: uncertain gains, more cost savings than “game-changer”

On AI and productivity:

• Acemoglu previously predicted only modest productivity gains over the next decade.
• He says he would revise estimates upward somewhat due to faster model progress.

But he stresses a key distinction:

• improving models is not the same as widespread adoption/application.
• deployment and real-world integration have been slower than expected.

He anticipates:

• uncertainty,
• possibly periods of faster growth,
• but not the dramatic outcome of “transform everything immediately.”

He also notes that current AI-driven improvements in economics research are mainly:

• labor-saving (e.g., data cleaning, converting qualitative information into quantitative measures),
• not automatically producing major new scientific breakthroughs.

---

“Pro-worker AI” and decentralization

Acemoglu supports the goal of “pro-worker AI”, but rejects the idea that regulators should authorize only “approved” pro-worker AI.

Instead, he favors policies and incentives that cultivate AI that expands workers’ capabilities, such as:

• enhancing problem-solving,
• training,
• enabling workers to do more sophisticated tasks.

He is wary of regulators deciding AI’s desirability case-by-case, while still emphasizing social outcomes.

He also worries that foundation-model approaches may be overly centralizing:

• even if models can be copied,
• market “winner-take-all” dynamics may concentrate power.

He suggests:

• more decentralization,
• more domain-specific models, as a way to preserve liberal-democratic values and avoid excessive concentration of control.

---

Fertility crisis: less worry than some mainstream concerns (with Lucas-type uncertainty)

On population and macroeconomic consequences, Acemoglu says he is less worried about the fertility crisis than Cowen suggests.

• He argues induced innovation can offset demand/supply concerns.
• Birth rate declines are slow-moving rather than immediate shocks.
• He also acknowledges Lucas-critique-type uncertainty: the worldwide demographic transition may create general equilibrium effects unlike historical expectations.
• He notes longer life and improved health can increase human capital accumulation over the life cycle.

---

Other issues: Armenia and a future research agenda

Armenia

Acemoglu argues:

• Armenia initially had strong human capital (education/engineering legacy from the Soviet system),
• but underperformed economically and squandered early democratic promise.

Armenia remains geopolitically squeezed between Russia, Turkey, and Azerbaijan, making its future difficult and risky.

He believes the AI age could offer opportunities, but stresses persistent instability.

Closing themes / agenda

He reiterates themes for what comes next:

• making AI more pro-worker,
• determining where societies need talent (not only at the top),
• exploring how ideas function in political economy—arguing that materialist frameworks often underplay how ideas and group-level beliefs shape outcomes.

---

Presenters or contributors

• Tyler Cowen
• Daron Acemoglu
• David Autor (mentioned)
• Pascual Restrepo (mentioned)
• Robert W. Beirne (mentioned as “Beirn”)
• Scott (mentioned)
• Barany, Patel, and Siegel (mentioned)
• McKinsey (mentioned)
• OpenAI (mentioned)
• Anthropic (mentioned)
• Nvidia (mentioned)
• Mercatus (mentioned)

Mentions of other economists/authors as references rather than participants: Rorty, Rousseau, Rawls, Sidgwick, Hayek, King and Plosser, Long and Plosser, Romer, Habakkuk.